Skip to content
All theses

The Sequential Monopoly Stack: Chips and Compute

ChipsJune 5, 2026medium conviction15 min read
View Chips & Compute names in screener

The Sequential Monopoly Stack: Chips and Compute

Every AI training run in the world runs on chips NVIDIA architectured, fabricated by TSMC, patterned by ASML machines. That sentence is not a metaphor — it is a supply-chain audit. The $725 billion 2026 hyperscaler capex programme, and the Mag 7's 31% S&P 500 weighting built on top of it, are not underwritten by diversified supply chains. They are underwritten by a sequential stack in which each layer is controlled by a single supplier or near-monopoly cluster, each unhedged against the geopolitical exposure that makes the stack fragile. The argument that follows proceeds from the most visible layer down to the most invisible, and then outward to the one risk none of the named companies has materially changed.

The three named monopolies

NVIDIA commands approximately 80% of the AI accelerator market. Data-centre revenue in fiscal 2026 came in at $193.7 billion, representing 88–90% of total revenue (NVIDIA Q4 FY2026 press release). Blackwell has been sold out through mid-2026 with an approximately 3.6 million unit backlog. Four customers individually exceeded 10% of revenue in the most recent disclosure period — three of them consumer-facing hyperscalers on look-through — which means a single-digit number of purchasing relationships controls the revenue side of a business the market has priced as a monopoly. The CUDA software moat, nearly two decades in the making, is the named switching cost: rewriting kernel code and retraining engineering teams against an alternative architecture that has not yet demonstrated frontier-grade training at scale keeps the installed base in place.

TSMC controls approximately 92% of global advanced-node foundry capacity. Its N3 node is fully booked through 2027; N2 entered high-volume manufacturing in Q4 2025 at roughly 40,000 wafer starts per month, scaling toward 100,000 through 2026 (TrendForce, April 2026). The CoWoS advanced-packaging layer — the integration step that binds high-bandwidth memory to the GPU die — ran at 35,000–40,000 wafers per month in late 2024, targeting roughly 130,000 by end-2026, with NVIDIA holding over 60% of allocated slots through 2027 (Digitimes, December 2025; FinancialContent, February 2026). Apple accounts for approximately 20% of TSMC's FY2025 revenue; NVIDIA accounts for approximately 19% — together over 40% of a foundry whose production is more than 90% concentrated in Taiwan (TSMC 2025 Annual Report; CNBC, January 26, 2026).

ASML holds 100% of EUV lithography and 100% of High-NA EUV. Forty-eight EUV systems were shipped in FY2025 against a year-end backlog of €38.8 billion; High-NA units price at approximately $370–400 million each (ASML Q4 2025 press release). China has received zero EUV systems since the Dutch government declined to renew ASML's export licence in 2019. ASML President and CEO Christophe Fouquet stated in December 2024: "By banning the export of EUV, China will lag 10 to 15 years behind the West" — adding that it took ASML roughly twenty years to develop EUV from prototype to production, which frames his estimate as optimistic rather than conservative (TrendForce, December 2024).

These three names constitute the sequential monopoly in its most visible form.

The second-order chokepoints

Below NVIDIA, TSMC, and ASML sits a tier of sequential single-source dependencies that the financial press has never had to name, because they have never failed. Each is a potential node of contagion as the AI capex cycle pushes throughput against physical limits.

Ajinomoto holds above 95% of the market for ABF laminating film — the dielectric resin used in every advanced CPU, GPU, and ASIC package substrate (TradingKey; Convergence Analysis). The company notified customers of a 30% price hike effective Q3 2026 against a projected 10% supply shortfall in H2 2026, widening to 42% by 2028. The single meaningful challenger, Sekisui Chemical, holds approximately 5%. Every chip in every AI rack passes through this resin chemistry before it reaches a substrate.

JSR, Shin-Etsu Chemical, and Tokyo Ohka Kogyo collectively hold approximately 95% of EUV-grade photoresist supply (TrendForce, November 2025; Fountyl). JSR is now state-owned via the Japan Investment Corporation. No Western peer carries qualified EUV-grade formulations at commercial scale. Photoresist is the most concentrated chemistry chokepoint in the entire compute stack.

