Citrini Research

All Along the AI Watchtower

A violent, leverage-driven unwind of the most crowded AI expressions — not open-source competition, Kimi3, or the Iranian conflict — is what explains the recent selloff, and the unwind has mispriced stocks by category rather than by fundamentals. The market has been willing to pay for infrastructure software but has indiscriminately clobbered application software, data businesses, and anything wearing an old label. Citrini therefore hunts for AI upside in names filed under the wrong category: Qualcomm as a data-center architecture play rather than a handset laggard, developer and context tooling that gets called more even as seats shrink, database and search vendors that meter agentic context maintenance, and the enterprise hardware and security complex that IBM's miss revealed is absorbing redirected capex. The unifying causal chain is that agentic AI shifts retrieval from an occasional preprocessing step to continuous execution-loop context maintenance, and pushes enterprises toward owning their stack — which routes spend to servers, storage, memory, and cybersecurity. Meta is the capstone: it is the only hyperscaler whose compute converts directly to P&L through advertising, so the bears subtracting capex first are getting the order of operations wrong.

• QCOM FY29 non-handset revenue guidance nearly doubled from $22bn to $40bn at the June 24 Investor Day
• Qualcomm High Bandwidth Compute (HBC) claims 6x bandwidth per watt vs HBM, 200x capacity per watt vs SRAM, 4-8x better decode performance per watt; arrives 2027
• QCOM automotive did $1.33bn in the March quarter, +38%; $65bn cumulative design pipeline; IoT target held at ~$14bn; robotics guidance embeds ~$700m
• MongoDB Atlas growth re-accelerated from ~26% to over 29% through fiscal 2026, crossing a $2bn annualized run rate; Atlas list pricing unchanged since at least 2020
• Elastic FY26 revenue +17%, Q4 +16%, FY27 guided ~14%; cRPO +20%, total RPO +28%; over a third of $100k+ customers using AI capabilities; SMB cloud +3%; ~19% operating margins
• Figma revenue growth 46% YoY with 139% net dollar retention, accelerating sequentially each quarter
• Fortinet: 900,000+ customers; product now low-to-mid 30s % of sales vs 40%+ a decade ago; consensus $763m Q4 2026 product sales; 70% OT billings growth in Q1
• Barclays survey: 88% of enterprise CIOs intend to move at least some public cloud workloads back to private cloud or on-prem
• Meta Andromeda co-design: 10,000x model capacity, 10x inference efficiency, 3x+ query throughput, 6% retrieval-recall gain, 8% ads quality gain; Advantage+ shopping +22% ROAS
• US CTV/online video sized at $45.2bn in 2025 (IAB); open-web programmatic display to $42.5bn by 2027 (EMARKETER); Meta planning ~14GW over three years

• Citrini's diagnosis of the selloff is positioning, not fundamentals: a leverage-fueled breakout in crowded thematic trades created its own fragility, and performance since has been driven almost entirely by categorical classification rather than idiosyncratic factors — which makes the dispersion a mispricing opportunity.
• The HBM trade contains its own antidote: every increase in HBM pricing strengthens near-term memory revenue while strengthening the case for near-memory architectures that use fewer HBM bytes per token. Qualcomm's HBC (2027) claims 6x bandwidth per watt vs HBM and removes the CoWoS packaging constraint.
• Citrini is more bullish on Qualcomm's automotive and IoT than management is: auto did $1.33bn in the March quarter (+38%) with content per vehicle rising eightfold between third- and fifth-generation platforms, while the IoT target was left unchanged at ~$14bn.
• GitLab's Orbit could become a toll on external coding agents — queries via MCP or REST consume GitLab Credits — but the economics are not yet real: Orbit Remote is beta, Duo Agent Platform queries are zero-rated, Orbit Local is free, and no per-query multiplier is published.
• MongoDB's edge is that Atlas list pricing hasn't moved since at least 2020 while new hot-capacity meters were bolted onto the same base curve — the toll booth was already built. Atlas re-accelerated from ~26% to 29%+ past a $2bn run rate.
• Elastic's numbers point in two directions: revenue decelerating to ~14% FY27 guidance while cRPO +20% and RPO +28% and SMB cloud grows just 3% — either enterprises are standardizing ahead of consumption (bullish) or contract duration is simply stretching (cosmetic).
• IBM's miss is the cleanest read on enterprise budget rotation: capex shifted to servers, storage, memory, and cybersecurity in the last few weeks of June, ahead of expected price increases — the empirical basis for the Enterprise Security and Sovereignty Basket.
• The Meta bull case is sequencing: it is the only hyperscaler growing fast without external cloud revenue, so compute hits the P&L directly — and the unmodeled leg is third-party CTV/programmatic inventory against a $40-50bn US pool where Meta owns the demand rather than the media.

Small Themes: June 2026

With virtually all analyst attention crowded into the AI trade, Citrini argues attention itself is the scarcest input -- so under-owned, under-modeled names outside AI are where thematic money gets made as 'AI exhaustion' rotation risk rises. The June 2026 Small Themes lays out five trades that need no view on AGI: US mainline airlines (DAL, UAL) finally getting their supply-constrained cycle as exogenous shocks fade; senior housing riding an unstoppable Boomer demographic wave into collapsed supply (WELL, JAN, BKD); live events and scarce sports assets where economics migrate from resellers to owners (long LYV/SPHR/MSGS/FWONK-type names, short SEAT/STUB); the end of CME's futures monopoly as 24/7 perps go legal (short incumbents, long HOOD/COIN/SPGI, with MIAX the exception); plus retail broker consolidation and a misunderstood, inflecting BNPL sector (AFRM, KLAR, SEZL).

• International arrivals to the US fell 5.5% to 68 million in 2025; Brent ran from $75 to north of $120 in 2026
• Delta record March quarter: $14.2B revenue (+9%), premium revenue +14%; Spirit shutdown removed 3.5 points of domestic capacity
• US 80+ population to grow >56% over the next decade vs ~5% total population growth
• Senior housing occupancy 89.5% (19th consecutive quarterly increase); construction starts -77% from peak; inventory growth 0.4%
• Welltower: $10B invested YTD while net leverage fell from 3.3x to 2.7x
• Dallas Cowboys valuation: $238M (1995) to ~$13B; league media rights run-rate $5.3B (2006) to $21.9B (2026)
• Live Nation: 107M+ concert tickets sold through April 2026, +11%; Sphere Q1 segment revenue $266M (+69%)
• World Cup: ~1.24M visitors expected at ~$4,000 average spend; average get-in resale price ~$1,000 (+80% in 30 days)
• Robinhood prediction-market revenue: $3M to $104M in one year; weekend crude perp volume spiked to $1.7B in a single day
• Affirm: $46B GMV in 12 months through March 2026 (~$50B pace, +35% YoY), 30+ day delinquencies high-2%; Sezzle EBITDA margins low-30s to ~38%; Klarna quarterly GMV $33.7B

• Attention is AI's scarcest byproduct: crowding leaves non-AI names under-owned and under-modeled, and with 'AI exhaustion' rotation risk high, Citrini has been trimming AI exposure and buying what the 'AI Dutch disease' left behind.
• Airlines earned through two exogenous shocks (2025 tourism collapse: arrivals -5.5% to 68M; 2026 Brent spike $75 to $120+): Delta posted a record March quarter and Spirit's May shutdown removed 3.5 points of domestic capacity -- DAL and UAL are set up for the cycle they deserve, helped by World Cup-driven sentiment reversal.
• Senior housing is a demographic certainty meeting collapsed supply: 80+ population grows 56% in a decade, occupancy at 89.5% (19th straight quarterly gain), construction starts -77%, inventory growth 0.4%, and rents must rise ~30% for new builds to pencil; WELL, JAN and BKD are the picks, and it is not merely a rates trade.
• In live events, own the creator and fade the reseller: FTC all-in pricing and anti-scalping laws are crushing middlemen (Vivid Seats EBITDA under $10M from $21.7M; StubHub ~$11 vs $23.50 IPO after a $10M FTC refund order), while Live Nation (107M+ tickets through April, +11%) and Sphere (revenue +69%) capture the economics.
• Scarce sports assets compound on fixed supply: Cowboys $238M (1995) to ~$13B; MSGS trades ~30-40% below Forbes-implied Knicks+Rangers value with a spin-off catalyst; MANU rumored to clear above 5 billion pounds vs ~3.5 billion market value; CHDN guided record Derby Week EBITDA on premium hospitality.
• CME's monopoly is ending as perps go legal: S&P 500 licensed to trade.xyz for 24/7 perps, weekend crude perp volume hit $1.7B/day during the Iran war, and CME is suing the CFTC -- short CME/CBOE/ICE, long Robinhood (prediction revenue $3M to $104M in one year), Coinbase, and S&P Global as the venue-agnostic licensor.
• MIAX is the exchange exception: it kept 10% of MIAXdx (the venue Robinhood/Susquehanna bought), has taken 17%+ of US options volume with Q1 revenue +40% and EBITDA +66%, and holds a decade-long exclusive Bloomberg index license -- the 10% stake could be worth more than the current market cap.
• BNPL's 'next domino' framing misreads upfront CECL loss provisioning against ratable revenue: Klarna printed its first GAAP profit in ~5 years ($33.7B quarterly GMV), Affirm hit FY25 GAAP profitability with 30+ delinquencies in the high-2% range and a path to $100B GMV, and Sezzle's adjusted EBITDA margins are expanding toward 38%.

Macro Memo: Spin Cycle

Citrini stays medium-term bullish on the US economy and equities despite a fresh inflation-and-rate-hike scare: the Iran/Hormuz energy shock created an 'Echo Shock' where measured inflation peaks just as source pricing pressure reverses, core inflation and wages show no spiral, and strong payroll prints overstate labor market heat. New Fed chair Kevin Warsh -- feared as a hawk and priced 70% for a December hike -- will do nothing, because no wage-price spiral exists and his own framework points dovish. The melt-up therefore continues (S&P 500 seen trading 8000 before real trouble) amid more frequent 10-15% drawdowns; trades are long SFRH7 to fade the priced hike and long gold as the credibility hedge.

• Short-term rate expectations up over 100bps since the war began February 28
• May payrolls: 172,000 jobs vs 88,000 consensus
• April FOMC vote split 8-4; ECB hiked 25bps
• December Fed hike ~70% priced into markets
• April shelter CPI +0.6% m/m, double its trend contribution (shutdown payback)
• ~60% of May payroll gains from restaurants and local government
• S&P 500 target: trades 8000 before worries meaningfully materialize; potential 10-15% further melt-up
• Effective Fed Funds ~3.6%; SFRH7 trade worth ~30bps of repricing in base case
• Gold fell 11% in a single session on Warsh's nomination
• Expect higher incidence of 10-15% drawdowns over the next 3-4 months

• Since the Iran war began February 28, short-term rate expectations jumped over 100bps and SOFR futures priced near-term hikes for the first time this cycle -- a repricing Citrini thinks is wrong and fadeable.
• The 'Echo Shock' concept: measured inflation peaks just as actual source pricing pressure reverses; spot energy has already rolled over and fertilizer, corn and wheat are cheaper than before the war started.
• Core inflation is calmer than it looks: April's 0.6% shelter jump was mechanical payback for the October government shutdown's zero-rent BLS assumption, core goods inflation turned negative, and real wage growth is sharply negative -- no wage-price spiral conditions exist.
• The May payrolls beat (172,000 vs 88,000 consensus) overstates labor strength: hire and quit rates haven't inflected, layoffs ticked up, hiring is below 2019 levels, and ~60% of May's gains came from restaurants and local government.
• The market has Warsh backwards: despite a 70%-priced December hike and G. William Miller comparisons, Warsh praised trimmed-mean inflation and AI-driven disinflation in confirmation; Citrini's call is he does absolutely nothing -- no hike, no balance sheet shrink.
• 2026 is not 1979: COLA contracts no longer transmit oil into wages, the US uses less than half the oil per real dollar of GDP, and is a net petroleum exporter -- tightening would stack a demand shock on an existing one.
• Equity view: the sell-off was a leverage flush, not a top; melt-up continues into late summer with more frequent 10-15% drawdowns, and the S&P 500 trades 8000 before worries meaningfully threaten the cycle.
• Trades: long SFRH7 (fade the priced hike, ~30bps base-case repricing, double if cuts return; closing the completed 2s30s flattener) and long gold as the credibility hedge after it paid the 'hawk tax' twice.

State of the Themes: June 2026

Citrini's June 2026 State of the Themes takes stock after a blistering AI-led rally: the dominant new narrative is 'Token Panic' -- exploding agentic token consumption colliding with labs' shift to usage-based pricing, so enterprise AI bills are blowing up budgets (Anthropic ARR hit $45B, up 5x YTD) and cost/efficiency themes now matter most. That favors edge AI and on-device inference, flash-based workarounds to the DRAM tax, agentic utilities that win on usage regardless of token cost, and selective 'AI loser' rebounds, while robotics custom silicon, EU Chips Act 2.0 beneficiaries, Colombia's rightward election, a tightening oil capex cycle, and new ideas (Shopify, Hyperliquid, healthcare's cyclical inflection) round out the thematic map. The unifying view: the AI trade broadens and rotates rather than ends.

