brianletort.ai
All issues

The AI Stack Weekly

Issue 04 · Week 20 of 2026.

/Industry brief · ~7 min read/Public sources onlyDownload brief

The Bottom Line

Frontier text took a breath. The constraint stack moved to supply, regulation, and pricing structure.

Flywheel arcAll three lenses

The model layer took a breath this week. No new GPT-class or Claude refresh between W19 and Google I/O (May 19-20), no headline frontier-text GA — but the specialist canopy widened sharply: Perceptron Mk1 priced video and embodied reasoning 80-90% below Gemini Flash Lite on May 12; NVIDIA's open-weights SANA-WM put a minute-scale 720p world model on a single RTX 5090 on May 15; OpenBMB's MiniCPM-V 4.6 collapsed multimodal SLMs to 1.3B running on-device across iOS, Android, and HarmonyOS on May 11.

Pricing structure became the dominant decision input. Anthropic emailed Max-20x subscribers a June 15 policy that moves Claude Agent SDK, `claude -p`, GitHub Actions, and third-party agents off subscription rate limits onto separate $20-$200/mo metered credit at API list prices — 12x-175x effective price increase per workload, ending Claude Code subscription arbitrage. OpenAI and Microsoft formalized a $38B cumulative cap on Microsoft's revenue-share through 2030 — the first hard ceiling on any hyperscaler's AI revenue-capture. xAI shipped Grok Build CLI and a SuperGrok Heavy tier at $300/mo. The unit of competition shifted from model intelligence to agent-runtime monetization.

Demand pulled forward into prints. Cisco Q3 FY26 raised FY26 AI infrastructure orders guide from $5B to $9B (Q3 AI orders $1.9B vs $600M YoY; ~$300M from non-hyperscaler neocloud / sovereign / enterprise buyers). Applied Materials Q2 FY26 printed a record $7.91B with record 50% non-GAAP gross margin and guided calendar-2026 semicap growth above 30%. Anthropic in talks at over $900B post-money on $30B April ARR (Bloomberg May 12), target close end-May. OpenAI launched a $4B+ Deployment Company with TPG, Advent, Bain, and Brookfield.

The supply chain wedged itself into the read. Samsung-union talks collapsed on May 13, locking in an 18-day general walkout May 21-June 7 across 50,000+ workers — Samsung is the sole HBM4 mass-shipper for NVIDIA's Vera Rubin platform, holds ~40% global DRAM share, and a single April 23 half-day rally already cut daily memory output 18.4%. TSMC's Taiwan Tech Symposium guided AI wafer demand 11x 2022-2026 and CoWoS capacity CAGR above 80% through 2027 — capacity is coming, but not before Q3.

The so-what for the four audiences: boards should treat federal procurement (Canada-TELUS BC sovereign AI factory May 11; HMRC-Quantexa £175M sovereign data and AI May 14) and capability gating (UK AISI's 4.7-month doubling rate on autonomous cyber) as durable vendor-risk dimensions. Investors should treat the Anthropic round close (not the headline valuation) as the binary read of the week and watch for Samsung-walkout impact on NVIDIA Q1 FY27 (May 20) and H2 hyperscaler capex. Architects and operators running Claude Code at scale need to model true API burn before the June 15 cutover this week, lock HBM allocation and Q3 GPU delivery slots, and add at least one specialist physical-AI model (Perceptron Mk1, MolmoAct 2) to 2H roadmaps before procurement budgets close.

JevonsMetcalfeGilderSoftwareJevonsHardwareHuangNetworkingMetcalfe + Gilder

The three lenses

What moved this week, and what to do about it.

12 events across the flywheel — 4 software, 4 hardware, 4 networking.

Software.