Mitsui Chemicals dominates the EUV pellicle market — the protective membrane that shields the reticle from contamination during lithography. Without a pellicle, the reticle contaminates after a few wafer passes and the fab loses the EUV machine to the contamination problem. The pellicle market stood at approximately $558 million in 2024 and is effectively single-source (imec partnership announcement; Grand View Research).

KLA Corporation controls approximately 60% of the semiconductor metrology and inspection equipment market (Yahoo Finance). As fabs transition to N3 and N2, defect rates become economically decisive, and KLA's share of total fab spend rises structurally with node complexity. The metrology and inspection market doubled from $6.0 billion in 2020 to $12.9 billion in 2025 (Fortune Business Insights).

Approximately 50% of ultra-high-purity neon — used as buffer gas in DUV laser lithography — has historically come from Ukrainian steel-mill byproduct capture (USITC briefing). The 2022 conflict-era supply shock prompted diversification, but geographic concentration persists as a stress-tested second-order chokepoint.

TSMC and ASML are the names. ABF resin, EUV photoresist, and the pellicle are the physical chokepoints that make TSMC and ASML possible. The sequential monopoly extends four layers deep before anyone in the financial press has to know what Ajinomoto makes.

The capex/revenue arithmetic

David Cahn's September 2024 analysis at Sequoia Capital identified the revenue arithmetic the infrastructure capex requires: at a 50% gross margin and 50% return-on-invested-capital hurdle, the AI capex implied an annual AI revenue requirement of $600 billion — against observed AI revenue, excluding chip sales that are the spend, of at most $10 billion (Sequoia, September 2024). By his June 2025 update, as capex accelerated, the required revenue had risen to $840 billion annually (Sequoia, June 2025). Observed AI revenue across OpenAI, Anthropic, Google AI products, Microsoft Copilot, and AWS Bedrock combined: under $50 billion. The gap is approximately sixteen times.

The commoditisation pressure compounds the revenue problem structurally. DeepSeek R1 in January 2026 demonstrated frontier-equivalent performance at a declared training cost of under $6 million on Huawei Ascend chips, vaporising approximately $500 billion of NVIDIA market capitalisation in a single session (arXiv:2501.12948; PIIE). Open-source enterprise AI deployment rose from 23% to 67% of new deployments through 2025. Chinese AI models reached approximately 15% of global use share within twelve months of near-zero (TrendForce, November 2025). The pricing power required to close Cahn's gap is being eroded at the platform layer before scale is recouped, while below the stack, the supply-chain pricing power is compounding in the opposite direction.

Goldman Sachs chief economist Jan Hatzius acknowledged that AI added "basically zero" to US GDP growth in 2025, projecting a contribution rising to 0.3–0.7 percentage points by 2027 — roughly a quarter of the 7% cumulative GDP figure the house published in 2023 (Tom's Hardware). Daron Acemoglu's task-based modelling brackets the plausible upper bound at 0.66% TFP gain over ten years — approximately one-tenth of what the capex programme requires to justify itself (NBER WP 32487). The bullish house has, on the macro side, conceded the productivity ceiling.

The unpriced geopolitical exposure

As of May 2026, more than 90% of TSMC's total manufacturing capacity remains concentrated in Taiwan. Even at full CHIPS Act execution by 2029, Taiwan will control approximately 61% of global sub-6nm capacity against 16% for the United States (TSMC 2025 Annual Report; TrendForce projections). NVIDIA's entire AI accelerator product line is manufactured at TSMC at 4nm and 3nm. Apple's entire premium device manufacturing is at TSMC at 3nm and below. AMD, Broadcom, and Qualcomm are similarly concentrated. None has disclosed an alternative foundry relationship for sub-5nm production.