• Anthropic ARR $45 billion in May 2026, up 5x since start of year
• Claude Opus 4.7/4.8 new tokenizer can use up to 35% more tokens for the same text; Chinese models 10-25x cheaper
• Prior calls: GOOG doubled in 9 months, KOSPI tripled, Taiwan outperformed QQQ by >60%
• DRAM contract prices up ~90% in a single quarter; flash ~55x cheaper per bit; HBF offers 8-16x HBM4 capacity
• Bandwidth up nearly 300% since Agentic Utilities; Fastly RPO +63%; Akamai $1.8B 7-year frontier-lab compute deal
• Amazon investing EUR 10 billion in European robotics; Figure AI initial peak capacity 12,000 robots/year
• Chips Act 2.0 includes EUR 15-30 billion to establish European memory production; Huber+Suhner backlog record CHF 432M
• Shopify: >14% of US e-commerce, ~$380B volume (+30% YoY), Shop Pay >$160B annualized GMV, Thrive Capital $100M investment
• Hyperliquid: $40-50B weekly perp volume, >$2B cumulative HYPE buybacks (~7% of market cap annually), $1B token burn, HIP-3 OI >$3B
• Molina normalized EPS seen >$28 (> $30 by 2028); Shinsegae luxury jewelry sales +147% YoY; Danaher Cytiva bioprocessing orders +30% YoY

• Token Panic is the new market narrative: Anthropic ARR hit $45B in May (5x YTD), Uber reportedly burned its full-year AI budget in four months, and OpenAI, Google and Microsoft all shifted to usage-based pricing between April 2 and June 1, 2026 -- making cost, routing, and observability the investable themes.
• Chinese and open-source price pressure is real: Qwen 3.7 and DeepSeek V4 are 10-25x cheaper than frontier US models, DeepSeek passed Anthropic on OpenRouter token volume, and Cursor's Moonshot-based model matches frontier quality at 10x lower cost per task.
• Edge AI is inflecting after being 'way too early' in July 2024: Nvidia's Spark AI PC (1 petaflop FP4, 128GB unified memory), OpenAI's io device roadmap, and Google eyewear point to a hybrid local/cloud model; the Edge basket has begun outperforming SMH.
• Flash is the way past the DRAM tax: DRAM contract prices rose ~90% in a quarter and HBM eats ~25% of DRAM wafer capacity, while flash runs ~55x cheaper per bit; Sandisk/SK Hynix's High Bandwidth Flash offers 8-16x HBM4 capacity -- bullish SIMO and Phison, medium-term negative for HBM scarcity premiums.
• Agentic Utilities validated: Bandwidth +~300% since the primer with a Salesforce Agentforce win, Fastly RPO +63%, Akamai signed a 7-year $1.8B frontier-lab compute deal, and SailPoint reports 11% of governed SaaS identities are already non-human.
• The AI Losers rally is mostly an undiscerning short squeeze (Chegg, Upwork, call centers bouncing despite impaired models), but FactSet could trade 50-70% higher on MCP-driven seats; FICO, EFX, Shopify and Guidewire are the selective buys.
• SPAD-based color LiDAR (Hesai Picasso, Ouster Rev8/L4) collapses the camera-vs-LiDAR debate and cuts robotics compute costs; Amazon is putting EUR 10B into European robotics and Figure AI began production with 12,000/year peak capacity.
• EU Chips Act 2.0 (June 3, 2026) abandons the 20% production-share goal for demand-side measures including EUR 15-30B for European memory; under-owned beneficiaries include Huber+Suhner (OCS, CHF 432M backlog), Canatu, BESI, ams-OSRAM and Nordic Semiconductor.
• New ideas: Shopify (14%+ of US e-commerce, ~$380B volume +30% YoY, Thrive's $100M stake), Hyperliquid's HYPE ($2B+ buybacks, ~7% annual buyback rate, HIP-3 open interest >$3B), and healthcare's cyclical inflection (Cytiva orders +30%, 173 AI-originated drug programs in the clinic vs ~24 in late 2023).

Semis Memo: Supply Chain Inheritance

The AI infrastructure trade has matured past first-order picks, so Citrini's third Semis Memo hunts mispriced second-order bottlenecks. The core idea is 'Supply Chain Inheritance': Nvidia's 800V DC rack architecture inherits the silicon carbide/GaN supply chain built for EVs and solar, so power semis still valued on a weak EV cycle (STM, ON, Infineon, and above all post-bankruptcy Wolfspeed) get a second demand cohort just as the first wobbles. Around that anchor, the memo argues CPUs and the orchestration plane become strategic in the agentic era (ASPEED, LOTES, Qualcomm), an inference shortage re-rates neoclouds (CoreWeave, Hut 8), AI materials bottlenecks keep worsening (Ibiden, WF6 gases via Kanto Denka), and Samsung Foundry's re-rating as TSMC's realistic second source unlocks Korean beneficiaries (Doosan Tesna, Wonik IPS, Samsung Electro-Mechanics, LG Innotek).

• SiC market: $300M (2018) to $3.5B (2024), ~12x; AI estimated at half of SiC demand by 2030
• Rack power BOM +16% in 800V transition; wide-bandgap + HV connectors from 0% to ~64% of rack power BOM
• Rubin Ultra GPU projected >2,500W, implying ~3,500 amps at 0.7V core voltage
• Wolfspeed spent $6.5B on SiC capacity before June 2025 Chapter 11; demonstrated 300mm SiC crystal in January 2026
• ASPEED: 80% BMC share (100% in Nvidia Hopper/Blackwell), 87 BMCs per NVL72, TAM revised to 65M chips by 2030
• OpenAI: compute 0.2 GW to 1.9 GW (2023-2025), $2B/month revenue by March 2026, 15B+ tokens/minute
• Anthropic run-rate revenue $45B by May 2026, up ~400% from ~$9B at end-2025
• CoreWeave RPO $99.4B in Q1 2026; $21B Meta expansion; Hut 8 $9.8B 15-year lease (352 MW)
• Korean CCL import prices +74.5% YoY; MLCC lead times 24 weeks (~3x production cycle); Rubin Ultra ~1.8M MLCCs (~90x H100)
• AEHR +300% with $41M Google TPU order; Himax +125%; 2027 AI spending now estimated above $3 trillion

• Supply Chain Inheritance: Nvidia's 800V DC architecture inherits the EV/solar SiC-GaN supply chain -- SiC grew 12x from $300M (2018) to $3.5B (2024) on EVs, and AI is estimated to reach half of total SiC demand by 2030 on the same wafers, while SiC names are still valued on the weak EV cohort.
• The 800V transition grows rack power BOM ~16% but shifts mix dramatically: wide-bandgap silicon plus high-voltage connectors go from 0% to roughly 64% of rack power BOM; VRM content compounds as Rubin Ultra GPUs (~2,500W) require ~3,500 amps at 0.7V core voltage.
• Wolfspeed is the 'crouching tiger' pick: $6.5B of capex bankrupted it in June 2025, but post-restructuring equity owns the world's only commercial-scale 200mm SiC fab and one of only two demonstrated 300mm SiC crystals (Coherent the other), with New York State billions sunk and not expecting repayment.
• Agents shift value to the CPU/orchestration plane (agents use 10-30x more tokens than chatbots); ASPEED has 80% BMC market share overall and 100% in Nvidia Hopper/Blackwell servers (87 BMCs per NVL72), and revised its 2030 BMC TAM from 46.5M to 65M chips.
• The inference shortage is visible in deals: Anthropic's run-rate revenue hit $45B by May 2026 (up ~400% from ~$9B at end-2025) and it signed Amazon (up to 5 GW), Google/Broadcom ($200B over 5 years), SpaceX (300+ MW, 220,000+ GPUs) and Akamai ($1.8B/7yr); CoreWeave RPO reached $99.4B in Q1 2026.
• AI materials bottlenecks all worsened: T-glass relief pushed to 2027+, Korean CCL import prices +74.5% YoY, Nvidia substrate area going from 3,190mm2 (Hopper) to ~8,000mm2 (Rubin); Ibiden is the top materials pick and WF6 specialty gas (Kanto Denka) the emerging bottleneck.
• Samsung Foundry is the realistic second source to a maxed-out TSMC, unlocking Korean beneficiaries: Doosan Tesna (Apple CIS + Nvidia LPU test capex of KRW 362B+), Wonik IPS (Taylor fab tooling, anchored by Tesla's $16.5B deal), Samsung Electro-Mechanics (FC-BGA + MLCC: Rubin Ultra uses ~1.8M MLCCs, ~90x H100), and LG Innotek (SiP substrate survivor economics).
• Prior memo picks validated and held: AEHR +300% (with a $41M Google TPU burn-in follow-on order and 2H bookings above $92M) and Himax +125%; Citrini's meta-lesson is that even 5x-consensus AI estimates proved conservative -- 2027 AI spend now estimated above $3 trillion -- and the cycle ends on macro, not technology.

Flash Note: Defense Production Act

On April 20, 2026 the President signed Presidential Determination 2026-10, a Defense Production Act Section 303 determination classifying grid infrastructure -- transformers, transmission, conductors, substations, breakers, relays and electrical core steel -- as essential to national defense. This converts Citrini's long-running Fiscal Primacy x AI power thesis into an actual federal priority regime: priority ratings compress lead times for grid customers, federal capital crowds in private investment, and a national-security frame replaces climate policy, making the spending durable across administrations. Winners are the already-known grid supply chain plus overlooked names, above all Cleveland-Cliffs, whose GOES electrical steel monopoly maps one-for-one onto the DPA designation, along with Quanta, Worthington Steel and Hammond Power.

• ~$323 million of FY26 DPA Title III funding remaining at announcement
• Korean Big Four transformer order backlog: $23.9B as of Q3 2025 (5-6 years of work)
• Large power transformer lead times: 3-4 years; distribution transformer backlogs ~1 year
• HD Hyundai Electric expanding Montgomery, AL to 150 transformer units/year by 2027
• Iljin Electric won a record $333M order from a US energy company
• GEV 1Q26 electrification backlog net addition nearly equal to annual additions from 2022-2025
• Cleveland-Cliffs restarting Weirton, WV as a $150M transformer plant
• Foreign GOES faces tariff stack reaching 50% on derivative products
• POWL nearly tripled since Citrini's Stargate Field Trip pick

• Presidential Determination 2026-10 (April 20, 2026) invokes DPA Section 303 for grid infrastructure, formally naming transformers, transmission, substations, breakers, relays, capacitor banks and electrical core steel as essential to national defense.
• The DPA's real power is not money (~$323M of FY26 Title III funds remain) but priority ratings that reroute supply to grid uses, crowded-in private capital, DOE Loan Programs Office access, and a national-security framing that survives administration changes.
• The grid backlog keeps accelerating: GE Vernova's 1Q26 electrification backlog net addition was nearly as large as the annual additions from 2022-2025; large power transformer lead times run 3-4 years.
• The marginal US transformer supplier is Korean: the Big Four (Hyosung, HD Hyundai Electric, LS Electric, Iljin) hold a combined $23.9B backlog -- five to six years of work -- and are building US plants.
• Cleveland-Cliffs is effectively the sole domestic GOES producer (Butler PA / Zanesville OH, only domestic high-permeability TRAN-COR), is converting Weirton WV into a $150M transformer plant, and is shielded by tariffs up to 50% -- a rare one-company map to a national-security determination.
• EPC contractors like Quanta Services should outperform equipment OEMs on a 24-month view: the DPA can fund factories but cannot quickly fund the human-capital pipeline of high-voltage linemen.
• Overlooked downstream/adjacent picks: Worthington Steel (toll-processing GOES into slit coils and lamination blanks, un-rerated) and Hammond Power Solutions (largest North American dry-type transformer maker, faster to deploy for data centers).
• Self-aware risk flag: success in adding capacity via the DPA eventually means lower margins for the complex, but likely not for another couple of years.

Strait of Hormuz: A Citrini Field Trip

Citrini sent a field analyst to the Strait of Hormuz during the live US-Iran conflict and concluded the market's binary 'open vs. closed' framing is wrong: Iran is running the strait as a toll road, not a blockade, approving vessels through a Qeshm-Larak checkpoint while the ground conflict continues. Hot war and commercial diplomacy are happening simultaneously -- US allies like France, India and Japan are negotiating passage with Tehran, mostly via underreported diplomatic workarounds rather than yuan or crypto payments. Traffic should keep rising (Citrini expects ~50% of pre-conflict volume within 4-6 weeks) even as the conflict escalates, so the trades are ones that win regardless of resolution: tankers (BWET), US petrochemicals (DOW, WLK), long SFRH7 rates, long December 2026 WTI versus front month, relative country pairs, and EU/Japan energy-independence and defense names (EUAD).