  • Anthropic emails Max-20x subscribers a June 15 policy: Claude Agent SDK, `claude -p`, GitHub Actions, and third-party agents (OpenClaw, T3 Code, Conductor, Zed, Jean) move off subscription rate limits onto separate $20-$200/mo metered credit at API list prices — community math shows 12x-175x effective price increase per workload, ending Claude Code subscription arbitrage

    support.claude.com, VentureBeat, XDA-Developers, Gate News

  • UK AISI publishes paired evaluations: GPT-5.5 hits 71.4% expert-task pass on AISI's cyber suite and becomes the second model after Claude Mythos Preview to solve the 32-step 'Last Ones' corporate-network attack end-to-end; companion post finds autonomous cyber task length is doubling every 4.7 months, aligned with METR's 4.2-month SWE figure

    aisi.gov.uk, The Register, CyberScoop, IT Pro, Metacurity, The Decoder

  • Google's Android Show: I/O Edition unveils 'Gemini Intelligence' (built on Gemini 3.1) and Googlebook AI laptops, pushing multi-step agentic Gemini — Chrome auto-browse, cross-app task automation, Rambler dictation, vibe-coded widgets — into the Android OS layer ahead of I/O 2026 (May 19-20)

    blog.google, TechCrunch, Engadget, 9to5Google, CNBC

  • OpenAI Codex changelog ships Codex remote access (Mac-to-mobile project handoff) plus Codex access tokens for trusted non-interactive ChatGPT Enterprise workspaces — Hooks now GA — formalizing a workspace-permissioned CI/CD automation rail distinct from raw API keys

    developers.openai.com/codex/changelog, OpenAI Help Center

What this means

W20 traded model novelty for structural posture. AISI's paired disclosures confirm autonomous offensive-cyber capability is now doubling every ~4.7 months (matched to METR's 4.2-month SWE figure) — CISOs and boards should treat that doubling rate as the new 2026 baseline for cyber-risk planning. Anthropic's $200 Agent SDK credit ends subscription-arbitraged Claude Code economics — architects running Claude Code at scale must model true API-rate burn before the June 15 cutover this week or shift workloads. Gemini Intelligence pushing the frontier-model surface into the Android OS layer and OpenAI Codex workspace-permissioned automation rails mean distribution and pricing structure are now the dominant procurement variables, not benchmark scores. See the Model Pulse for the three new tree rows (Perceptron Mk1, SANA-WM, MiniCPM-V 4.6) that widened the specialist canopy this week.

Hardware.

  • Government-mediated Samsung-union talks collapse after a 17-hour final round, locking in an 18-day general walkout May 21-June 7 covering 50,000+ workers at the only company currently mass-shipping HBM4 for NVIDIA's Vera Rubin (Samsung holds ~40% global DRAM share; a single April 23 half-day rally already cut daily memory output 18.4%)

    Korea Herald, Seoul Economic Daily, TechTimes

  • TSMC at Taiwan Tech Symposium guides AI wafer demand up 11x 2022-2026, CoWoS capacity CAGR >80% through 2027, 14-reticle CoWoS with 20 HBM stacks in 2028 / >14 reticles with 24 HBM stacks in 2029, SoIC Chiayi line scaling 5x to 50K wpm by 2027, and the world's first 200Gbps Micro Ring Modulator using COUPE silicon photonics entering production this year

    TrendForce, Reuters, Commercial Times, Digitimes

  • TrendForce reports MediaTek adopting a dual advanced-packaging strategy (Intel EMIB + TSMC CoWoS) for AI ASICs, with Google's TPU v8e expected to use Intel EMIB; EMIB-M is scaling toward 6-12x effective reticle size vs CoWoS-S at 3.3x, opening a credible second source for hyperscaler custom-silicon supply outside the CoWoS bottleneck

    TrendForce, Digitimes

  • Applied Materials and TSMC announce a partnership at AMAT's $5B EPIC Center in Silicon Valley to co-develop materials, equipment, and process integration for next-gen 3D transistor and interconnect structures, with TSMC named a founding partner getting early access to AMAT's roadmap

    Applied Materials IR / GlobeNewswire press release

What this means

Architects and procurement leads should lock HBM allocation and triage Q3 2026 GPU delivery slots this week rather than waiting for the May 20 NVIDIA print: the Samsung walkout (May 21-June 7) threatens the sole HBM4 mass-shipper for Vera Rubin, while TSMC's 14-reticle CoWoS and 24-stack HBM trajectory (2028-29) sets the density envelope to plan 2028+ platforms against. MediaTek's validation of Intel EMIB as a real CoWoS alternative for AI ASICs (with TPU v8e named) should reopen the ~33-36% custom-silicon-share assumption upward — when packaging supply diversifies, hyperscaler ASIC volume scales faster, and the hardware-to-software flywheel turns again because new packaging unlocks architectures (24-stack HBM, EMIB-bridged ASICs) that weren't economic last quarter. Investors should model HBM supply continuity — not GPU pricing — as the binding swing variable on the May 20 NVIDIA print and H2 hyperscaler capex.