In July 2023, CIA Director William Burns and Director of National Intelligence Avril Haines convened classified briefings for senior technology executives in Silicon Valley. Those briefed included Apple CEO Tim Cook, NVIDIA CEO Jensen Huang, AMD CEO Lisa Su, and Qualcomm CEO Cristiano Amon. The briefings warned that China's military buildup could position Beijing to move on Taiwan as early as 2027. Cook told officials after the briefing that he sleeps "with one eye open." The existence and content of these briefings were not publicly reported until Tripp Mickle's New York Times investigation in February 2026 (NYT, February 24, 2026). Those briefings did not materially change any company's supply-chain architecture on a short-term basis. The bet was known and held.

The US Intelligence Community's March 2026 annual threat assessment stated that China "currently does not plan to invade Taiwan in 2027" but maintains contingency plans (Focus Taiwan, March 2026). The 2027 military ready-date — CIA Director Burns's February 2023 congressional testimony disclosing that Xi Jinping ordered the military ready to invade by 2027 — is a planning milestone, not an assessed probability of action; the distinction is required. Bloomberg Economics' full-conflict modelling quantified the first-year global cost at approximately $10.6 trillion, roughly 9.6% of global GDP, exceeding the COVID-19 pandemic's economic shock (Bloomberg Economics, February 2026). A confidential SIA/McKinsey analysis from 2022, disclosed by the NYT in February 2026, found that losing access to Taiwan's chip supply would trigger "the largest economic crisis since the Great Depression," with US output falling 11%.

This is not a prediction of conflict. It is an identification of a known, unhedged structural exposure. The market prices Taiwan Strait stability as a given. The CIA, CSIS, and Bloomberg Economics price it differently. The question is not whether Taiwan is invaded. The question is whether the market is compensating equity holders for carrying the probability. The evidence is that it is not.

The CHIPS Act: real in direction, immaterial on the relevant timeline

The CHIPS and Science Act appropriated $52.7 billion, of which $33.7 billion across twenty awards had been committed as of January 2025 (Commerce OIG Report OIG-25-021-I). The policy is genuine and bipartisan; the Trump administration did not roll back grants already finalised, and TSMC CEO C.C. Wei jointly announced an expanded $165 billion Arizona commitment with the President in March 2025.

The execution timeline does not match the 2025–2028 AI capex exposure window.

TSMC Arizona Phase 1 (4nm) is operational and reportedly at parity yield with comparable Taiwan nodes — a genuine manufacturing achievement (Tom's Hardware). Its capacity of approximately 10,000 wafer starts per month represents less than 1% of TSMC's Taiwan capacity. Phase 2 (3nm) targets volume production in 2H 2027; Phase 3 (2nm) targets 2029. The full six-fab Arizona cluster at completion would account for approximately 30% of TSMC's 2nm capacity — but that is a 2030s timeline, not a 2026–2027 timeline (TrendForce, December 2025).

Intel Ohio's first fab has been delayed to 2030 — eight years from announcement (Tom's Hardware; Intel Newsroom). Samsung's Texas facility is attempting a 2nm GAA transition from a paused 4nm plan; current yield is reported at 10–20%, well below the approximately 80% threshold required for economically viable high-volume manufacturing (Tom's Hardware; TechSpot). Micron's New York Fab 1 targets 2029. US advanced-node capacity as a share of global leading-edge stands at approximately 2–3% today. Even after Phase 2 Arizona, early Micron, and Intel 18A ramp, the projected US share of sub-6nm by 2029 is approximately 16% — against Taiwan's projected 61%. The policy is real. The timescales are not commensurate.

China's constrained catch-up

TechInsights' teardown of the Huawei Mate 60 Pro confirmed SMIC manufactured the Kirin 9000S using its N+2 process at 7nm-class dimensions — entirely without EUV lithography, using multi-patterning DUV (TechInsights, September 2023). The Mate 70 Pro (2024) made no advance to 5nm. The ITIF assessed in August 2024 that China remains a decade behind semiconductor leaders (ITIF, August 2024). The DUV-only path is technically feasible at 7nm (proven), marginally feasible at 5nm (contested), and not scalable to 3nm or below — EUV is a physical necessity at sub-5nm, not a commercial preference.