• 15 ships confirmed crossing the strait on April 2; 15-18 on April 4, vs 2-5/day two weeks prior and >100/day pre-conflict
• Net hit to global commercial oil stocks estimated at 10.6 million barrels per day
• Front-month WTI ~$110 vs December 2026 WTI ~$71
• BWET has more than quadrupled since the January tanker call; Iran Conflict Beta basket +17% since inception
• Up to 50% of global polyethylene supply offline, constrained or impacted (Dow CEO); 250-275 days to normalize supply post-reopening
• Dow polyethylene price hikes: $0.10 in March, $0.30 in April (double the announced $0.15)
• SFRH7 fell to 96.05 on rate overreaction, recovered to ~96.40
• March 26: Iran granted passage to China, Russia, India, Iraq, Pakistan; Malaysia, Thailand, Philippines, France, Japan added within a week
• If the strait stays effectively closed, global equities seen falling ~15-20%
• Expectation of up to ~50% of pre-conflict strait traffic within 4-6 weeks

• Iran is running a toll road, not a blockade: vessels submit ownership/flag/cargo/crew details through brokers, pay via cash, crypto or -- most commonly and underreported -- diplomatic workarounds like unfreezing Iranian assets settled through Bank of Kunlun, then transit the Qeshm-Larak channel with a confirmation code.
• Hot war and commercial diplomacy coexist: traffic rose from 2-5 ships/day to a confirmed 15 on April 2 and 15-18 on April 4 even as a US F-15 and A-10 were shot down over Qeshm -- though this remains far below >100 ships/day pre-conflict, and mostly smaller than Aframax vessels.
• No ship with IRGC clearance has been struck, and the Houthis are being deliberately restrained from closing the Bab al-Mandab -- evidence of a centralized Iranian command holding escalation cards in reserve.
• The 'insurance is the bottleneck' narrative is wrong; per a vessel captain, the bottleneck is not wanting to die and not wanting to lose an irreplaceable ship in the hottest rate environment ever -- while idle ships print money as floating storage.
• Iran's endgame is a Montreux Convention-style sovereignty regime (like Turkey's Bosphorus management since 1936): tolls and rules under Tehran's control with commercial traffic flowing, positioning the US as the disruptive actor.
• US petrochemicals is the most asymmetric trade: up to 50% of global polyethylene supply is offline/constrained per Dow's CEO, who sees 250-275 days to normalize even post-reopening; Dow hiked polyethylene $0.10 in March and $0.30 in April; longs are DOW and WLK, still in ~50% drawdowns.
• Tankers (BWET, up more than 4x) haven't topped -- rates stay elevated in nearly every scenario -- and the Iran Conflict Beta basket is +17% since inception.
• The oil curve misprices duration: front-month WTI at $110 vs December 2026 at $71 implies a short sharp shock, but Citrini sees a longer disruption with a permanent risk premium, expecting ~50% of pre-conflict traffic within 4-6 weeks; the crisis reads as a symptom of a multipolar world (upcoming 'Age of Empires' primer).

Agentic Utilities: Always On, Never Prompted

AI agents reached an inflection point with the viral adoption of OpenClaw in late 2025, converting theoretical agentic AI into a real workload that consumes tokens around the clock. Because agents generate up to 25x the network traffic of chatbots, run always-on persistent sessions, and route traffic 'east-west' between servers rather than 'north-south' to humans, the structure of the internet itself must be rebuilt. The winners are 'Agentic Utilities' -- infrastructure names (CDNs, optical, DNS, data centers), business-to-agent ecosystem tollbooths (payments, telephony, e-signature), and governance vendors (observability, identity, security) -- many of which are legacy stocks left for dead that are already seeing revenue inflections. Citrini is long AKAM, FSLY, CRCL and NET in the Citrindex and publishes a broad Agentic Utilities basket.

• Cisco estimate: agentic AI generates up to 25x more network traffic than a chatbot
• Nokia: agentic M2M traffic to grow from 66 EB/month (2025) to 537 EB/month (2034)
• HPE networking segment +152% YoY, now 30% of total revenue and half of operating profits
• VeriSign: 450 billion DNS queries/day, 67% operating margins
• Akamai API security revenue +100% YoY; $250M capex for Akamai Inference Cloud
• Equinix: 500,000 interconnects; 60% of large new contracts from AI workloads
• US call centers employ 3.6 million people (~2.5% of workforce), >$100 billion annual labor spend
• DocuSign IAM: >$350M ARR (~11% of ARR) in FY2026, expected to nearly double in FY2027
• PayPal PYUSD circulation grew to over $4 billion in 2H25
• Rackspace carries $2.75 billion in long-term debt; Palo Alto's CyberArk deal ~$25 billion

• Agentic AI went from buzzword to reality with OpenClaw (formerly ClawdBot) in November 2025; its trick of storing context in a text file and looping AI calls drove astronomical GitHub adoption and token usage in two months.
• Cisco estimates agentic AI generates up to 25x more network traffic than a chatbot, and Nokia projects agentic machine-to-machine traffic rising from 66 EB/month in 2025 to 537 EB/month in 2034 -- traffic shifts from north-south (human-server) to ~80% east-west (server-server).
• Legacy internet plumbing is inflecting: HPE networking grew 152% YoY to 30% of revenue, Ciena raised capex ~2.5x its historical average to meet (not anticipate) demand, and VeriSign (450 billion DNS queries/day, 67% operating margins) flagged agentic AI on its last two earnings calls.
• CDNs become 'Compute Delivery Networks': Akamai's API security revenue is up 100% YoY and it launched an Inference Cloud with $250M capex; Fastly turned FCF-positive on agentic traffic and offers the highest convexity after being left for dead.
• A new B2A (business-to-agent) category is emerging in payments: Coinbase/Cloudflare's x402 standard revives the HTTP '402 Payment Required' code, Circle Nanopayments enables gas-free USDC transfers as small as $0.000001, and Cloudflare launched NET Dollar and pay-per-crawl.
• Telephony is a regulatory tollbooth agents cannot vibecode away: Bandwidth (PSTN interconnects, OpenAI Realtime API integration) and Twilio (Segment data, ConversationRelay, 85% of its own inbound sales calls handled by AI) are the picks.
• DocuSign's IAM platform generated over $350 million ARR (~11% of ARR) in fiscal 2026 and is expected to nearly double in fiscal 2027, with MCP/Claude integrations plus FedRAMP and ESIGN/UETA/eIDAS regulatory moats.
• Governance is the third leg: observability and security converge into 'Agentic Governance', favoring Palo Alto (Chronosphere acquisition plus ~$25 billion CyberArk deal), SailPoint, Zscaler and Datadog; Citrini remains unconvinced on Okta despite cheapness.

Let There Be Light: Revisiting the Optics & AI Connectivity Theme

Citrini's contrarian mid-2024 optics calls (Interconnects 101, Can You Hear Me Now?) have tripled the connectivity basket, but the obvious winners (Lumentum, Coherent, AAOI) are now priced for perfection, so the next asymmetry sits deeper in the photonics stack. Both pluggable transceivers and co-packaged optics converge on the same upstream substrates, components and equipment — so regardless of CPO timing, MOCVD tools (Aixtron), bonding equipment (SUSS MicroTec), SOI substrates (Soitec), InP fabs (Nokia) and optical attach hardware (Furukawa, Sumitomo Electric) capture the buildout. The highest-conviction single name is Himax, whose nanoimprint-lithography micro-lens arrays and V-groove baseplates appear (per a Hunterbrook patent investigation) to sit inside TSMC's COUPE fiber-array-unit supply chain for Nvidia CPO, with a second underappreciated kicker in Apple smart-glasses optics.

• Hyperscaler AI capex estimated at $650B-$700B for 2026; connectivity basket has more than tripled since publication and is +35% YTD
• Compute performance +60,000x over two decades vs electrical interconnect bandwidth only +30x; CPO cuts per-port power from ~30W to as low as 9W
• AI-cluster optical transceiver market on track to exceed $10B in 2026, roughly doubling from 2024; LightCounting: pluggable optics $15.5B in 2025 growing 17% CAGR to >$34B by 2030, CPO >$9B by 2030
• AAOI: ten customers = 96.6% of 2025 revenue (Microsoft 28.8%); >$1B 2026 revenue guided but supply-constrained; 'major laser shortage' with 1-year lead times; $209M 2025 capex vs $120-150M expected
• NVIDIA invested $2B each into Lumentum and Coherent (March 2026) plus $1B into Nokia for 6G; Lumentum expanded EML production 40% in 2025 with another 40% planned in 2026; Tower Semi investing $650M to triple SiPh shipments
• Himax: CPO revenue could exceed $100M annualized in early mass production (2026 first MP year); Gen-2 FAU targets >6.4T for GPUs in late 2027/early 2028; Morgan Stanley projects 5,000 Rubin Ultra racks in 2027 and 28,000 in 2028; HIMX at ~3-4x EV/EBITDA on 2028 estimates
• Apple smart glasses: est. 3-5M launch-year units; Himax WiseEye ASP ~$10-15, combined AR optics ASP ~$70-90 by ~2028; non-driver segment seen going from <20% to ~50% of Himax sales by 2028
• Aixtron: 70-90% MOCVD share, >85% GaN share; photonics guided to more than double YoY (~€120M optoelectronics in 2025 toward +€250M)
• SUSS MicroTec: ~50% share of Samsung/Micron HBM temporary bonding, ~90% of TSMC photomask cleaning; 2030 targets €750-900M revenue at 20-22% EBIT margin (~6x forward EBIT at midpoint) vs BESI ~40x
• Furukawa DFB laser chip capacity +500% by 2028 vs 2025; Sumitomo Electric info-comm operating profit ¥46.1B vs ¥10.4B (+360%), connector capacity +7x 2023-2028, InP substrate capacity +2.4x by 2028; STMicro photonics revenue 'nicely above' $500M this year and well above $1B next year on a multi-billion AWS contract

• The equipment/substrate layer is architecture-agnostic: pluggables and CPO consume the same upstream MOCVD reactors, InP crystals and SOI wafers, so the 'which architecture wins' debate doesn't matter for Aixtron, SUSS, Soitec and peers — Hock Tan's CPO skepticism is irrelevant to the thesis (and Broadcom itself has shipped three CPO generations to Meta).
• Citrini's original optics calls tripled the connectivity basket (+35% YTD, led by Lumentum), but the firm now advises taking profits in the crowd favorites: AAOI — 96.6% of 2025 revenue from ten customers, Microsoft 28.8%, CPO laser only at sample stage, mid-2027 targets requiring 5x capacity on unshipped technology — could trade 50% lower and is a candidate short against the new basket.
• The supply chain is validating the bottleneck: AAOI cites a 'major laser shortage' with 1-year lead times, Lumentum is expanding EML output 40% for a second straight year, Tower is spending $650M to triple SiPh shipments, and Nvidia invested $2B each into Lumentum and Coherent in March 2026.
• Himax is the highest-conviction asymmetric play: patent forensics (two Himax patents and a TSMC inspection patent sharing an unusual 22-channel FAU design) plus a former Lumentum engineer's read suggest Himax stamps the optics blocks and V-groove plates that FOCI assembles into FAUs for TSMC's COUPE and Nvidia's CPO platform; management guides $100M+ CPO revenue in early mass production.
• Himax's second kicker is Apple smart glasses: WiseEye sensors at $10-15 ASP on an expected 3-5M unit launch, then LCoS microdisplays and nanoimprint waveguides at $70-90 combined ASP for ~2028 display-equipped AR glasses — potentially rivaling CPO revenue; non-driver sales could reach ~50% of revenue by 2028, putting HIMX at ~3-4x EV/EBITDA on 2028 numbers.
• Aixtron is 'a photonics equipment monopoly with the GaN ramp for free': 70-90% MOCVD share, photonics orders guided to more than double in 2026 (potentially +€250M, becoming the largest segment), and 85%+ GaN share as Nvidia's 800V HVDC Rubin Ultra architecture forces GaN tool orders from H2 2026.
• SUSS MicroTec is the purest advanced-packaging equipment play: ~50% of Samsung/Micron HBM temporary bonding, ~90% of TSMC photomask cleaning, ~70% of revenue tied to advanced packaging, the XBS300 doing CPO heterogeneous integration — at ~16-18x trough EBIT (~6x on 2030 targets) versus BESI at ~40x forward EBIT.
• Soitec's moat is proven by its would-be competitor — Shin-Etsu, the world's largest silicon wafer maker, licenses Smart Cut rather than replicating it (~10x trough EV/EBITDA for an ~80% share SOI monopoly) — while Nokia is a hidden photonics semi company: the Infinera deal gives it the West's only vertically integrated InP fab, CHIPS-subsidized San Jose and Bethlehem buildouts, and it uniquely both receives a $1B Nvidia 6G investment and benefits from Nvidia's $4B spend across Lumentum and Coherent.

The 2028 Global Intelligence Crisis: A Thought Exercise in Financial History, from the Future

Written as a fictional macro memo from June 2028, the piece models what happens if AI bullishness is right and that is precisely what is bearish: agentic AI triggers a self-reinforcing 'intelligence displacement spiral' in which capability improvements drive white-collar layoffs, savings are redeployed into more AI, displaced high earners cut the discretionary spending that anchors the consumer economy, and the loop has no natural brake because AI is OpEx substitution, not cyclical capex. The disruption spreads from SaaS repricing to agent-led disintermediation of payments, insurance and marketplaces, then to systemic risk via private credit (insurer-funded 'permanent capital' holding defaulting software LBO debt) and finally to the $13 trillion prime mortgage market as 780-FICO borrowers' incomes are structurally impaired. It is a scenario, not a prediction — published in February 2026 so investors can audit which portfolio assumptions won't survive the decade.