Networking.

  • 3M, AMD, Arista, Cisco, Meta, Oracle (co-chair), Molex, Amphenol, TE Connectivity, Senko, Sumitomo, Source Photonics, Accelink, Aperion, Nexthop-AI, viaPhoton, and Xscape Photonics launch the Expanded Beam Optical (EBO) MSA at ebomsa.org to standardize open, contamination-resilient connector specs for hyperscale AI fabrics

    3M, PR Newswire, Converge Digest, SDxCentral, RCR Tech

  • TELUS and the Government of Canada unveil a three-site British Columbia Sovereign AI Factory cluster (Kamloops, Vancouver Mount Pleasant, 150 W Georgia) scaling to 60,000+ NVIDIA GPUs and 150 MW by 2032, networked with NVIDIA Quantum InfiniBand and Spectrum-X Ethernet under the federal Enabling Large-Scale Sovereign AI Data Centres program

    TELUS, CNW Newswire, CBC, Globe and Mail

  • Lumen launches NorthLine, a ~2,000-mile Seattle-Minneapolis long-haul fiber route (100G/400G wavelengths now, architected for future 800G and 1.6T) — its first major long-haul build in decades — targeting AI east-west DCI traffic across emerging northern data-center corridors

    Lumen, BusinessWire, Fierce Network, Telecompaper

  • Equinix announces global expansion of Fabric Geo Zones across five continents (preview live in AU/BR/CA/JP/CH/UK/US; EU in June), positioning network-layer multicloud sovereignty enforcement that blocks rerouting across non-compliant jurisdictions on its 77-metro software-defined fabric

    Equinix Newsroom, PR Newswire

What this means

W20's pattern is network-internal compounding — AI fabric matured at the connector layer (17-vendor EBO MSA pulling AMD, Cisco, Meta, and Oracle into a shared open spec), the inter-DC fiber map (Lumen NorthLine, first major US long-haul build in decades), and sovereignty enforced at the wire (Equinix Geo Zones; TELUS' federal-backed InfiniBand + Spectrum-X cluster). Architects and procurement leads writing 2027 AI RFPs should treat 800G as capacity-bound through mid-2027 (per Applied Optoelectronics Q1 commentary), start EBO connector qualification and sovereignty-aware routing requirements now, and stop treating both as application-overlay problems. Investors should watch interconnect-category pricing power compound — Equinix Geo Zones is a wire-level enforcement product, not a feature.

Capital flow

Money in, revenue out.

4 categories tracked. Capital deployment up in 2 of 4; revenue follows at multiples of 0.21 to 0.6.

The four-category scorecard. Where capital is going in, where revenue is coming out, and how much of it is real. The one chart for the boardroom.

  • Frontier Labs

    OpenAI, Anthropic, Google DeepMind, xAI

    Capital In

    ~$52B →

    vs ~$52B

    Revenue Out

    ~$48B ↑

    vs ~$45B

    Burn / Rev

    0.9

    Movement

    Anthropic in talks to raise at least $30B at a >$900B valuation (Bloomberg May 12), target close end-May, on April annualized run-rate revenue of ~$30B — boards on Anthropic-dependent contracts should underwrite ~30 months counterparty runway conditional on close; investors should treat the close (not the headline valuation) as the binary read of the week.

  • Hyperscaler-Hosted

    Microsoft Azure, Google Cloud, AWS, Meta, Oracle

    Capital In

    ~$58B →

    vs ~$58B

    Revenue Out

    ~$15B ↑

    vs ~$13B

    Burn / Rev

    0.26

    Movement

    OpenAI and Microsoft formalized a $38B cumulative cap on Microsoft's revenue-share payments through 2030 on May 12 — the first hard ceiling on any hyperscaler's AI revenue-capture; architects building on Azure-OpenAI should plan for Microsoft pushing native MAI-line and direct API products into more enterprise workloads through 2027, and operators should expect Azure-OpenAI commercial terms to harden, not loosen, post-cap.