Huawei's Ascend 910C delivers roughly 60% of NVIDIA H100 inference performance on DeepSeek's own published evaluation but remains "unattractive for training AI models" (Tom's Hardware; CSIS). Huawei's CloudMatrix 384 cluster surpasses NVIDIA's NVL72 on certain throughput metrics by using five times more chips — a systems engineering achievement, not chip-level parity.

April 2026 brought the DeepSeek V4 preview: a 1.6 trillion parameter model optimised for Huawei's Ascend processors, the first major frontier release not designed for NVIDIA hardware, with ByteDance, Tencent, and Alibaba scrambling to order Huawei Ascend 950 processors in response (Fortune, April 24, 2026; TrendForce, April 2026). The divergence between China's algorithmic capability curve — advancing rapidly — and its manufacturing capability curve — constrained by EUV denial — is the forward-looking variable the thesis turns on. CXMT's DRAM production remains at 16nm, approximately three years behind Samsung, SK Hynix, and Micron at 10nm and below (TechInsights). Fouquet's 10–15 year lag estimate remains the correct structural anchor: it took ASML approximately 20 years to develop EUV from prototype to production. China starts from DUV-only without the foundational IP, supply chain, or field engineering. The estimate is optimistic on his own framing.

The synthesis: one geopolitical bet wearing three costumes

The Mag 7 at 31% of the S&P 500 is, on supply-chain look-through, a concentrated, unhedged geopolitical bet on Taiwan Strait stability wearing three costumes: an AI capex story, a consumer-revenue story, and a productivity story.

Strip the look-through: Apple's entire premium device manufacturing runs at TSMC at 3nm and below; no disclosed alternative foundry. NVIDIA's entire AI accelerator product line runs at TSMC at 4nm and 3nm; no announced alternative. The combined Apple and NVIDIA market capitalisation exceeds $6 trillion as of May 2026. Both companies' most critical products depend on a single production cluster more than 90% located 100 miles from mainland China — a cluster the CIA briefed their CEOs about as carrying material conflict risk in the 2027 planning window.

Chris Miller framed it at the CommonWealth Economic Forum in January 2025: "today, when it comes to advanced chips, around 90 percent of them are made in one island, Taiwan, which is an extraordinary concentration, both from the perspective of earthquake risk, but, more importantly, from geopolitical risk" (Commonwealth Magazine, January 13, 2025). His CFR interview (January 2023) had already identified the GPU-specific dependency: "almost all of the graphics processing unit chips that run artificial intelligence applications in data centres are made in Taiwan" (CFR, January 3, 2023).

Leopold Aschenbrenner's compute-scaling bull case — the premise underwriting both Cahn's revenue gap and the Mag 7's multiple — requires uninterrupted access to TSMC's production. Taiwan Strait stability is the hidden denominator of every AI bull-case calculation. The energy constraint compounds the same stack from the other direction: NERC's January 2026 Long-Term Reliability Assessment revised 10-year US summer peak demand upward by +224 GW (+69%), the largest single-year revision in the assessment's history, and PJM cleared at the FERC price cap ($329.17/MW-day) for two consecutive years (NERC LTRA 2025; PJM RPM). The silicon constraint and the power constraint are independent bindings on the same thesis, pointing at the same 2026–2027 window.

The desk monitors Honeywell (HON), IBM, and IonQ (IONQ) as the quantum-computing layer where, over a much longer horizon, an alternative compute architecture could theoretically evolve — HON as pre-IPO proxy for Quantinuum (the trapped-ion platform with disclosed logical-qubit milestones), IBM with its Starling 2029 roadmap and Qiskit middleware moat, and IONQ as the trapped-ion pure-play that has stapled a Trusted Foundry fab and photonic-interconnect capability onto its listed vehicle. That surveillance is a 2030s question at the earliest and does not change the chips-and-compute bottleneck that is binding in 2025–2028.