• Scenario: unemployment 10.2% in June 2028; S&P 500 down 38% cumulatively from October 2026 highs (~8000 peak, Nasdaq above 30k)
• Median American consuming 400,000 tokens per day by March 2027 — 10x the end-2026 level
• White-collar workers: ~50% of employment, ~75% of discretionary consumer spending; top 10% of earners >50% of all consumer spending; a 2% white-collar employment decline = 3-4% discretionary spending hit
• Private credit grew from under $1 trillion in 2015 to over $2.5 trillion by 2026; Zendesk's $5B direct lending facility marked to 58 cents in the scenario
• US residential mortgage market ~$13 trillion; scenario Zillow home values -11% YoY San Francisco, -9% Seattle, -8% Austin
• Scenario 10-year yield falls from 4.3% to 3.2% over four months; initial jobless claims surge to 487,000
• Labor share of GDP: 64% (1974) to 56% (2024) to 46% in the 2028 scenario; federal receipts running 12% below CBO baseline
• India IT services exports >$200 billion annually; rupee -18% vs the dollar in four months in the scenario
• Full mortgage-crack scenario: 57% peak-to-trough S&P drawdown (GFC-scale) to ~3500
• A Claude agent doing a $180,000 product manager's work for $200/month — the unit economics driving the displacement spiral

• The core mechanism is a feedback loop with no natural brake: AI capability improves, firms cut white-collar jobs and redeploy savings into AI, displaced workers spend less, margin pressure forces more AI adoption — and because it is OpEx substitution rather than capex, falling demand does not slow the buildout.
• Incumbents don't resist the disruptive technology this time — they cannot afford to: the scenario's ServiceNow cuts 15% of its workforce and funds the very AI disrupting it, while seat-based revenue mechanically shrinks with customers' AI-driven layoffs.
• The consumption math is the crux: white-collar workers are ~50% of employment but ~75% of discretionary spending, and the top 10% of earners drive >50% of all US consumer spending, so a 2% white-collar employment decline maps to a 3-4% discretionary spending hit — with a dangerous savings-buffer lag before the data confirms it.
• Agentic commerce destroys 'habitual intermediation': friction-based moats (subscription inertia, insurance renewals worth 15-20% of premiums, 5-6% real estate commissions, delivery-app home-screen loyalty) dissolve when agents compare everything 24/7, and agents route payments around 2-3% interchange into stablecoins.
• Private credit is the transmission channel to systemic risk: $2.5T by 2026 with heavy SaaS LBO exposure; the fictional Zendesk default ($5B facility marked at 58 cents) exposes that 'permanent capital' is actually annuity policyholder money at manager-owned insurers layered through opaque Bermuda reinsurance.
• The endgame question is whether prime mortgages are money good: $13T of loans underwritten on income durability for 780-FICO borrowers — 'the loans were good on day one; the world changed after they were written' — with a full crack implying a 57% GFC-scale equity drawdown to S&P ~3500.
• Government is structurally unfit for this crisis: the revenue base is a tax on human time, receipts run 12% below baseline as labor share of GDP falls from 56% (2024) to 46%, and automatic stabilizers assume workers will be reabsorbed at prior wages — which they are not.
• The investable asymmetries inside the scenario: pure AI-infrastructure economies (Taiwan, Korea) outperform massively while India's $200B IT services surplus evaporates (rupee -18% in four months); the piece's real-world instruction is to audit portfolio assumptions built on scarce human intelligence before the repricing.

Thematic Update: Modern Warfare

Modern warfare is being reshaped by three headline-driven trends — directed energy weapons finally working, cheap subsea drones threatening multi-billion-dollar naval assets, and a great-power scramble for Greenland and the Arctic — and the winners are smaller 'enablers' rather than the big primes. nLIGHT, up nearly 6x since Citrini's May 2025 entry, is the only public pure-play vertically integrated directed-energy company targeting a megawatt laser, and Iron Beam's operational deployment plus Golden Dome funding validate a spending wave barely in consensus estimates. Subsea drone offense and defense (Exail, Kraken, Oceaneering and others) and Arctic exposure (miners like Critical Metals and Energy Transition Minerals, icebreaker supply chains, Iridium's polar satcom moat) round out three new baskets built to capture accelerating defense budgets with structural tailwinds beyond headline beta.

• LASR entered at $8.47 in May 2025; stock has nearly 6x'd (from $6 to $51 over the past year), market cap ~$2.9B at ~11x trailing revenue
• HELSI contract: $171M (originally $86M, awarded May 2023) to scale from 300kW to 1MW, scheduled completion 2026
• nLIGHT Q3 2025: revenue $66.7M (+19% YoY), A&D revenue ~+50% to record $46M, defense product sales +70%, record 41% product gross margin; Q4 preliminary revenue $78-80M
• Golden Dome: $24.4B federal funding through OBBBA with $13B allocated for FY2026
• Iron Beam funding: ~$1.2B US Congressional supplemental, >2B shekel (~$500M) Rafael/Elbit contract, plus $200M award to Elbit
• Double case: FY2027 revenue toward $450M and EBITDA of $75-100M; bear case ~30% downside to ~$35 on $320M FY2027 revenue at 6x sales
• Global high-energy laser market projected at roughly $10-15B by 2030; nLIGHT's ~$165M A&D run-rate is <2% of that TAM; 5% share = $500-750M defense revenue
• US Navy ~290 deployable ships; $2-3M laser content per ship on half the fleet = $290-435M revenue opportunity
• Russia operates 40 icebreakers (8 nuclear) vs one US heavy polar icebreaker; NSR cuts transit distance ~40% and 10-15 days
• Iridium's EMSS contract: 7 years, $738.5M; Ukraine's Operation Spiderweb used 117 quadcopter drones to destroy or damage at least 40 Russian aircraft

• Directed energy hit its operational inflection in a six-week window: UK DragonFire to deploy on Type 45 destroyers by 2027 (five years early), Israel's Iron Beam entered IDF service December 28, 2025 as the world's first operational laser air defense, and the US Navy's top surface warfare officer committed to 'a laser on every ship.'
• DEW economics solve the cost-asymmetry problem: a $500 rocket forces a $50,000 Iron Dome interceptor, while laser interception costs single-digit dollars per shot with an unlimited magazine; CRS power thresholds — 100kW for drones, 300kW for cruise missiles, 1MW for ballistic/hypersonic threats.
• nLIGHT is the only public pure-play targeting the megawatt: the $171M HELSI contract (expanded from $86M, completion scheduled 2026), full vertical integration from diodes to beam directors via the 2019 Nutronics deal, an Iron Beam supply role, and a ~$200M raise to build capacity ahead of a program of record.
• LASR is up nearly 6x from Citrini's $8.47 entry in May 2025, yet the double case needs only two of three: 40%+ A&D growth toward $450M FY2027 revenue, HELSI/Golden Dome validation, and EBITDA inflecting to $75-100M; bear case ~30% downside to ~$35 — a roughly 3:1 risk/reward.
• Golden Dome carries $24.4B of federal funding through the One Big Beautiful Bill Act ($13B allocated FY2026) with no directed-energy awards made yet — an open catalyst for nLIGHT, which confirmed it is responding to RFPs.
• Ukraine's December 2025 UUV strike on a KILO-class submarine inside the enclosed Novorossiysk base proves stealthy subsea drones can penetrate defended harbors — driving spending on detection, barriers and autonomous 'guard dog' systems; Oceaneering (12x forward P/E, record DoW contract ramping 2026/27, SIIPS) is the favorite cheap play, Exail the most levered integrator.
• The Greenland trade is not about annexation: it is about accelerating Western investment in Arctic minerals (CRML's Tanbreez, ETM's Kvanefjeld after cutting Chinese ties, Amaroq), icebreakers (Russia has 40 including 8 nuclear vs one US heavy polar icebreaker; ABB's Azipod is the design lynchpin), and polar satcom.
• Iridium's risk premium should unwind: the stock lost a third when SpaceX bought Echostar spectrum, but Iridium runs the only true pole-to-pole LEO constellation, is the GMDSS backbone for the Northern Sea Route, equips every UUV Citrini researched, and won a December 2025 Space Force ground-modernization contract ahead of the EMSS airtime rebid.

Atoms vs Bits: What Can't You Prompt Your Way Out Of?

Agentic AI has forced the market to apply an 'AI Disruption Discount' to software, rotating capital from bits into atoms — the physical materials, metals and chemicals that AI, defense, aerospace and electrification all need but that no prompt can conjure. The obvious physical winners (power equipment, copper, nuclear, optics) are crowded, so the next leg is in overlooked upstream materials with oligopoly or monopoly supply, multi-year qualification moats, and converging cyclical-plus-structural demand: uranium conversion (Solstice), beryllium (Materion), titanium sponge (Osaka/Toho), aerospace aluminum, tungsten (Kennametal), superalloys, carbon fiber (Hexcel), grain-oriented electrical steel, mining capex (Komatsu), and AI electronic chemicals (Nittobo, Asahi Kasei, Entegris). These names are still priced as commodity producers even as they become critical enablers, and the re-rating 'won't be gentle.'

• SOLS Metropolis plant: ~15,000 MTU nameplate, ~20% of global conversion capacity; legacy contracts at $20/kgU vs ~$64/kgU spot; ~25% of legacy book repricing per year; $2B AES backlog; Citrini 2028E AES EBIT ~$500M and fair value $95-105/share
• Materion supplies >40% of global beryllium and >70% of freshly mined production; defense bookings +40%
• Osaka Titanium investing $210M to expand sponge capacity from 40,000 to 50,000 tonnes by 2028; export sponge volumes +26% YoY in H1; US imports >95% of its titanium sponge
• Copper/aluminum cost ratio at 4.2x vs ~3.5x switching threshold; US produced ~670,000 tons of primary aluminum in 2024 (<1% of global); no new US smelter since 1980
• Tungsten prices have tripled; Kennametal at ~$40/share and ~8.5x EV/EBITDA with Metal Cutting margin up from 6.0% to 9.6% YoY
• Power transformer lead times 128 weeks (vs 20 weeks two years ago), GSU lead times 144 weeks; transformer prices up 4-6x since 2022; GSU demand +274% since 2019
• Hexcel 2026 guidance: $2.0-2.1B revenue, EPS $2.10-2.30 (~25% growth); ~$500M incremental revenue at full OEM production rates; $600M buyback; Airbus sources >80% of its carbon fiber from Hexcel
• Komatsu at ~15x forward earnings / 8x EV/EBITDA vs Caterpillar at 31x forward / 27x EV/EBITDA; ~900 FrontRunner AHS trucks across 23 mine sites; CAT projects mining capex +50% by 2030
• Nitto Boseki +75% YTD with >70% share in high-end electronic glass fiber; operating profit +26.4% YoY to ¥14.9B; dividend nearly tripled to ¥114/share
• Gold miners' AISC averaging ~$1,600/oz vs gold above $5,000/oz; central banks bought 755+ tons of gold in 2025; F-35 is ~20% titanium by weight, F-22 ~39%

• Solstice (SOLS) is 'the cheapest nuclear play in the US': sole domestic UF6 converter (~20% of global capacity), Q4 revenue +39% YoY on only 15-20% volume growth, legacy $20/kgU contracts repricing toward $64+ spot with identical costs — Citrini sees ~60% AES margins, a $2B backlog, and $95-105/share fair value at 12-14x versus Centrus at 33x EV/EBITDA.
• Materion is a near-monopoly on beryllium (>40% of global supply, >70% of freshly mined), with defense bookings +40% and a Commonwealth Fusion Systems agreement giving a free call option on fusion.
• Western titanium sponge is a Japanese duopoly (Osaka + Toho ~80% share with the Saudi JV) trading at single-digit EV/EBITDA in a Boeing-destocking trough; the US imports over 95% of its sponge and Chinese material is blocked by defense law and 4-6 year qualification cycles.
• Tungsten prices have tripled and Kennametal is the only vertically integrated Western tungsten carbide maker with its own powder and recycling; at ~8.5x EV/EBITDA with Metal Cutting margins up 360bps (6.0% to 9.6%), 'the market hasn't figured this out yet.'
• Grain-oriented electrical steel is the grid's hidden chokepoint: power transformer lead times went from 20 weeks to 128 weeks (144 for GSUs), prices up 4-6x since 2022, GSU demand +274% since 2019, and the GOES core is 25-30% of transformer cost — JFE (6.5% silicon steel monopoly, ~5x EV/EBITDA), POSCO and Nippon Steel are the plays, with CLF's $400M government GOES contract a turnaround signal.
• Hexcel is a coiled spring: right-sized costs, 2026 guidance of $2.0-2.1B revenue and $2.10-2.30 EPS (~25% growth), ~$500M of incremental revenue when OEM rates normalize, sole-source positions (Airbus sources >80% of carbon fiber from Hexcel) and unmodeled space upside as launch cadence grows 20-25%/year.
• Komatsu vs Caterpillar is 'one of the more mispriced discrepancies in global industrials': ~15x forward and 8x EV/EBITDA versus CAT's 31x/27x, nearly 900 autonomous trucks deployed across 23 mines with zero safety incidents in 17 years, and CAT itself projecting mining capex +50% by 2030; room to double in 12-18 months.
• The December 'AI materials' upstream call is being validated in real time: WSJ confirmed Nitto Boseki's T-glass bottleneck (>70% share, stock +75% YTD, capacity booked by Apple/Nvidia vendors into 2027, dividend nearly tripled to ¥114), Asahi Kasei's PIMEL PSPI and Q Glass are the next chokepoints, and the copper/aluminum ratio at 4.2x (above the ~3.5x switching threshold) plus data centers outbidding smelters for power ($115/MWh vs the $30-40/MWh smelters need) favor aluminum incumbents like Constellium.