  • Neoclouds

    CoreWeave, Nscale, IREN, Crusoe, Lambda, Applied Digital, Nebius

    Capital In

    ~$30B ↑

    vs ~$28B

    Revenue Out

    ~$3.0B →

    vs ~$3.0B

    Burn / Rev

    0.10

    Movement

    Nscale closed $790M committed senior debt (+$790M accordion) from five Nordic banks for the Narvik, Norway expansion on May 11 — operators evaluating non-CoreWeave neocloud capacity now have a fresh credit comp; investors should treat this as the template for second-tier neocloud refinancings, not a one-off, and watch the accordion draw cadence as a real-time demand signal.

  • On-Prem / Hybrid

    Enterprise GPU clusters, sovereign and national programs, Cisco / Dell / HPE

    Capital In

    ~$45B ↑

    vs ~$42B

    Revenue Out

    Indirect →

    vs Indirect

    Burn / Rev

    n/a

    Movement

    Cisco Q3 FY26 raised FY26 AI infrastructure orders guide from $5B to $9B and AI revenue guide from $3B to $4B on May 13, with Q3 AI orders of $1.9B vs $600M YoY (~$300M from non-hyperscaler neocloud / sovereign / enterprise) — architects should re-baseline enterprise AI fabric demand from a 2027-2028 thesis to a real 2026-print risk; operators should expect lead-time pressure on Silicon One, Nexus 9000, and Hypershield SKUs.

Burn-to-Revenue is revenue divided by committed capital. Lower means more capital is going out than coming in.

See the full breakdown

Signal vs noise

What’s real, what’s noise.

5 claims this week — 4 signal, 1 noise.

Each claim is scored 1–5 on source quality and triangulation. Anything 2 or below is flagged as noise. Where consensus is wrong, we say so.

  • 5 / 5

    Cisco raised FY26 AI infrastructure orders guide from $5B to $9B and AI revenue guide from $3B to $4B (Q3 FY26 release May 13, 2026); Q3 AI orders alone were $1.9B vs $600M YoY, with ~$300M from non-hyperscaler neocloud / sovereign / enterprise buyers against a $3B pipeline.

    Sources: Cisco PR Newswire (primary press release), CNBC, FT.com data feed

    Real and material — enterprise / on-prem AI fabric demand pulled forward into 2026 prints; operators on legacy refresh budgets should re-baseline networking lead-times this week, not next quarter. What would change the read: Q4 FY26 AI orders failing to deliver the implied ~$3.7B step needed to hit the $9B annual guide.

  • 5 / 5

    Anthropic in talks to raise at least $30B at a valuation exceeding $900B, target close end-May 2026, on April annualized run-rate revenue of ~$30B (doubled from $14B in February); previous Feb 2026 round was $30B at $350B.

    Sources: Bloomberg (primary May 12), Yahoo Finance / Investing.com / Techzine / Moneycontrol (echo)

    Real but unsigned — term sheet not finalized, so investors should treat 'closes' vs 'collapses' as the binary read; boards on Anthropic-dependent vendor contracts can underwrite ~30 months counterparty runway conditional on close. What would change the read: lead investor list leak, a haircut to <$700B, or a slip past the end-May target.

  • 4 / 5

    OpenAI and Microsoft formalized a $38B cumulative cap on Microsoft revenue-share payments through 2030 (May 12); follows the late-2025 contract amendment that allowed OpenAI to add Anthropic-style multi-cloud commitments (SpaceX, Akamai, Amazon).

    Sources: The Hindu / Reuters-derived, ResultSense secondary; no MSFT 8-K inside-window

    Real and structural — first hard ceiling on any hyperscaler's AI revenue-share capture. Architects building on Azure-OpenAI should plan for Microsoft pushing MAI-line and native direct-API products into more enterprise workloads through 2027; operators should expect Azure-OpenAI commercial terms to harden post-cap. What would change the read: a counter-disclosure from MSFT IR walking back the cap framing or a different framing in the next MSFT 10-Q.

  • 4 / 5

    Samsung Electronics and union talks collapsed May 13 after 17 hours; an 18-day general walkout May 21-June 7 covering 50,000+ workers is locked in at the only company currently mass-shipping HBM4 for NVIDIA Vera Rubin; Samsung holds ~40% global DRAM share and a single April 23 half-day rally cut daily memory output 18.4%.