What would change my mind

Three named thresholds, each with a mechanism and a named primary source.

Supply-side falsification: US advanced-node capacity as a share of global sub-6nm exceeds 25% before 2030, verified by TrendForce or SemiAnalysis capacity reports. The current trajectory — approximately 16% by 2029 — needs to beat the projection by ten percentage points for the CHIPS Act to have materially closed the exposure window ahead of schedule.

Demand-side falsification: AI-attributable revenue grows to within 3× of hyperscaler annual AI capex within two years, or Mag 7 AI-service gross margin holds above 60% for three consecutive years. Either outcome would demonstrate that the commoditisation curve — DeepSeek R1 at under $6 million, open-source enterprise deployment at 67%, Chinese use-share at 15% — has not eroded platform-layer pricing power at the rate the structural evidence implies.

Geopolitical falsification: A verifiable, commercially significant alternative TSMC-equivalent foundry at sub-3nm outside Taiwan reaches 10% of advanced-node global capacity before 2030. TSMC Arizona Phase 2, Samsung Taylor, and Intel 18A are the three candidates. Samsung's current 10–20% yield at 2nm and Intel Ohio's 2030 timeline mean that threshold is not close.

China ceiling revision: A confirmed SMIC or domestic-Chinese EUV machine achieving production-grade yields at 5nm or below before 2028 would require updating Fouquet's 10–15 year estimate. The constraint is real; the timeline is the question. No such confirmation exists as of May 2026.

Until one of these fires, the read stands. The Mag 7 is not seven stocks. It is a single unhedged bet on Taiwan Strait stability — priced as if the CIA, CSIS, and Bloomberg Economics had not all quantified what happens when that bet resolves adversely — layered on a capex arithmetic that requires $840 billion of annual AI revenue against a current run-rate under $50 billion, held together by a sequential monopoly stack whose most important members are an amino-acid company's dielectric resin and a Dutch machine that no one else on the planet knows how to build.

The math is the math. The timing is the question.

Surfaced Names

6920.T

research_scan theme='chips-compute'; value_chain_position='upstream-enabler'; Sole vendor of EUV actinic mask-blank inspection equipment — a second-order monopoly nested inside the ASML EUV stack. The Sept-8 ASML tri-customer High-NA collaboration announcements widen the addressable market for mask-inspection at every High-NA customer (TSMC + Intel Foundry + Samsung). Non-US listing noted; Tokyo Stock Exchange ticker.

ALAB

research_scan theme='agi-infrastructure'; value_chain_position='upstream-enabler'; PCIe-CXL retimer and memory-expansion fabric picks-and-shovel; sits directly on the accelerator-to-CPU interconnect layer and benefits from custom-silicon multi-source dynamics regardless of merchant-GPU capture. Fabric-retimer positioning is adjacent to NVLink Fusion adoption cadence (d-Matrix Raptor XPU 2026-09-10 per chips-compute 2026-09-14).

AMZN

research_scan theme='autonomous-mobility'; value_chain_position='pure-play'; Zoox sits inside Amazon; parent balance sheet cushions the pure-play robotaxi capex from C-23 rate-cycle exposure similar to Alphabet-Waymo. No verified fresh Zoox operational disclosure this cycle; retained for structural completeness.

ANET

research_scan theme='agi-infrastructure'; value_chain_position='picks-and-shovels'; Ethernet / UEC back-end AI-network spine — the multi-vendor alternative to NVDA NVLink captive fabric. Beneficiary of the same hyperscaler multi-sourcing dynamic that promotes AVGO/MRVL custom silicon; sells into all hyperscalers regardless of which accelerator wins each slot.

ASML

research_scan theme='chips-compute'; value_chain_position='upstream-enabler'; Single-supplier EUV / High-NA EUV monopoly gating every leading-edge logic node at TSMC + Intel Foundry + Samsung. The 2026-09-08 press-release triple (Intel Foundry × ASML High-NA collaboration, Samsung × ASML expanded strategic collaboration, Eindhoven campus construction) widens the customer base for High-NA EUV to three logic customers simultaneously — a monopoly-rent moat that strengthens further this cycle. Q2 2026 print of €9.3B / 54.0% GM (July 23 scan) showed no compression.