Semis Memo: Muscle Memory

As returns from mainstream AI 'layups' (Nvidia, Micron, SK Hynix, Broadcom) moderate, alpha migrates to less-followed layers of the semiconductor supply chain, and Citrini has hired dedicated semis analysts to hunt there. The memory supercycle structurally lifts demand for memory testing (a 16-name basket spanning testers, probe cards, handlers and sockets) because HBM must be tested at every assembly layer and the HBM4E hybrid-bonding transition raises complexity further, while semicap subsystems ('picks and shovels for the picks and shovels') capture disproportionate growth as process steps per wafer multiply. Silicon photonics re-rates legacy foundries GlobalFoundries and UMC, and a watchlist adds legacy memory squeeze names, oversold cybersecurity, SpaceX's Taiwanese supply chain, and a bullish Nvidia earnings setup now that the king is under-owned.

• NVIDIA Blackwell Ultra GB300 carries roughly $4,000 of HBM3E content per accelerator
• HBM microbump probing at 40µm pitch; MEMS probe cards supplied almost exclusively by FormFactor and Micronics Japan
• TSMC expected to spend $190B-$200B in capex over the next 3 years; Intel targeting 100k wafers/month of 18A-P + 14A capacity by 2029-30
• HBM wafer starts growing ~50% annually equal 100-200% more process steps per wafer; 1,500+ steps at 3nm and below; ~30% fab cost increase per node
• BESI hybrid bonding share potential: ~30% at Samsung, 50-60% at SK Hynix, >80% at Micron; Hanmi holds ~90% of HBM4 TCB equipment share
• SK Hynix ordered 30+ TC bonders from ASMPT and is targeting ~100 TC bonders total for HBM4
• MLC NAND global capacity projected to drop ~41.7% YoY in 2026; Macronix raised NOR prices ~30% in 1Q26
• UMC 3Q25: wafer shipments +3.4% QoQ, utilization 78%; Powertech guiding 2026 capex over NT$40B
• Cohu: 2025 Neon HBM inspection revenue raised to ~$10-11M; Eclipse handler tests up to 16 devices in parallel, -55°C to +155°C, ~3,000W power dissipation for a likely-Intel AI processor
• Citrindex Dynamic AI basket +16% YTD, +278.60% since inception; IWM has outperformed SPY by ~10% since Citrini turned constructive on small caps

• Memory testing is a structural winner of the HBM cycle: HBM requires known-good-die screening, stack-level and package-level testing, and Nvidia has reportedly forced suppliers to fully test HBM packages before foundry integration; FormFactor and Micronics Japan near-exclusively supply the MEMS probe cards that touch 40µm-pitch microbumps.
• Hybrid bonding arrives with HBM4E (2026/27): Samsung has effectively confirmed adoption with yields comparable to TC-NCF; BESI may get only ~30% of Samsung's tool demand (Semes ~70%) but potentially 50-60% at SK Hynix and >80% at Micron — while large-scale hybrid bonding threatens Hanmi Semiconductor's ~90% share in HBM4 TCB equipment.
• Subsystems scale disproportionately: HBM wafer starts growing ~50% annually translate into 100-200% more process steps per wafer, 3nm-and-below requires 1,500+ steps, and each new node lifts fab cost ~30% — favoring ICHR/UCTT-style vendors over already-consensus WFE OEMs.
• TSMC is expected to spend $190-200B of capex over the next three years and Intel could reach 100k wafers/month of 18A-P + 14A capacity by 2029-30, with Intel's 2026 capex front-loaded toward equipment — the fab buildout is another player's revenue.
• GlobalFoundries and UMC are commodity-multiple foundries quietly assembling silicon photonics and power-semi platforms (GF: Fotonix, Singapore fab, Cisco, TSMC GaN license with Navitas/ON Semi in 2026; UMC: imec iSiPP300, risk production 2026-27), setting up multiple re-ratings.
• The legacy memory squeeze is spreading: DDR4 EOL has produced panic buying (DDR4 briefly traded at a premium to DDR5), tightness is spilling into DDR3 and NOR (Macronix ~30% NOR price hike in 1Q26), and MLC NAND capacity is projected to fall 41.7% YoY in 2026 — but these names are crowded, so buy the unwind.
• Selling cybersecurity on agentic coding fears is 'stupid': AI agents are privileged identities that must be secured — CrowdStrike (Falcon Next-Gen Identity, SGNL and Pangea acquisitions), Rubrik (Agent Rewind), Cloudflare and Zscaler are beneficiaries mistakenly sold with SaaS.
• Nvidia is now the neglected trade: sideways for six months while investors chase second-order winners; China re-entry, model competition and coding-driven compute demand set up an upside guidance surprise best played with longer-dated call options — and Citrini's Dynamic AI basket (+16% YTD, +278.6% since inception) stays long semicap while short consumer electronics on unpriced memory cost pass-through.

Macro Memo: Running Hot

The US economy is re-accelerating despite a stagnant labor market because AI-driven productivity gains are increasingly showing up in output, while pro-growth pre-midterm policy (MBS purchases to lower mortgage rates, teased stimulus) and the Fed's pivot from QT to ~$40B/month of balance sheet expansion add fuel. That mix supports risk assets, argues rates are too low (hence a contrarian 2s30s flattener), and favors housing-sensitive longs like Rocket. Abroad, the 'Don-roe Doctrine' — a transactional US retrenchment into its hemisphere after the Venezuela intervention — makes LatAm (especially Chile), Southeast Asia (long Vietnam/Indonesia/Malaysia/Philippines), and a reflating China the places to be, while Iran tension plus structural fleet tightness supports long oil and especially tankers.

• Real GDP +3.8% annualized Q2 and +4.3% Q3; Atlanta Fed GDPNow forecasting 5.1% for Q4
• $200 billion in announced MBS purchases as pre-midterm housing stimulus
• Record $29.4 billion single-day Standing Repo Facility draw on Oct 31, 2025
• Fed balance sheet expansion opening at a pace of $40 billion per month
• 2-year Treasury at 3.5% with two cuts priced by end-2026 that Citrini expects removed
• Long WTI crude (CLH6) initiated at $59.15 on January 12, 2026
• VENZ 9.25% 9/15/27 bonds rallied 37% after Maduro's capture; +21% since publication of 26 Trades
• Chile produces 25% of the world's copper; Chilean inflation expected at 3% by end of Q1 2026
• HGZ6 copper up another 20% since the last memo's roll; QQQ outperformed IBIT by ~8% during the pair trade
• Moore Threads and MetaX traded up 6x on their December market debuts; NGF27 natural gas bought at 4.42 after falling from above 5.2

• AI productivity is the memo's core macro call: real GDP grew 3.8% (Q2) and 4.3% (Q3) annualized with GDPNow at 5.1% for Q4 despite stalled job creation — Citrini argues output per hour is rising because AI tools are working, and 2026 will make it obvious.
• History supports the 'weak labor, strong GDP, rising market' combination: 1992, 1975, 1958 and the post-WW2 demobilization years all saw it.
• Housing is the administration's midterm lever: $200B of announced MBS purchases are 'just the beginning,' with the GSE MBS purchase-cap removal still possible; plays include RKT, UHAL, RH, HD and LOW.
• The Fed has crossed from base-money destruction to creation: after repo stress including a record $29.4B single-day SRF draw on Oct 31, 2025, balance sheet expansion is running ~$40B/month — directionally positive for risk assets.
• Contrarian 2s30s flattener: Powell can remain a governor after his chair term, the steepener is crowded, and the 2-year at 3.5% prices two 2026 cuts Citrini thinks get removed as nominal growth runs hot; also implies a stronger dollar.
• Tankers offer asymmetric, multi-catalyst exposure: record Iranian crude on tankers, an aging fleet, tightening shadow-fleet sanctions and multi-year-high VLCC rates work even without a Hormuz disruption; long WTI from $59.15 with a stop at entry.
• Don-roe Doctrine: post-Venezuela (VENZ 9.25% 9/15/27 bonds +37% after Maduro's capture), the US retrenches into its hemisphere — Chile (25% of global copper, Kast inaugurated March 11, 2026, inflation reaching 3% by end-Q1) is the favorite; long VIMP replaces long Indonesia/short Vietnam.
• Bitcoin flip: after the long QQQ/short IBIT trade returned ~8%, Citrini went long IBIT on January 5, expecting BTC back above 100K with a possible retest of all-time highs.

26 Trades for 2026: A Thematic Watchlist for the New Year

Citrini's annual 'Twenty-Something Trades' exercise drops the pretense of a precise year-ahead outlook and instead publishes 26 thematic trade ideas for 2026 as a watchlist designed to minimize intellectual blind spots. The 2025 edition validated the format — over half the ideas beat the S&P 500 and single-stock picks averaged +26.5% — so the 2026 list continues it, delivered as a long PDF with accompanying thematic basket spreadsheets.

• 26 trade ideas for 2026, each about 6 pages in the PDF
• More than half of the 25 Trades for 2025 outperformed the S&P 500
• More than 80% of the 2025 ideas had positive returns
• 2025 single-stock picks averaged +26.50% equally weighted
• Single-name average beat the S&P 500 by more than 10 percentage points
• Most 2025 drone basket names roughly doubled

• The 26 Trades format is explicitly a thematic watchlist, not a forecast: 26 ideas published without the 'this has to happen or else I'm wrong' constraint, intended to surface differentiated views and minimize blind spots.
• 2025 scorecard: slightly more than half of the prior year's ideas outperformed the S&P 500 and more than 80% posted positive returns.
• 2025 single-stock picks averaged +26.50% equally weighted, beating the S&P 500 by more than 10 percentage points.
• Standout 2025 winners: most drone basket names doubled, Ukraine normalization picks 'crushed it,' South Africa beat Turkey on the gold rally, Allient doubled on Rockwell's return to growth and robotics tailwinds, and Citrini correctly anticipated the success of Arcutis (ARQT) Zoryve.
• Acknowledged 2025 loser: the remittances idea, which conflicted with the Trump administration's deportation ramp.
• The full 2026 content (each trade ~6 pages) is distributed as a PDF with a separate thematic baskets spreadsheet; this post is the paywalled front door.

Carving Up the TPU: Leftovers for Jensen or Just Gravy on the AI Trade?

Gemini 3's success and reported TPU adoption by Anthropic and Meta have vindicated Citrini's August 'Long GOOGL' call, but the market's read that this is simply bearish Nvidia is too crude. A TPU scale-up is more bullish for Google and for the shared supply chain than it is bearish for Nvidia, because both giants will compete for the same TSMC CoWoS capacity, HBM, advanced packaging, PCBs and optics, boosting suppliers' pricing power. Citrini believes the market's reaction will effectively force Google to sell TPUs as a merchant vendor, which turns a walled-garden asset into revenue and optionality. The trade is therefore a basket of first-order TPU suppliers (packaging, memory, dense PCBs, optical circuit switch players) plus second-order winners like Apple (as a prospective Google TPU/Gemini customer), SiTime and MACOM riding the 800G-to-1.6T optical transition.

• Google internal TPU forecast for next cycle revised from ~2 million to ~4 million units
• 1.6T optical module demand expected to jump from ~3 million modules in 2025 to ~20 million in 2026, with Google consuming 6-10 million
• SiTime MEMS oscillator content per 1.6T module roughly doubles to ~$2, implying ~$40M of 2026 TAM from 1.6T modules alone
• SiTime guiding to $100-103M Q4 revenue at ~60% gross margin, implying >$400M annualized run-rate; quarterly revenue +45% YoY
• SiTime Apple 5G modem exclusive MEMS oscillator win estimated at ~$240M of 2027 revenue
• MACOM holds >60% share in 200G photodiodes used in 1.6T modules
• Fluidstack/Terawulf TPU deployment backed by a $3.2 billion Google backstop with Google taking 14% equity
• Hyperscaler capex minus D&A gap has exploded to over $200 billion in the past twelve months
• Nvidia B200: 192 SMs totaling 768 tensor cores and 24,576 CUDA cores (architecture comparison)
• Citrini reduced from +100% to +80% net long into year-end, significantly hedged

• Citrini views the TPU narrative as more bullish for GOOGL than bearish for NVDA: Google and Nvidia will fight for the same TSMC CoWoS capacity, so the shared supply chain gains pricing power regardless of who wins share.
• Switching costs are the underrated moat: moving from CUDA/PyTorch to JAX is non-trivial and expensive, and Google's 10 years of TPU optimization is not easily replicable by third parties — Google itself is still one of Nvidia's largest customers.
• The market's reaction will reflexively force Google to sell TPUs; deals with Anthropic, Meta and Fluidstack (a Terawulf-hosted deployment with a $3.2B Google backstop and 14% equity) show the merchant/franchise decision is largely made.
• Citrini's 25-name TPU supplier basket has been the best performing AI category YTD and still trades cheaper than the GPU-tilted 125-name Phase 1 AI Infrastructure basket, leaving room for a re-rating in PCBs, substrates and networking.
• Google's OCS architecture (Lumentum-built optical circuit switches using MEMS mirrors fabbed by Silex/Sai MicroElectronics, with Huber+Suhner the other key player) replaces power-hungry electrical switching, a key structural difference from Nvidia clusters.
• SiTime: Google's TPU forecast doubling to ~4M units helps drive 1.6T optical module demand from ~3M (2025) to ~20M (2026); at ~$2 of MEMS timing content per module that is ~$40M of 2026 TAM, of which SiTime may capture 25-50%, on top of a ~$240M 2027 Apple 5G modem revenue opportunity.
• Apple is flagged as a surprising second-order TPU winner: TPU pods' static determinism suits a fast, reliable cloud-inference Siri, and Apple is not locked into Nvidia's ecosystem.
• Macro caution: hyperscaler capex now exceeds D&A by over $200 billion in twelve months, overstating earnings versus cash flow, and buyback capacity is eroding — a narrative Citrini expects to surface in a downturn; positioning cut to +80% net long and hedged.