    Sources: Korea Herald, Seoul Economic Daily, TechTimes (May 16 synthesis)

    Real and consequential. The Vera Rubin H2 ramp leans on Samsung HBM4 — an 18-day walkout almost certainly compresses Q2-Q3 HBM4 output by single-digit-billion-dollar revenue impact and risks NVIDIA's May 20 print color. Operators planning H2 capacity should triage HBM allocation this week; investors should size the supply-chain wedge against Goldman's $80B Q1 forecast.

  • 2 / 5 — noise

    Alpha Compute closed a $32.2M two-year lease with a 'leading frontier AI lab' for 504 NVIDIA B200 GPUs at a Canadian data center (May 12), generating ~$16.1M annualized recurring revenue.

    Sources: Markets Insider / Business Insider (PR-led, single primary, counterparty unnamed)

    Mostly noise — small ARR contract dressed up with 'frontier AI lab' framing; counterparty undisclosed, deal size irrelevant at category level. Operators should not treat this as a comp for capacity pricing or deployment density; investors should discount until the lab is named and the deal is corroborated against the lab's public compute roadmap.

Early warning panel

The levers we monitor.

10 metrics tracked — 1 rising, 0 falling, 9 steady.

Current vs prior period. Each metric has a threshold where the read materially changes — this panel flags the inflection before it lands in headlines. Click any metric for the methodology and this-week read.

  • Frontier lab cash position (avg months runway, top 3)

    ~30 mo→vs ~30 mo

    Threshold: <18 mo triggers re-rating risk

    What this measures

    Top 3 frontier labs (OpenAI, Anthropic, Google DeepMind) by disclosed runway, computed from cash on hand divided by trailing-12-month operating burn. Anthropic's $30B / >$900B round (Bloomberg May 12) would extend top-3 average materially if closed by end-May; the round is unsigned. OpenAI Deployment Company launch (May 11) adds an embedded services cost layer that consumes some runway. Carry W19 figure flat until Anthropic round prices. <18 months triggers re-rating risk; <12 months forces consolidation, acquisition, or revenue reset.

  • Hyperscaler capex / AI revenue ratio (top 4 weighted)

    ~5.0-5.2→vs ~5.0-5.2

    Threshold: >6.0 invites investor pushback at next earnings

    What this measures

    Top 4 hyperscalers (MSFT, GOOG, META, AMZN) weighted aggregate of total capex divided by AI-attributable revenue. No fresh capex print in W20 — the next pulse points are Oracle's June print and Q2 FY27 / Q2 CY26 cycle in late July through mid-August. Custom-silicon ramp continues (Trainium $20B+ annualized, MTIA Gen 2 broad production, Maia 200 deployed, TPU v8 split SKUs). Ratio crosses the >6.0 threshold inside a quarter if AI revenue growth flinches.

  • CoreWeave revenue backlog

    $99.4B→vs $99.4B

    Threshold: Conversion velocity matters more than gross figure

    What this measures

    Booked but unrecognized revenue. The May 7 audited Q1 disclosure ($99.4B as of March 31, 2026) and the May 8 10-Q both landed inside W19 — the W19 watchlist 'CoreWeave 10-Q likely files inside W20' resolved one week early. No fresh CoreWeave disclosure inside W20; carry W19 figure. The next read is the Q2 print in early August. Conversion velocity (capacity online + customer ramp pace) remains the metric to watch.

  • NVIDIA Q-over-Q data center revenue

    $62.3B (Q4 FY26)→vs $62.3B (Q4 FY26)

    Threshold: Q1 FY27 print Tuesday May 20 — Goldman forecasts ~$70B+ DC implied

    What this measures

    Data-center segment revenue, sequential Q-over-Q. Q1 FY27 print scheduled for Tuesday May 20, 2026 (W21). Cross-currents heading into the print: Samsung HBM4 walkout (May 21-June 7) threatens H2 ramp; AMD Helios + Meta 6 GW shipping H2 adds credible second source on training. NVIDIA Vera Rubin mass-production reportedly finalized on TSMC N3 (May 11 supply-chain reports, grade 3 — unconfirmed). Beat or miss against Goldman's $70B+ DC implied sets the slope for H2 hyperscaler capex; circular-flow framing (CoreWeave + IREN warrants) gets stress-tested on the call.