ASTS

research_scan theme='space-economy'; value_chain_position='pure-play'; Bluebird Block 2 launch cadence remains slipped into 2027 on Blue Origin New Glenn constraint (1–3 flights/yr). HALO Europa dual-use direct-to-cell ASIC funding (Feb 2026) established. Watch: FCC SCS docket resolution and MNO partnership deployment count. Structurally novel direct-to-cell competitor to Starlink Direct-to-Cell.

AVGO

research_scan theme='agi-infrastructure'; value_chain_position='picks-and-shovels'; Custom AI ASIC design partner for Google TPU, Meta MTIA, and multiple additional hyperscaler captive-silicon programs — 18–36 month qualification cycle protects rent against merchant-GPU price competition. As Sonnet 5 promotional pricing signals platform-layer margin compression, custom-ASIC rent captured by AVGO is one layer out and structurally more durable than NVDA merchant-GPU margin under C-04/C-26 rent-migration frame.

BESI.AS

research_scan theme='chips-compute'; value_chain_position='upstream-enabler'; Dominant hybrid-bonding equipment vendor into HBM stack assembly and CoWoS-adjacent advanced-packaging. Sits directly upstream of the SK Hynix / Micron / Samsung HBM4 ramp and TSMC's binding CoWoS-S/L packaging capacity — the durable-rent picks-and-shovels layer that captures spend regardless of which foundry or HBM supplier wins the next architectural cycle.

CACI

research_scan theme='defense-modernization'; value_chain_position='picks-and-shovels'; SkyValor $500M counter-drone 'detect and defeat' contract (July 31, 2026) — Pentagon Counter-Drone Task Force + Domestic Shield initiative — brings systems-integrator prime with cleared-labour pool into direct competition with autonomy pure-plays (AVAV, Anduril) at the C-UAS platform-integration layer. Also positioned in the DoD enterprise-software layer against Oracle + Accenture + PLTR.

COHR

research_scan theme='quantum-computing'; value_chain_position='upstream-enabler'; Laser and photonic components across ion-trap, neutral-atom, and photonic quantum stacks; sits alongside Toptica (private) in the modality-agnostic upstream. Also carries optical-interconnect exposure to AI-datacentre buildout — a two-theme picks-and-shovels position.

CRWV

research_scan theme='chips-compute'; value_chain_position='downstream-beneficiary'; AVOID per C-08 posture. Neocloud in the Vera Rubin four-cloud production ramp (2026-07-21, alongside GCP/Azure/OCI); preferential merchant-GPU allocation tightens coupling to NVDA product-cycle success ahead of the Q3-Q4 2026 material refinancing window against the Blackstone $11.5B facility (March 2025). C-08 transmission channel to private-credit / insurance-balance-sheet exposure into C-21 default chain.

DDOG

research_scan theme='agi-infrastructure'; value_chain_position='picks-and-shovels'; LLM observability extended off the installed observability footprint — moat is the pre-existing agent footprint, not model-specific IP. Structurally advantaged versus standalone LLM-eval (LangSmith, Arize, Braintrust, W&B) which now face the Anthropic-AWS-MSFT-GOOG-Glasswing cross-lab consortium eval-primitive control per 2026-06-30 Fable 5 industry framework.

HON

research_scan theme='quantum-computing'; value_chain_position='downstream-beneficiary'; Retained economic interest in Quantinuum post-IPO (residual stake percentage not disclosed at press-release grain — S-1/424B5 verification queued for next scan). Public-equity proxy for pure-play quantum hardware capital-gain optionality; Honeywell industrial cash-flow base underwrites the exposure without single-name volatility on Quantinuum.

HXGBY

research_scan theme='robotics-humanoids'; value_chain_position='pure-play'; Sensor + reality-capture + factory-automation platform with humanoid + mobile-robotics adjacency; the diversified public-market analogue for OEM-tier exposure at investable liquidity. Retained across scans as non-Figure calibration entry. Not a headline humanoid pure-play — treat as OEM-adjacent exposure with automation-adjacent cushion.