Macro Memo: Mo-vember

With the shutdown ending and a deluge of delayed data incoming, Citrini expects continued labor deterioration on private data, a cautious Fed that skips the December cut, risk-off pressure on speculative assets (crypto, quantum, miners) into mid-December, and then a Santa rally as 2026's fiscal impulse ($380-545B of OBBBA tax savings plus possible tariff dividends) gets priced. The durable structural call is that AI datacenter load exhausts every 'spare grid capacity' workaround — storage, demand response, BYOG microgrids — forcing real generation buildout that transforms utilities' earnings runway; hence a new long XLU/short SPY pair, alongside long QQQ/short IBIT (reflexive Bitcoin to ~75K), profit-taking in natural gas (+15%) and gold puts (+100% on half), rolling copper long on undermodeled datacenter demand, and re-entering the SOFR H6H7 flattener on aggressive cuts under Powell's successor.

• Market prices 43% odds of a December 25bp cut; Citrini expects a hold
• OBBBA 2026 tax savings estimated $380-545B (~$75B incremental new savings)
• Tariff revenue ~$30B/month (~$360B/year), ~20% of the FY2025 deficit; SCOTUS upholding odds priced at 25%
• Natural gas: NGZ5 long from ~$3.9, 3/4 taken off for +15%; BTM demand +1 Bcf/d YoY
• Copper: 5GW → 25GW/year of DC additions ≈ 500kt incremental demand at 25kt/GW; HGZ5→HGZ6 roll booked +7%
• Gold: GLD 350 Dec 31 puts (bought 10/15), half closed for ~+100%
• SOFR H6H7: curve prices ~50bps of cuts Mar26-Mar27; Citrini targets 75bps, fat tail 100bps
• BYOG requires ~1.5-2x load in firm capacity; Stargate 1.2GW load vs ~360MW planned generation
• QQQ/IBIT: BTC target ~75K, stop on close above 100K; Nikkei short watchlist trigger below 50K JPY (stop 51.5K)
• Sentiment: AAII 39% bullish; VIX ~16 at recent high vs >25 at late-90s peaks; 62% of S&P above 200dma; ES 50-day EMA 6722

• The Fed likely skips December: market prices a 43% cut probability but Citrini's tally of FOMC member views suggests lower odds, and a hold should push additional cuts into 2026 pricing — fueling the re-entered SOFR H6H7 flattener (curve at ~50bps vs Citrini's 75bps base, 100bps tail).
• The 2026 fiscal impulse is large and underappreciated: $380-545B of OBBBA tax savings (~$75B incremental beyond TCJA extension) plus possible tariff-dividend checks mean the wall of worry likely resolves into a Q1 2026 rally, with a Santa rally expected as this gets priced.
• Grid substitutes are temporary — 'storage fixes shape, not supply' — and BYOG microgrids require ~1.5-2x load in firm capacity (Stargate: 1.2GW load vs ~360MW planned generation), so persistent AI load ultimately forces real generation buildout through utilities: go long XLU/short SPY, with defensive beta as the hedge.
• Datacenter copper demand is massively undermodeled: moving from 5GW to 25GW of annual DC additions implies ~500kt of incremental copper demand at a conservative 25kt/GW — equivalent to a year of China entering the WTO — while US scrap export restrictions add a resource-nationalism bid; rolled HGZ5 to HGZ6 (+7%).
• Behind-the-meter gas demand is already visible but miscategorized: EIA books it as commercial demand, and summer res+comm gas demand ran +1 Bcf/d over 2024 without unusual weather — early confirmation of the structural gas thesis; 3/4 of the NGZ5 long booked at +15%.
• Bitcoin trades on pure reflexivity and MicroStrategy can no longer act as buyer of last resort: long QQQ/short IBIT targets BTC ~75K with a defined stop on a close above 100K.
• The contrarian gold-put trade (GLD 350p bought 10/15 against crowded macro momentum longs) paid ~+100% on the half closed, validating the view that gold's fundamental narratives were exaggerated.
• Sentiment is far from euphoric — AAII bulls at 39%, VIX ~16 at the high versus 25+ at late-90s peaks, 62% breadth — supporting the 1998 analog of a bull that grinds higher with rising volatility; ES's 50-day EMA at 6722 is the key tactical level after holding all of May-October.

I Can See Clearly Now... The Rain Is Gone?

Fifteen months into the Healthcare Innovation basket, the feared Trump-era policy damage never materialized — funding cuts were walked back, landmark approvals landed, diagnostics reimbursement improved — yet healthcare positioning sits at 20-year lows with post-COVID-high short interest and depressed multiples, while guidance beats run near record highs and 2025 M&A tracks ahead of any year on record. Citrini argues this is the start of a multi-year upswing led by small-cap medtech crossing into profitability (the IRTC/ATEC re-rating template applies to the STAX names), insurers' medical-cost crisis funneling business to value-based care (TOI, ALHC, new add ASTH), GLP-1 second-order winners (AIRS, WW, NGVC, LLY vs shorts in low-end food), and a diversified life-sciences upcycle sleeve (TKNO, NOTV, AVTR, plus new FTRE and CYRX).

• Healthcare long positioning around 20-year lows; short interest testing post-COVID highs
• 2025 healthcare M&A deal count tracking substantially ahead of any prior year on record
• TOI trades at a 50%+ discount to peers such as Privia (PRVA); guided to sustainable profitability before end of 2025
• CVRX BAROSTIM saves at least $9,000 per patient vs standard of care (US and European studies)
• ALHC exceeded full-year EBITDA guidance in Q2 alone, then tripled analyst consensus in Q3
• ASTH: ~10% free-cash-flow yield, 20+ payers across 13 markets, path to 25%+ revenue growth
• XGN at just over 4x trailing revenue vs diagnostic peer group now above 9x (both ~4x in July)
• STIM has invested nearly 2x Brainsway's R&D over the last five years; TMS search growth at multi-year high
• AIRS: return to 2024 utilization implies over $60M of free cash flow over three years
• NGVC customer median income near $100,000; nearly one-third of SKUs are vitamins/supplements

• The feared policy damage never came: funding cuts walked back, pharma-government manufacturing deals struck, landmark approvals granted — yet healthcare long positioning is at ~20-year lows with post-COVID-high short interest, creating the setup for a multi-year upswing.
• 2025 healthcare M&A deal count is tracking substantially ahead of any prior year on record, and depressed small-cap valuations make several basket names (XGN, TOI, CVRX) plausible strategic targets — Labcorp has explicitly named autoimmune diagnostics as an acquisition focus and JNJ is CVRX's largest shareholder.
• The profitability-inflection template drives the STAX thesis: iRhythm and Alphatec re-rated from peer discounts to premiums upon turning profitable, and all four STAX names are guided to cross that threshold in coming quarters.
• Insurers' MLR crisis is the value-based care demand engine: Cigna, Centene, CVS and Elevance all named VBC as core strategy and Molina blamed oncology drug costs specifically — TOI, the vertically integrated oncology VBC leader, trades at a 50%+ discount to Privia despite guiding to profitability by end-2025.
• ALHC validated the VBC bet by exceeding its entire annual EBITDA guidance in one quarter (Q2) then tripling Q3 consensus; new addition ASTH offers ~10% FCF yield, 20+ payer relationships in 13 markets and a path to 25%+ revenue growth at the lowest multiple in VBC.
• XGN's peer group re-rated from ~4x to >9x trailing revenue since July while XGN itself still trades just over 4x — the market is paying up for AI-leverageable diagnostic data moats and XGN owns the autoimmune niche.
• GLP-1 second-order effects are inflecting: aesthetics demand is rebounding (Needham's 250+ patient survey, WSJ/McKinsey coverage) benefiting AIRS (>$60M 3-year FCF potential at 2024 utilization), while low-end food shorts (SHAK, DPZ, EAT, HRL) keep working and NGVC's supplement-heavy, ~$100k-income customer base makes the sell-off a buying opportunity.
• Life sciences recovery is diversified for beta rather than timed: ICON's backlog has stabilized three quarters running, TMO/DHR guide above consensus on reshoring, NVDA signed drug-discovery partnerships with LLY and Verily — sleeve now TKNO, NOTV, AVTR, plus new FTRE and CYRX.

Power Struggle: Natural Gas

The US natural gas market is flipping from fifteen years of price-insensitive supply-push to demand-pull: ~35 Bcf/d of high-confidence demand growth from LNG exports and AI/electrification power generation over the next decade collides with only 15-20 Bcf/d of supply growth at current strip prices, because the Permian's 'free' associated gas is oversubscribed (~7-10 Bcf/d left) and dry-gas producers need $4-5+/MMBtu to dispatch growth from deteriorating rock. With early-2030s NYMEX futures at ~$3.20 in steep backwardation, the curve is mispriced and the repricing gets more violent the longer it waits; Citrini positions in a broadened basket of deep-inventory upstream E&Ps (US and Canadian), midstream, utilities, compression, power-generation equipment and E&C, most trading at 8-12x earnings.

• ~35 Bcf/d high-confidence demand growth next decade vs 15-20 Bcf/d supply growth at current strip; expansive demand view ~70 Bcf/d (+70%)
• US data centers: ~17GW consumed in 2022 → ~130GW by 2030 (~12% of US demand)
• GE Vernova Q2 2025: 9GW new gas turbine orders, 29GW backlog, 25GW slot reservations (~55GW ≈ 6-9 Bcf/d of burn)
• Gas turbine prices up ~50% since summer 2024; turbines last 30+ years vs GPUs replaced 10+ times
• LNG: ~14 Bcf/d operating capacity; under-construction projects add ~5.3 Bcf/d by 2027; ~8 Bcf/d FID'd in 2025 (+~3.5 expected); developers seek ~40 Bcf/d more (~$260-340B capex)
• Permian: ~7-10 Bcf/d of associated gas growth remaining; >$6/MMBtu needed for gas-directed drilling; gas only ~10% of revenue, ~5-7% of cash flow for top operators
• Dry gas trigger prices: ~$4.00/MMBtu for maintenance, $5+/MMBtu for meaningful growth (EQT pursuing $5 fixed-price deals)
• Enverus 2030 inventory: Haynesville ~6 years, Marcellus <10 years of 'good' inventory
• Early-2030s NYMEX at ~$3.20/MMBtu; Transco Zone 6 spiked to $97/MMBtu in January 2025
• PJM ATC power prices rose from $0.04/KWh (2024) to ~$0.06/KWh ≈ ~$300/year higher household bill; basket names at 8-12x earnings

• The gas market is flipping from supply-push to demand-pull: ~35 Bcf/d of high-confidence decade-ahead demand growth (LNG + power) versus 15-20 Bcf/d of supply growth at current strip — and a fuller demand view approaches 70 Bcf/d, a 70% increase on current production.
• 'The Permian will take care of it' is the consensus cop-out and it's wrong: only ~7-10 Bcf/d of associated-gas growth remains, gas is just ~5-7% of Permian operators' cash flow, and gas-directed Permian drilling needs >$6/MMBtu — associated gas doesn't respond to gas prices at all.
• Dry-gas producers are openly naming their price: Aethon says meaningful development 'starts with a $5-handle,' Expand pegs marginal Haynesville breakeven at $3.50 with materially higher needed for growth, and Comstock is ~3 years from legacy inventory exhaustion — these firms represent ~30% of US dry gas.
• The early-2030s NYMEX strip at ~$3.20/MMBtu in steep backwardation is self-defeating: cheap-forever pricing keeps demand projects penciling, so the longer the back of the curve stays low, the more violent the eventual repricing.
• Turbine orderbooks lock in demand: GE Vernova's 9GW of new Q2 2025 orders, 29GW backlog and 25GW of reservations (~55GW total) imply 6-9 Bcf/d of future burn, and reconstructed delivery schedules point to ~1 Bcf/d of new gas demand per year for a decade.
• LNG stacks on top: operating capacity of ~14 Bcf/d will more than double, the Trump administration FID'd ~8 Bcf/d in one year (more than Biden's whole term), and developers want another ~40 Bcf/d (~$260-340B of liquefaction capex) — while tolling contracts push price risk onto buyers.
• Geology is the quiet second engine of the thesis: cores of the Haynesville and Marcellus are drilled up (~6 and <10 years of good inventory by 2030 per Enverus), F&D costs are rising after 15 years of declines, and full-cycle Appalachian costs exceed $3/MMBtu with MVP transport.
• The basket shifts up the supply chain from construction winners (AGX doubled YTD) to commodity-levered upstream with deep inventory (e.g., EQT, Comstock, Canadian E&Ps) plus midstream, utilities, compression and E&C — most trading at 8-12x earnings.

Macro Memo: Red October

October's tariff flare-up (China's rare-earth export controls, Trump's 100% tariff threat) is escalation-to-negotiate, not the end of the bull market: both sides will repeatedly go to the brink and seek off-ramps, constrained by debt, demographics, recession risk, midterms and defense. Meanwhile the K-shaped economy keeps 'slowing without breaking' — private data show deteriorating jobs during the shutdown data blackout — which supports continued Fed cuts, while the AI bull market at 77% above its 10-year moving average is still well short of the 100-120%+ extremes that marked 1929, 1987, 2000 and Japan 1989. Citrini holds course: rare earths as a fiscal-subsidy watchlist trade (+193% in 5 months), tactical index puts as October insurance, and vigilance on emerging credit cracks (Tricolor, First Brands, subprime auto).