  • Open vs closed gap on SWE-Bench Pro (coding)

    Closed +19pp→vs Closed +6 to +20pp

    Threshold: Sustained open lead reshapes enterprise procurement

    What this measures

    Coding benchmark differential between top open-weight and top closed model. May 13 BenchLM snapshot: Claude Mythos Preview (gated) 77.8 / Claude Opus 4.7 (Adaptive) 64.3 vs top open Kimi K2.6 / GLM-5.1 / MiMo V2.5 Pro at 58.6 / 58.4 / 57.2. Top-of-board gap (Mythos vs Kimi) is 19.2pp; production-tier gap (Opus 4.7 vs Kimi) is 5.7pp. Bifurcation continues — open-weight is competitive at GA tier but losing ground to frontier preview / research-tier closed models.

  • Sovereign AI commitments (count / aggregate $)

    12 / ~$80B+↑vs 10 / ~$80B+

    What this measures

    Two new sovereign initiatives in W20: Canada-TELUS BC Sovereign AI Factory (May 11, 60K+ NVIDIA GPUs by 2032 under federal Enabling Large-Scale Sovereign AI Data Centres program), and HMRC-Quantexa £175M / 10-year sovereign data and AI partnership (May 14). Program count nudges from 10 to 12; aggregate $ flat ($175M and Canada funding amounts not yet quantified). Sovereign-AI gravity continues to broaden beyond the UAE concentration.

  • PJM 2026/27 capacity auction price ($/MW-day)

    $329.17→vs $329.17

    Threshold: 11x in 24 months — power is the new binding constraint

    What this measures

    The 2026/27 BRA cleared July 2025 at $329.17/MW-day, unchanged. The 2027/28 BRA cleared Dec 17, 2025 at $333.44/MW-day at the FERC-approved cap; without the cap it would have cleared at $529.80 against the first RTO-wide reliability shortfall in PJM history. No FERC ruling or auction in W20 specifically — carry-forward methodology.

  • Time-to-power, busiest US markets (months)

    60-84 (new PJM); 36-48 (existing PJM queue)→vs 60-84 (new PJM); 36-48 (existing PJM queue)

    What this measures

    Months from new-load interconnection request to energization. PJM bifurcation holds: existing pipeline projects retain 36-48 month timeline; new large-load applications in Dominion territory face 5-7 year (60-84 month) windows. No fresh interconnect-queue disclosures inside W20. Architects siting greenfield AI capacity should pre-secure interconnect and assume the 'find power, build there' forcing function applies for any new application this cycle.

  • Cost-per-task, frontier reasoning model

    ~$0.10-$0.15 (effective, with hidden reasoning tokens)→vs ~$0.10-$0.15 (effective, with hidden reasoning tokens)

    Anthropic Agent SDK June 15 cutover will reset effective Claude Code economics

    What this measures

    Median cost across the frontier-tier reasoning models for a benchmark complex task. W19 methodology correction holds — published per-token rates understate true cost because hidden 'thinking' tokens are billed at output rate. W20 wrinkle: Anthropic's June 15 Agent SDK cutover moves Claude Code, GitHub Actions, and third-party agents off subscription rate limits onto separate $20-$200/mo metered credit at API list prices, raising effective cost 12x-175x per workload for arbitrage users. Architects running Claude Code at scale need to model the cutover this week.

  • Custom silicon share of incremental AI compute

    ~33-36% (estimated)→vs ~33-36% (estimated)

    Threshold: >35% materially compresses merchant GPU pricing

    What this measures

    Approximate share of newly deployed AI compute capacity using custom silicon (TPU, Trainium, MTIA, Maia) versus merchant silicon (NVIDIA, AMD). W20 catalysts adjacent but no fresh share print: MediaTek dual EMIB+CoWoS strategy for AI ASICs (TPU v8e expected on Intel EMIB) opens a second packaging source for custom silicon supply — when packaging diversifies, hyperscaler ASIC volume scales faster. Past 35% the compression is material and re-rates merchant gross margins; the next read is NVIDIA Q1 FY27 (May 20) commentary on customer-mix and Goldman's $70B+ DC implied number.

Predictions

What we expect next.

5 predictions for the next 30-90 days, confidence 55%-70%.

Each prediction is falsifiable, time-bounded, and tied to a specific signal we will watch. Future issues score these hit, miss, partial, or pending and build a public track record.

Prediction 01

55%

confidence

Hardware

Samsung HBM4 shipments to NVIDIA Vera Rubin drop by more than 25% in May vs April due to the May 21-June 7 walkout, with at least one downstream hyperscaler publicly delaying a Q3 capacity ramp.