IBM

research_scan theme='quantum-computing'; value_chain_position='picks-and-shovels'; Only mature-cap incumbent running a publicly-committed logical-qubit roadmap (Loon 2025 -> Kookaburra 2026 -> Cockatoo 2027 -> Starling 2029 -> Blue Jay 2033) using bivariate-bicycle qLDPC codes; hyperscaler quantum cloud channel (IBM Quantum) alongside AWS/Azure/GCP. Quantum sits inside IBM's aggregate revenue rather than dominating it — a lower-risk exposure to the theme than the pure-plays. IBM Quantum blog primary source returned 403 this scan; roadmap milestones inferring on prior published cadence.

IONQ

research_scan theme='quantum-computing'; value_chain_position='pure-play'; Trapped-ion pure-play with $260-270M FY26 revenue guide (raised at Q1 2026 print +755% YoY), $3.1B cash, $470M RPO; pending SkyWater Trusted-Foundry acquisition close Q2-Q3 2026 vertically integrates control-electronics supply. IonQ Clavis XG Multiplex (2026-06-17) opens a PQC-migration/QKD revenue bridge on regulated federal spend. Named-equity C-02 valuation-regime test (relevant-but-not-injected) at ~37x forward revenue multiple.

META

research_scan theme='ai-capex-ceiling'; value_chain_position='pure-play'; Q2 CY2026 capex $31.08B; FY2026 capex guide narrowed to $130-145B — low end RAISED $5B from prior $125B floor. Q2 total revenue $60.80B (+28% YoY), Family of Apps ad revenue $59.36B. The low-end raise is the load-bearing new datum this cycle — moves C-04 arithmetic contradicts-ward. AVOID per Druckenmiller 13F exit list and C-09 ad-revenue-lag risk.

MRCY

research_scan theme='defense-modernization'; value_chain_position='picks-and-shovels'; Rugged compute / radar / EW module supplier at the intersection of chips-compute and defense-modernization. Beneficiary of the classified programs $12.1B line and the airborne-targeting/space-networks $4B line in the July 2026 supplemental. Sells into every prime radar and ISR system; the durability read is on Golden Dome MEO tracking-sat AMDT3 build-out.

MRVL

research_scan theme='agi-infrastructure'; value_chain_position='picks-and-shovels'; Custom silicon partner for AWS Trainium/Inferentia — the second-order beneficiary of hyperscaler multi-sourcing away from NVDA. AWS Bedrock margin captured by Amazon rather than by Anthropic under the Sonnet 5 promotional-pricing configuration; MRVL sells the silicon that makes that capture possible.

MSFT

research_scan theme='ai-capex-ceiling'; value_chain_position='pure-play'; Q4 FY2026 earnings scheduled 2026-07-29 (verified) — first hyperscaler capex-guide read on whether the ~16x capex/revenue gap begins to close in CY2026; direct falsification-observable candidate for criterion #26. MSFT sits on supports_avoid per C-04 posture ahead of the print.

MU

research_scan theme='chips-compute'; value_chain_position='upstream-enabler'; US-listed HBM oligopoly member (with SK Hynix + Samsung); HBM4 ramp qualifies for Blackwell Ultra / Vera Rubin. SK Hynix Indiana fab groundbreaking (2026-08-28) accelerates US-manufacturing footprint at the HBM tier, structurally supportive of Micron's Idaho + New York expansion posture and of the passive-flow marginal-buyer thesis at the HBM oligopoly.

NOW

research_scan theme='agi-infrastructure'; value_chain_position='pure-play'; Now Assist rides the CMDB installed-base moat — enterprise workflow orchestration where the customer's process graph is not substitutable within one refresh cycle. More insulated than lightweight orchestrators against Anthropic Cowork forward integration, but ~60× fwd P/E prices durability that requires enterprise IT budget conditions Essay 04 §II questions.