• Rare earths watchlist +193% in roughly 5 months
• S&P 500 currently 77% above its 10-year moving average; historic tops occurred at 100-120%+ (8150 = 120%)
• Trump threatened 100% tariffs on Chinese goods after China's Oct 9 REE export controls
• Dallas Fed breakeven job growth estimate ~30,000/month (others ~10,000)
• US imported only $170M of rare earth elements last year; F-35 uses ~450kg REE, 110 jets ≈ 50 tonnes vs 10,000 tonnes US annual demand
• Neo Performance: 2K tonnes scaling to 5K (plans 20K); 2K tonnes ≈ $20M EBITDA; Solvay could serve ~30% of European demand
• Japan's rare-earth reliance on China cut from 90% to 60%, heading to 50%
• First Brands: at least $2B owed to factoring financiers unaccounted for; auto ABS spreads +30bps since early September
• Crypto liquidation cascade losses estimated at $20-40B, mostly retail
• Positions: SMH January 2026 puts sized up, SPX November puts added, long TLT Dec 95 calls (half sold)

• China's rare-earth export controls are a negotiating card played before it expires, not a bubble-popping strike: Citrini expects tit-for-tat escalation into China's Oct 20-23 plenum, then deal-making before/at the Trump-Xi APEC meeting — a view distinct from the complacent 'TACO' narrative.
• The rare-earth basket is up +193% in ~5 months, but it is purely a fiscal-subsidy macro trade: the US imported just $170M of REEs last year, the defense angle is overstated (110 F-35s ≈ 50 tonnes vs 10,000 tonnes of US annual demand), and many watchlist names will never produce revenue.
• Bubble check: the four great tops (1929, 1987, 2000, Japan 1989) all peaked >100% above the 10-year moving average with 120% the lowest threshold; the S&P 500 is at 77% (8150 = 120%), so significant upside can remain before a true mania top.
• Private data during the shutdown blackout show continued labor deterioration — two negative ADP months (biggest in two years), decade-low hire/quit rates — but breakeven job growth may have fallen to ~10-30k/month, muting the unemployment impact.
• Rent's 'integer problem' anchors services inflation: no matter how rich the top decile is, they occupy one apartment, so softening wages contain rent CPI; bond markets price at least two more 25bp cuts despite the data blackout.
• Credit cracks are the canary: Tricolor and First Brands collapsed weeks apart with ~$2B missing at First Brands, subprime auto ABS spreads widened 30bps, repossessions are at 2009 levels, and Carvana's mid-2023 ABS vintages are losing at twice rating-agency expectations.
• An estimated $20-40B of mostly-retail crypto losses in Friday's liquidation cascade could have negative wealth-effect implications given how levered speculative spending has been.
• Positioning is insurance, not capitulation: sized-up SMH Jan 2026 puts plus SPX Nov puts, still long TLT Dec 95 calls and the H6H7 flattener, and the USDJPY short is abandoned on Takaichi's pro-growth agenda; 5-10% pullbacks are 'natural' wall-of-worry rebuilding.

Stargate: A Citrini Field Trip

Citrini sent a drone (with Hunterbrook Media) to OpenAI's Stargate site in Abilene, Texas to physically verify the AI datacenter buildout, and concluded the scale — eight 'ludicrous buildings' totaling ~4M sq ft and 1.2GW, a footprint the size of lower Manhattan, one of tens-to-hundreds of similar megaprojects — makes the industrial supply chain around the data hall the next leg of the trade. Because reliability trumps everything, these sites run on natural gas (simple-cycle turbines available in 12-18 months vs 5+ years for combined cycle and 10+ for nuclear), and with >30 Bcf/d of demand chasing ~10 Bcf/d of 'free' Permian gas growth, gas prices must move structurally higher. The actionable takeaway is 'more power, more natural gas, more thermal management and more construction': a 50-name high-confidence (plus 200 low-medium) beneficiary list across power, cooling, MEP, building envelope, fiber and construction services, with EQT added to the portfolio.

• Stargate Abilene: 8 buildings, ~4 million sq ft, 1.2GW total power capacity by mid-2026; targeted full size ~5GW
• Behind-the-meter plant: 10 turbines (5x 38MW Titan 350 + 5x 34MW GE LM2500) = 361MW continuous
• Lead times: simple-cycle turbines 12-18 months; combined-cycle 5+ years; nuclear 10+ years; UPS systems 18-24 months
• >30 Bcf/d of gas demand lined up vs ~10 Bcf/d of 'free' Permian growth
• EQT fixed-price supply negotiations at ~$5/MMBtu, ~70% above Henry Hub spot
• Captured Permian flared gas would power only ~5% of the ~5GW complex
• Prysmian $118M medium-voltage cable contract; substation expanding from 200MW/138kV to 1.2GW with 345kV
• 624 cooling towers across eight buildings (39 per side per building)
• Each major project uses 5-10 million sq ft of insulated metal panels
• Beneficiary list: 50 high-confidence and 200 low-medium-confidence public companies across six baskets

• Seeing is believing: Stargate Abilene alone will be eight buildings, ~4M sq ft and 1.2GW by mid-2026 — a footprint the size of lower Manhattan — and it is one of tens-to-hundreds of similar megaprojects (Hyperion, Colossus 2, Fairwater, Project Rainier), making this the largest US industrial investment cycle since WWII.
• Natural gas is the only dispatchable answer: simple-cycle turbines ship in 12-18 months versus 5+ years for combined-cycle and 10+ for nuclear, so Stargate runs 10 GE/Caterpillar simple-cycle units (361MW continuous) and xAI Memphis runs 30+ — reliability beats both efficiency and net-zero pledges.
• US gas demand is 'backed by committed infrastructure,' not estimates: >30 Bcf/d of demand from LNG exports plus data centers is chasing only ~10 Bcf/d of remaining 'free' Permian associated-gas growth — the consensus 'Permian will take care of it' is wrong.
• EQT is negotiating fixed-price gas supply at ~$5/MMBtu — ~70% above Henry Hub spot — signaling hyperscalers will pay up for flow surety and producers finally get compensated for dispatching growth, a radical change in supply dynamics.
• Meta is telegraphing a volatile gas/power future: it added legendary gas trader John Arnold to its board, is funding >2GW of new Louisiana gas generation, formed power-trading arm Atem Energy, and sited Hyperion at the crossroads of four major gas pipelines.
• Supply-chain scarcity cascades down tiers: UPS lead times have hit 18-24 months, Powell Industries' 3-year backlog should lengthen further, Prysmian signed a $118M cable contract for Abilene alone, and Tier 2/3 vendors will catch orders as Tier 1 lead times blow out.
• The thermal layer is enormous and quantifiable: 39 cooling towers per side of each building means 624 school-bus-dwarfing HVAC racks across the eight Abilene buildings, with Vertiv the clear leader and Schneider a distant second.
• Captured Permian flared gas would cover only ~5% of the full ~5GW Stargate complex, and data centers siphoning associated gas is detrimental to LNG exporters — the gas market is a closed loop that pulls forward a natural gas bull market.

Single Stock Long Thesis: Acuren (TIC)

Acuren (TIC US) is a mispriced, multi-thematic toll road on the AI/infrastructure buildout: its core non-destructive testing business converts stressed, can't-go-offline infrastructure (data centers, grids, pipelines, aging power plants) into >90% recurring, mandated inspection revenue, and its opportunistic $1.7B acquisition of NV5 adds hyperscaler data-center MEP design, grid engineering, geospatial data and digital twins just as those businesses inflect. Because of SPAC stigma, misperceived cyclicality and merger complexity, the combined entity trades under 8x EBITDA versus peers at 11-30x; with Sir Martin Franklin running the proven APG playbook (guidance update, investor day, re-rating), Citrini sees 100-200% base-case upside and a path for shares to double or triple in 18-36 months.

• TIC combined entity valued just over $3B, under 8x consensus NTM EBITDA
• NV5 acquired for $1.7B, <10x EBITDA; NV5 peers valued 15-30x EBITDA
• TIC peers SGS/Intertek/UL Solutions trade 11-18x EBITDA; TIC has traded 4-8x
• ASCE: $9.1T US maintenance need vs $5.4T funding = $3.7T shortfall
• >90% of core Acuren revenue recurring; service <2% of customer operating expenses; 75% fewer safety incidents than industry average
• NV5 data-center unit: <$50M 2024 revenue, management target $400M+ annually; >1GW designed/commissioned; ~300MW client capacity unlocked in 2024
• NV5 standalone 2028 revenue target $1.6B; $20M identified cost synergies (~90bps margin)
• Scenario: $600M+ combined EBITDA in 2028, ~30% annualized 4-year growth, 20%+ above consensus
• APG analog: 250% 3-year return post-merger; 30%+ annualized over 5 years; 40%+ move in weeks after investor day
• Base case 100-200% upside; shares seen doubling or tripling over 18-36 months

• Acuren is a 'toll road' on infrastructure stress: >90% recurring revenue, mandated by insurers/codes/uptime guarantees, costing customers under 2% of opex — small, predictable inspection dollars protecting very large uptime dollars.
• The valuation gap is stark: combined TIC trades below 8x consensus EBITDA versus TIC peers at 11-18x (SGS, Intertek, ULS) and NV5 peers at 15-30x (Willdan, Stantec, AECOM, Trimble) — and MEP peer Legence just IPO'd near $5B on roughly NV5's standalone EBITDA versus $3B for all of TIC.
• The ASCE's $3.7 trillion US infrastructure maintenance shortfall ($9.1T need vs $5.4T funding) plus AI-driven deferral of power-plant retirements structurally accelerates demand for life-extending NDT work.
• NV5 was bought at an inflection: >1GW of data-center design/commissioning completed, ~300MW of client computing capacity unlocked in 2024, and management targets 10x growth in the data-center unit from <$50M to $400M+ annual revenue.
• Cyclicality fears are misplaced — revenue fell only marginally in 2008, 2015-16 and 2020 because exposure is downstream (pipelines, refiners, power) and non-discretionary; 2024 softness was deliberate low-margin contract exits while margins expanded 100bps+ to 17%.
• Consensus implicitly assumes legacy TIC revenue shrinks >20% by 2028 if NV5 hits its $1.6B target inside ~$2.5B combined estimates — a sharp contrast to consistent historical growth, making estimates look badly mis-set.
• The Martin Franklin/APG playbook is the catalyst map: APG returned 250% in three years after its transformational Chubb deal and gained 40%+ in weeks after its investor day; TIC's CFO came from APG and an investor day plus November guidance are imminent.
• Scenario math: NV5 at target + legacy TIC at 8% growth + 20% combined margin yields $600M+ 2028 EBITDA (~30% 4-year CAGR, 20%+ above consensus); a peer-multiple re-rate alone implies 100-200% equity upside.

Macro Memo: Summer is Over

The US economy is slowing without breaking — labor demand is sagging (more unemployed than job openings for the first time since August 2021) while tariff revenue (~$360B annualized) is quietly halving the primary deficit and supporting bonds. Citrini's key insight is that the Trump administration cares less about Treasury yields than mortgage rates ahead of the 2026 midterms, and will likely lift post-GFC caps on GSE MBS holdings — unleashing up to $1 trillion of quasi-QE MBS demand that compresses mortgage spreads without touching the Treasury market. That favors mortgage originators like RKT and long-duration bonds, while seasonal September/October fragility argues for a tactical equity short and long natural gas on electricity-price inflation.

• Tariffs on track for ~$360B annualized revenue; FY2025 deficit ~$1.7 trillion
• Primary deficit ~2% of GDP; tariffs cut it roughly in half
• Job-opening rate back below 4.5%; more unemployed than openings for first time since Aug 2021
• 10Y Treasury anchored ~4.2%; 30Y range-bound 4.75%-5.00% all summer
• Lifting GSE caps could free up to $1 trillion of incremental MBS demand with 5x leverage
• Trade: Long NGZ5 @ 3.945, stop 3.545, target 5
• Trade: Short H6H7 @ -55bps, stop -45bps, target -90bps
• Trade: Short ES @ 6450, stop 6525, target 6240 (~3:1 reward/risk)
• Long TLT December 2025 95 calls, cost basis 0.40

• For the first time since August 2021 there are more unemployed Americans than job openings, and the job-opening rate is back below the 4.5% threshold that historically precedes further rises in unemployment.
• ADP has proven more reliable than the BLS: post-revision NFP growth converged onto the private ADP series, and Waller flagged that September 8 annual benchmark revisions will likely push official job creation down further.
• Tariffs are quietly fixing the fiscal picture: ~$360B annualized revenue against a ~$1.7T FY2025 deficit, cutting the ~2%-of-GDP primary deficit roughly in half — so a Supreme Court tariff overturn would be net negative for bonds.
• The administration's real target is mortgage rates, not Treasury yields: lifting GSE MBS caps could unlock up to $1T of incremental MBS demand — quasi-QE via housing credit that sits entirely within Treasury/FHFA authority, timed one year before the midterms.
• Prefer mortgage originators (RKT) over homebuilders: existing-home inventory returning to pre-pandemic levels erodes builders' 'only game in town' advantage while lower rates plus tighter spreads revive origination volumes.
• The secular 'four Ds' — Demographics, De-globalization, Defense, Debt — are mutually reinforcing and floor inflation and long rates, which is why ultra-long global yields are blowing out while the 10Y stays anchored near 4.2%.
• September is the only month with a negative long-run average US equity return, and crash risk clusters in March/September/October around fiscal deadlines and thin liquidity — justifying a tactical short ES at 6450 targeting 6240 with a 6525 stop.
• Compressed bond volatility (30Y stuck in a 25bps range all summer) tends to resolve violently to lower yields; stay long TLT December 95 calls at a 0.40 cost basis.