Deadline: By July 31, 2026

Trigger: TrendForce / SemiAnalysis May-June supply data; NVIDIA Q1 FY27 commentary May 20; hyperscaler Q2 capex / supply commentary.

Prediction 02

65%

confidence

Software

Anthropic discloses or third-party developer surveys document a greater than 20% drop in active Claude Code third-party agent users within 30 days of the June 15 pricing cutover.

Deadline: By August 15, 2026

Trigger: Anthropic earnings / blog commentary; GitHub stars / npm downloads on Claude Agent SDK; third-party agent vendors (OpenClaw, T3 Code, Conductor, Zed) public statements.

Prediction 03

60%

confidence

Software

At least one Fortune 500 enterprise discloses a production deployment greater than $10M annualized of a specialist video / embodied multimodal model (Perceptron Mk1, MolmoAct 2 successors, or analogous) as a primary tier rather than a general LMM.

Deadline: By September 30, 2026

Trigger: Vendor case studies; enterprise architecture announcements; quarterly earnings color from Perceptron, Ai2 partners.

Prediction 04

70%

confidence

Capital

Cisco's Q4 FY26 AI orders confirm the $9B annual run rate (Q4 AI orders at least $2.5B), validating the enterprise AI fabric demand-pull narrative.

Deadline: By August 20, 2026

Trigger: Cisco Q4 FY26 earnings release (early August).

Prediction 05

60%

confidence

Networking

Expanded Beam Optical MSA publishes a v1.0 spec within 90 days of launch (May 12), with at least one in-production deployment announced by a hyperscaler member (AMD, Cisco, Meta, Oracle).

Deadline: By August 12, 2026

Trigger: ebomsa.org publications; member vendor blogs; OCP Future Technologies Symposium filings.

Track record

Scoring prior predictions.

5 prior predictions: 0 hit, 0 miss, 1 partial, 4 pending. Hit rate 0%.

5 predictions across issues so far. Hit rate: 0%. Hits 0, misses 0, partials 1, pending 4.

Prediction 01

65%

confidence

Software

At least one Fortune 500 enterprise discloses replacement of an existing voice / IVR platform with a frontier voice model (GPT-Realtime-2, Inworld TTS-2, or competitor) on a contract greater than $25M annualized.

Deadline: By July 31, 2026

Trigger: Enterprise customer-disclosure announcements; Q2 vendor commentary from OpenAI / Anthropic / Inworld; named Zillow / Priceline / Deutsche Telekom contract values.

pending— OpenAI realtime trio (GPT-Realtime-2 / Translate / Whisper) went broadly available May 11 with 128K context and adjustable reasoning effort. No fresh F500 disclosure of >$25M annualized contract this week.

Prediction 02

70%

confidence

Networking

At least one major hyperscaler other than Microsoft publicly discloses a production AI fabric running MRC at greater than 50,000-GPU scale.

Deadline: By August 31, 2026

Trigger: Google I/O 2026 (May 19-20) fabric session; AWS re:Invent prep coverage; OCP Future Technologies Symposium filings; vendor blog disclosures.

pending— No second-hyperscaler MRC adoption inside W20. Google I/O 2026 fabric session next week (May 19-20) is the next signal window.

Prediction 03

75%

confidence

Capital

Anthropic's pending raise closes at a final post-money valuation between $850B and $1T, with the lead investor and check size publicly disclosed.

Deadline: By June 15, 2026

Trigger: Anthropic primary disclosure; SEC Form D if applicable; secondary market activity; lead investor naming.

partial— Bloomberg May 12 confirms talks at >$900B with target close end-May; round still unsigned. On track for the June 15 deadline; lead investor not yet disclosed. Score moves to hit when the round prices and the lead is named publicly.

Prediction 04

55%

confidence

Hardware

A second public open-weights frontier-class model trained end-to-end on AMD silicon (MI300X or MI400) is announced, confirming AMD-as-training-substrate is structural rather than a one-off proof.

Deadline: By September 30, 2026

Trigger: Vendor announcements; arxiv preprints; AMD customer-list updates at Advancing AI 2026 (July 22-23); Rackspace + AMD MOU production milestones.

pending— No second open-weights frontier-class model on AMD silicon in W20. Advancing AI 2026 (July 22-23) is the next likely disclosure venue.