NVDA

research_scan theme='autonomous-mobility'; value_chain_position='Layer 2 (AV inference SoC — DRIVE Thor)'; DRIVE Thor Blackwell-derived is OEM-agnostic AV silicon platform. Same die family and distribution strategy as Jetson Thor for humanoids — 'Android of physical AI' extends across both themes. Durably profitable at compute-provider tier. BPA has NVDA on supports_avoid side of C-04 at cap-weighted peak (posture 41), but on AV-specific chessboard NVDA is least-exposed compute-side incumbent. Held as calibration exposure.

OXIG.L

research_scan theme='quantum-computing'; value_chain_position='upstream-enabler'; Public-market cryogenics oligopolist (dilution refrigerators, vacuum systems, cryogenic optics) selling into every quantum modality regardless of architecture — the durable picks-and-shovels tier. Sales cycle 12-24 months, capacity constrained by specialist manufacturing; benefits from D-Wave gate-model build-out on top of existing trapped-ion, superconducting, and photonic customer footprint.

PLTR

research_scan theme='defense-modernization'; value_chain_position='downstream-beneficiary'; Marine Corps Maven Smart System mandate (June 23, 2026) is the operational-deployment inflection at the tactical-app layer. Updated framing this scan: the July 9 Accenture $821M War Data Platform award confirms PLTR is not the sole DoD enterprise-data-platform winner — durable at tactical-app, capped at enterprise-integration. Wallet-share ceiling is lower than the June 25 reading implied.

QNT

research_scan theme='quantum-computing'; value_chain_position='pure-play'; Trapped-ion hardware-leading pure-play; IPO priced 2026-06-03 and closed 2026-06-05 at $60/share, 28M shares sold on Nasdaq for $1.68B gross proceeds; 2026-06-21 HPE strategic collaboration extends the HPC-hybrid enterprise channel beyond BMW/BP/Synopsys/Mitsubishi Electric. Sits inside the C-02 valuation-regime test (relevant-but-not-injected) at named-equity level. Post-IPO revenue disclosure needed to establish comparable-set multiple vs IONQ FY26 $260-270M guide.

SYM

research_scan theme='robotics-humanoids'; value_chain_position='downstream-beneficiary'; Rigid-goods logistics automation with Walmart flagship + expanding footprint. Public-market analogue for the apparel-DC / material-moving deployment surface humanoids target via Catalyst Brands Reno; the rigid-goods variant is already operating at scale and provides the payback-period reference against which humanoid-DC economics will be measured.

TEL

research_scan theme='robotics-humanoids'; value_chain_position='picks-and-shovels'; Sensor + connector supplier across humanoid BOMs (force-torque, tactile, position). Trades ~16x forward with sensor-content-per-unit rising as OEMs move to 25-DoF hands (1X 2026-07-09) and multi-vertical deployments (Catalyst apparel DC extends beyond auto). Under-priced picks-and-shovels edge — captures rent regardless of which OEM/model wins.

TSM

research_scan theme='chips-compute'; value_chain_position='picks-and-shovels'; Sole leading-edge foundry for Blackwell/Blackwell Ultra/Vera Rubin plus binding CoWoS-S/L packaging capacity. July 2026 revenue +44.7% YoY, August 2026 +53.3% YoY / +10.1% MoM, YTD +39.3% — the foundry-tier print continues to accelerate rather than compress, contradicting the C-04 capex-side compression path on criterion #26 while ratifying C-26 rent migration to the digital upstream-monopoly tier.

UBER

research_scan theme='autonomous-mobility'; value_chain_position='downstream-beneficiary'; Rideshare network-rake beneficiary regardless of which AV OEM wins domestically; but Waymo's Aug 13 Vegas Raiders / Allegiant Stadium official-partner deal plus the Allianz-underwritten London+Munich international launch are the first evidence Waymo internalises payments+matching+insurance at the metro level — narrows the domestic US rake case, expands the international rake case. Under-priced 2026 exposure at ~15-18x fwd.