State of the Themes: Long GOOGL, Asian Outperformance and the Age of Virality

Citrini's quarterly State of the Themes argues that 2025's thematically dispersed market rewards selectivity: GPT-5's incremental improvement confirms AI is shifting from infrastructure builders to implementers, which favors Google (research depth, TPUs, ad monetization at 20x forward P/E) and data-and-distribution owners like fintech super-apps, while the remaining infrastructure upside sits in specific bottlenecks (interconnects, memory, AMD) and cheap Taiwanese suppliers. Simultaneously, fiscal policy is the other dominant driver — the AI Action Plan and OBBB tax provisions subsidize the US buildout, Korea's 'KOSPI 5000' governance push warrants broad long exposure, and beaten-down solar offers asymmetric recovery. With rate cuts imminent and SMID caps deeply underowned, Citrini re-launches its Small Themes basket around rate-sensitive names and 'virality'-driven retail winners, while pressing high-conviction healthcare (STAX) and GLP-1 second-order positions.

• AMD +127% since 4/8/25 close vs NVDA +87% and SMH +64%
• Google ~20x forward P/E; $200B ad revenue vs OpenAI <$4B consumer revenue
• JPC Connectivity 13.6x EV/LTM EBITDA vs Amphenol 28.0x; Lite-On 10.7x EV/EBITDA vs MPWR 40x
• Tariffs annualizing ~$360B of federal revenue vs ~$1T 2024 primary deficit
• AI Action Plan earmarks ~US$50B; each public dollar historically crowds in 3-7 private dollars
• OBBB: ~$210M year-one tax shield per $1B of qualifying kit; chip manufacturing credit raised 25% to 35%
• Kaspi (KSPI) at 8.3x forward earnings with 61% YoY revenue growth
• Inseego expected ~$200M 2026 revenue, valuation <0.8x sales
• WW International trades at <5x EBITDA with EBITDA growth over 60% YoY
• Community picks after one month: long basket +11.4%, short basket -7.8% (ABVX +600%)

• GPT-5's incremental improvement confirms the shift to AI 'Phase 2': gains from pure LLM scaling will moderate while LLM + reinforcement learning + data compound, so positioning should favor implementers over infrastructure builders.
• Google is the mispriced AGI call option: $200B of search/video ad revenue vs OpenAI's <$4B consumer revenue, 6-month Gemini iteration cycles, TPU vertical integration, and only ~20x forward P/E; Citrini's math implies $1T+ of unpriced AGI optionality.
• Selectivity beats broad AI beta: AMD +127% since 4/8/25 vs NVDA +87% and SMH +64%, because inference is memory-bound and MI300X is the only at-scale second source (Azure and Meta already deploying).
• The Taiwan/US discount gap is extreme — JPC Connectivity trades at 13.6x EV/LTM EBITDA vs Amphenol at 28x despite nearly identical 3-year sales growth (21.724% vs 21.673%); Citrini sees no good reason for the 2x spread.
• Fiscal and AI now intersect: the AI Action Plan earmarks ~$50B, OBBB gives ~$210M year-one tax shield per $1B of datacenter equipment and lifts the section 48D chip-manufacturing credit from 25% to 35%, quietly underwriting hyperscaler capex.
• Citrini swaps nuclear for solar (FSLR, ARRY, NXT, ENPH added) on three-part asymmetry: political pressure from rising electricity bills, solar projects in Republican districts, and solar+storage maturity — plus the favorable IRS Notice 2025-42 safe-harbor clarity.
• Korea long via EWY: export pillars (semis, autos/EVs, batteries, shipbuilding, defense) plus the Lee administration's 'KOSPI 5000' governance campaign should re-rate a market discounted for chaebol capital inefficiency.
• Healthcare STAX positions get maxed out: XGN at 3x revenue (35% discount to VCYT), STIM at <2x revenue (75%+ discount to Brainsway), TOI implied at $20-25+/share by McKesson/TPG private-market comps, AIRS at $14+ on peer 15x EBITDA.

Macro Memo: The Powell Struggle

The US economy will keep slowing without breaking while disinflation persists and tariff inflation stays modest and transitory, so the path is toward lower rates. The key dynamic is 'The Powell Struggle': Trump's failed public pressure on Powell is morphing into a strategic fiscal-monetary convergence (tariff truces, back-loaded spending, disinflationary regulatory nudges) that manufactures conditions justifying cuts, opening the door to easing and gradually lower long-term yields, which would eventually unlock housing. Trade tensions continue taking off-ramps (EU/Japan 'deals' as strategic acquiescence; US-China a stabilized strategic divorce). Markets risk either a seasonal late-summer correction or a 1998-style melt-up into a dotcom-like bubble phase. Positioning leans into the obvious rather than fading extremes: SOFR curve spreads betting the next Fed chair cuts more aggressively, long TLT calls as a hedge, and watchlist interest in agency mREITs.

• Markets price ~1.75 rate cuts by year-end 2025
• 2-year breakeven inflation fell 30bp; 10-year term premium fell ~20bp from May highs
• Official job openings rate ~4.5% (threshold below which unemployment historically rises)
• EU trade framework: 15% blanket tariff except autos; US-China deadline extended 90 days (talks in Stockholm)
• 1 Year Yield 1 Year forward roughly 82bps lower; March 2026 policy rate ~3.25% priced
• Annaly Q2 GAAP NIM 1.04% (+17bps YoY), economic NIM ex-PAA 1.71%; prepay 6.8% vs 4.1% effective mortgage rate
• AGNC shock table: 50bp parallel decline trims tangible book ~3.8%, 75bp move ~7%
• At March 2000 top only 35% of S&P 500 above 200-day MA; margin debt grew >75% early 1999-early 2000 vs S&P <25%
• Mortgage originator triggers: 75-100bp rate rally, MCAI above 110, or mortgage rates below 6%; VIX under 15
• Palladium: sold half above $1,300, stop at $1,250; LatAm vs Argentina trade ~10% outperformance before flipping long Argentina

• The core thesis is 'The Powell Struggle': Trump's failed public pressure on Powell is morphing into strategic fiscal-monetary coordination (tariff truce, back-loaded OBBB spend, disinflationary regulatory nudges) that manufactures conditions justifying rate cuts.
• Citrini expects Powell to open the door to a September cut at the July FOMC or Jackson Hole; markets price ~1.75 cuts by year-end, and long yields should trend lower regardless of who ultimately cuts.
• The labor market shows resilience masking softening: recent-college-grad joblessness at a ten-year high, job openings near the 4.5% threshold that historically precedes rising unemployment, and decelerating wage growth keeping rent inflation contained.
• EU and Japan trade 'deals' (15% blanket EU framework except autos) are strategic acquiescence, not negotiated settlements; US-China is a stabilized strategic divorce with the tariff deadline extended 90 days and talks moving to Stockholm.
• Prior rare-earth picks worked: MP Materials re-rated sharply after the DoD became a large shareholder and Apple signed on as a customer, prompting Citrini to recommend taking profit.
• Markets face a fork: a seasonal late-summer correction or a 1998-style melt-up; breadth (weakest since late 1998) on new highs could signal the start of a dotcom-like 'bubble phase,' though margin debt remains benign versus 2000 and 2007.
• New positioning leans into the obvious: a SOFR U5H6H7 fly (sell H6) betting more cuts price into March 2026-27, plus long TLT December 2025 95 calls as a hedge and fairly-priced upside optionality.
• Watchlist: agency mREITs (NLY, AGNC) broke out on spread dynamics (Annaly economic NIM ex-PAA at 1.71%), best on a mild bull steepener; mortgage originators (RKT, UWMC) need mortgage rates below 6% or MCAI above 110 to work.

GLP-1s: The Second Serving - Revisiting our GLP-1 Basket

GLP-1 obesity drugs have reached a tipping point of adoption, so the second-order effects Citrini predicted two years ago (which the market has now grown complacent about) are about to show up clearly on income statements. As a critical mass of patients has been on the drugs long enough, lagged effects create investable dislocations: aesthetic/cosmetic surgery demand is inflecting up (loose skin, 'Ozempic Face'), food/beverage consumption is shifting toward protein and away from junk/alcohol, and access/sustainability services and fitness benefit, while sleep-apnea, dialysis and diabetic-complication names face lagged headwinds. Among manufacturers, Eli Lilly is favored over Novo Nordisk on a superior, more durable pipeline. Citrini re-establishes its GLP-1 basket (9% of Citrindex) with longs on beneficiaries, shorts on losers, and selective manufacturer/supply-chain exposure, backed by detailed clinical-trial analysis.

• ~4% of US adults had a GLP-1 prescription at end-2024; likely 2-3x higher counting grey-market/compounded; over 30% of US adults obese
• Prior basket: 75% gain over 14 months (exited Nov 2024); new basket = 9% of Citrindex
• LLY ~26x conservative 2026 EPS; tirzepatide at 15% share of a $125B global market; orforglipron -7.9% weight at week 40; retatrutide -24% weight
• Retatrutide visceral fat reduction 29.4% (8mg) / 31.4% (12mg); arrhythmia 12% at 8mg, 7% at 12mg, 6% overall
• AIRS: $750M+ revenue at full utilization, path to 30% EBITDA margins, under 2x EBITDA; fat grafting ~50% annual growth
• WW: net debt ~$300M (<3x EBITDA), telehealth ~20% of revenue (from <5%), <4x 2024 EBITDA vs Hims 6x revenue
• NGVC same-store traffic +5.3% vs Sprouts ~4% vs grocery <1%; supplements >20% of revenue
• XPOF ~7x 2025 EBITDA vs fitness franchisors >20x; GLP-1 drugs cost $900-1,200/month uninsured, up to $200 insured
• Chinese data: BGM0504 19.78% placebo-adjusted at 24 weeks; Hengrui HRS9531 ~18% at 32 weeks; UBT251 16.6% at 12 weeks
• Efficacy progression: early GLP-1s 5-10%, dual agonists 15-20%, next-gen compounds 20%+

• GLP-1 adoption has hit critical mass (nearly 4% of US adults prescribed at end-2024, likely 2-3x higher counting grey-market peptides), so predicted second-order effects will now show up clearly on income statements just as the market has grown complacent.
• Eli Lilly is favored over Novo Nordisk: at ~26x conservative 2026 EPS, tirzepatide alone at 15% share of a $125B market justifies the valuation, with orforglipron, retatrutide and tablet economics as free upside.
• Aesthetic surgery demand is inflecting after an 18-month bust misread as GLP-1 demand destruction; Google Trends turned positive YoY in January 2025 and McKinsey found ~two-thirds of post-GLP-1 aesthetic patients are new to the market.
• Airsculpt (AIRS) is the top aesthetics pick: its network can generate $750M+ revenue at full utilization with a path to 30% EBITDA margins, valuing it under 2x EBITDA at full utilization.
• Citrini adds a diversified 'funding short' basket of fast-food names (EAT, DPZ, QSR, SHAK, YUM, WING) against healthy-food longs like Natural Grocers (NGVC), whose 5.3% same-store traffic beat Sprouts' ~4%.
• WW International (post-bankruptcy) is a favored access/sustainability play: net debt cut to ~$300M (<3x EBITDA), GLP-1 telehealth now ~20% of revenue, trading under 4x 2024 EBITDA versus Hims at 6x revenue.
• New shorts on lagged losers: ResMed (RMD, ~half the sleep-apnea device market), DaVita (DVA), Fresenius (FMS) and MiMedx (MDXG), facing eventual GLP-1-driven demand decline plus reimbursement pressure.
• Weight-loss efficacy keeps climbing (early GLP-1s 5-10%, dual agonists 15-20%, next-gen 20%+); Chinese players are competitive (BrightGene's BGM0504 at 19.78% placebo-adjusted at 24 weeks; Innovent's Mazdutide first China-approved obesity GLP-1).

Robotics Update: Teradyne's Vulcan Win, China Supply-Chain Tour, and Basket Winners

The robotics supply chain is a secular growth story hidden inside cyclically depressed, auto-exposed suppliers, and Citrini's May primer thesis is playing out via three converging forces: a collapsing robot cost curve, VLA models finally giving robots a usable brain, and key suppliers trading at trough valuations. The report's headline reveal (with Hunterbrook) is that Teradyne (TER) supplies the Universal Robots arms behind Amazon's breakthrough Vulcan robot, a potentially catalytic contract on top of a cyclical rebound in TER's semiconductor-test business, implying large upside. A China supply-chain tour confirms rapid Chinese progress and a declining cost curve but a leading edge of autonomy still in the US, so Citrini favors non-China, IP-rich Western/Japanese suppliers with design-win proof. The basket is narrowed toward names converting real orders into booked revenue (sensors/LiDAR, power semis, warehouse bots, actuators) over prototype-stage OEMs.