Prediction 05

70%

confidence

Capital

At least one major equity research firm or audit-grade publication explicitly frames NVIDIA's 2026 customer-equity stakes (CoreWeave + IREN, plus any subsequent counterparty) as a circular-financing risk in a written report or audit qualifier.

Deadline: By July 31, 2026

Trigger: NVIDIA Q1 FY27 earnings call (May 20); Q2 audit cycle commentary; sell-side and short-seller report releases.

pending— No equity research firm formally framed circular-financing risk in W20. The May 20 NVIDIA Q1 FY27 call is the next venue likely to surface the question.

Watchlist

On the radar this week.

5 catalysts to watch, starting May 19-20.

Specific catalysts that would change the read materially. Watching these tells us whether the thesis is strengthening or weakening.

  • May 19-20

    Google I/O 2026 — Gemini 3.2 Flash / Pro confirmation + fabric session

    Gemini 3.2 Flash leaked May 5 at ~$0.25 / $2.00 per Mtok with ~92% of GPT-5.5 capability; if confirmed at I/O, becomes the new price anchor for the cheap-frontier tier and forces OpenAI / Anthropic price moves within 60 days. Watch the fabric session for any second-hyperscaler MRC adoption (tests p13).

  • May 20

    NVIDIA Q1 FY27 earnings

    First disclosure on Vera Rubin ramp velocity, Samsung HBM4 supply commentary post-walkout context, and warrant-stake terms for CoreWeave / IREN / Corning equity investments. Goldman's $80B total / ~$70B+ DC implied bar sets the test. A beat resets H2 hyperscaler capex tail; a miss flattens the doubling slope; circular-flow framing stress-tests on the call (tests p16).

  • May 21 - June 7

    Samsung HBM4 walkout impact on daily output

    18-day general walkout locked in across 50,000+ workers at the sole HBM4 mass-shipper for NVIDIA Vera Rubin. A single half-day rally April 23 cut daily memory output 18.4%; an 18-day stoppage at the same scale puts single-digit-billion of Q2-Q3 HBM4 revenue at risk and threatens Vera Rubin H2 ramp commitments. Tests p17.

  • May 19-22

    Microsoft Build 2026

    First major Microsoft venue since the OpenAI revenue-share cap formalized. Watch for MAI-line product announcements that signal Microsoft pushing direct-API and native models into enterprise workloads; Azure-OpenAI commercial terms color; any new Bedrock-like distribution moves through OCI / Oracle. Conditions architects' Azure-OpenAI procurement read through 2027.

  • By end-May

    Anthropic $900B round close or pricing slip

    Bloomberg-confirmed target end-May close on $30B / >$900B round. The price ($900B vs OpenAI's $852B) and the lead investor's identity re-rate the entire frontier-lab valuation curve. A slip past end-May, a pricing haircut below $700B, or a walk-away from the price would be the loudest re-rating signal in a year. Tests p14.

Companion reads

The rest of the spine.

The AI Stack Weekly is the cross-stack flywheel read. Pair it with the model-and-tree spine and the working framework to get the full picture.

Edits this issue

  • — First Sunday-cadence publish. Window shifts from Friday-publish (Mon-Fri primarily) to Sunday-publish (full ISO Mon-Sun), catching Saturday and early Sunday news. W17 / W18 / W19 stay Friday-stamped historically.
  • — Sovereign AI commitments lever count nudged from 10 to 12 (Canada-TELUS BC + HMRC-Quantexa added).
  • — Cost-per-task lever detail updated to flag the Anthropic Agent SDK June 15 cutover as a 12x-175x effective-cost reset for Claude Code arbitrage users.
  • — Predictions p7-p11 (W18) outcomes were scored in W19; W19 priors p12-p16 carried forward with W20 notes — p14 (Anthropic round close) moves to partial.

About this brief

Compiled from public announcements, SEC filings, earnings transcripts, and official lab and vendor publications. Every quantitative claim is graded 1–5 on source quality. Claims graded 2 or below are flagged as noise. The thesis the brief defends is published separately and updated only when a hypothesis materially changes.

Authorship

Written by Brian Letort. Independent analysis. All sources cited are public. Not investment guidance.

Operate. Publish. Teach.