brianletort.ai
All issues

The AI Stack Weekly

Issue 02 · Week 18 of 2026.

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

The Bottom Line

Q1 broke the bear thesis. The marginal dollar started earning more — and the network still compounds.

Flywheel arcAll three lenses

The W17 read was 'the marginal dollar earns less at every layer except the network.' Q1 2026 hyperscaler prints (MSFT / GOOG / META / AMZN, all on Apr 29) reset the math. Aggregate 2026 capex stepped to $695-725B (+77% YoY) but AI revenue grew faster — Microsoft Cloud commercial RPO doubled YoY to $627B, Microsoft AI run rate +123% YoY to $37B, Google Cloud +63%, AWS +28% (a 15-quarter high).

The hyperscaler capex / AI revenue lever compressed from ~5.5 to ~5.0-5.2 — the first Q-over-Q compression in five quarters. The network hypothesis got an audited proof point in the same 48 hours: Equinix Q1 Fabric revenue +26% YoY with bookings +70% YoY at a record 51% adjusted EBITDA margin, raised guidance, ~60% of the largest Q1 deals AI-related.

Custom silicon at ~31% (Trainium's $225B forward book, Maia 200 going GA in Iowa, MTIA expanded with Broadcom and AMD lanes) is starting to compress merchant GPU pricing power. Capability gating became market structure rather than research-org concern: GPT-5.5-Cyber and Claude Security both shipped this week as gated, separately-priced enterprise products.

The arc the buildout is now on: less hype, more earn-out — boards should reset their AI-ROI thresholds upward, investors should track the capex / AI-revenue ratio (now ~5.0, threshold >6.0) as the wedge it has stopped being, and architects should pull custom-silicon ASIC TCO into procurement bake-offs that previously assumed merchant GPU dominance.

JevonsMetcalfeGilderSoftwareJevonsHardwareHuangNetworkingMetcalfe + Gilder

The three lenses

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

13 events across the flywheel — 5 software, 4 hardware, 4 networking.

Software.

  • UK AISI publishes GPT-5.5 cyber evaluation (71.4% Expert pass rate, highest tested; beats Mythos Preview at 68.6% and Opus 4.7 at 48.6%); OpenAI launches GPT-5.5-Cyber as restricted-access, adopting the same vetted-customer gating Sam Altman previously criticized when Anthropic used it for Mythos

    aisi.gov.uk, deploymentsafety.openai.com/gpt-5-5/cybersecurity, simonwillison.net

  • Anthropic ships Claude Security in public beta — an Opus 4.7-powered codebase vulnerability scanner with auto-patching, scheduled scans, and audit integrations; embedded with CrowdStrike, Microsoft Security, Palo Alto Networks, SentinelOne, Wiz, plus Accenture, BCG, Deloitte, Infosys, PwC

    claude.com/blog/claude-security-public-beta, securityweek.com

  • Mistral releases Medium 3.5 (open-weight modified MIT, 128B dense, 256K ctx) at 77.6% SWE-Bench Verified and $1.50/$7.50 per Mtok; replaces Devstral 2, Magistral, and Medium 3.1 in one consolidated release with Vibe remote cloud coding agents and Le Chat Work mode on the same runtime

    mistral.ai/news, the-decoder.com, awesomeagents.ai, decrypt.co

  • IBM releases Granite 4.1 LLM family (Apache 2.0, 3B / 8B / 30B dense decoders, 512K context, ~15T training tokens); 8B Instruct matches the prior 32B-A9B MoE Granite 4.0-H-Small at a fraction of the parameter count

    huggingface.co/blog/ibm-granite, ibm.com/granite/docs

  • GPT-5.5 + Codex + Bedrock Managed Agents land natively on AWS Bedrock in limited preview; OpenAI inference now billable against AWS commitments under IAM, PrivateLink, and CloudTrail — the same week Microsoft / OpenAI exclusivity formally ends and OpenAI's revenue-share is capped through 2030

    openai.com/index/openai-on-aws, aws.amazon.com What's New, blogs.microsoft.com

What this means

Two arcs converged in one week. Software-internal: AISI's GPT-5.5 cyber numbers and the simultaneous launch of OpenAI's gated GPT-5.5-Cyber and Anthropic's Opus-4.7-powered Claude Security make frontier-grade cyber an explicit, separately-priced product line — CISOs and security architects must update agentic threat models this quarter and pick a defender stack now. Software-distribution: GPT-5.5 going Bedrock-native plus the end of Microsoft exclusivity decouples closed-frontier intelligence from cloud residency for the first time, so boards previously locked in by Azure-only constraints should reopen OpenAI procurement on AWS or GCP within the limited-preview window before pricing hardens at GA. See the Model Pulse for the full architecture read on the six new tree rows that landed this week.

Hardware.

  • Amazon Q1 2026 8-K: Trainium revenue commitments hit $225B; chip business at >$20B annualized run rate (now top-3 data-center silicon globally); Trainium3 nearly fully subscribed, Trainium4 already partly reserved; Anthropic 5 GW + OpenAI ~2 GW commitments confirmed

    Amazon 8-K (sec.gov), Amazon Q1 press release, About Amazon (Jassy commentary), BusinessWire

  • Microsoft FY26 Q3 8-K: FY26 capex raised to ~$190B (+61% YoY); Q3 capex $31.9B; Maia 200 went GA in US Central (Iowa) with US West 3 (Phoenix) next, claiming '30% better tokens per dollar' vs current fleet silicon; Microsoft cited a ~$25B memory-cost headwind

    Microsoft 8-K (sec.gov), Microsoft IR FY26 Q3 release, news.microsoft.com, CNBC

  • Meta Q1 2026 8-K: 2026 capex band raised to $125-145B (from $115-135B); management named memory pricing as the primary driver; Zuckerberg confirmed broader rollout of Broadcom-co-developed MTIA silicon plus AMD Instinct GPUs alongside NVIDIA in 2026 fleets

    Meta 8-K (sec.gov), Meta Q1 press release, Bloomberg, AP

  • Samsung Q1 2026: confirmed industry-first HBM4 mass production for NVIDIA Vera Rubin began Feb 2026; HBM4E sampling targeted Q2 2026 (16 Gbps pin, 4.0 TB/s/stack); HBM revenue projected to more than triple YoY in 2026 with HBM4 forecast >50% of HBM mix from Q3

    Samsung Q1 2026 release (news.samsung.com), CNBC, MarketScreener earnings transcript, Digitimes

What this means

Three SEC-disclosed CSP earnings inside a 48-hour window made W17's thesis tangible: custom silicon's share of incremental compute is pushing past the ~30% baseline (Trainium's $225B forward book + Maia 200 in two regions + Meta's expanded MTIA and AMD lanes), and the binding constraint is now memory on the record — Microsoft's $25B memory headwind and Meta's capex hike both explicitly cite memory pricing, while Samsung's Apr 30 HBM4 mass-production confirmation for Vera Rubin finally adds a qualified second source against SK Hynix's monopoly. Architects and operators planning H2 2026 capacity should re-bid GPU/ASIC mix assuming Samsung HBM4 is now real (de-risks Vera Rubin allocation); investors should model HBM pricing — not GPU pricing — as the swing variable on hyperscaler ROI through 2026.

Networking.

  • Equinix Q1 2026 8-K reports Fabric revenue +26% YoY, Fabric bookings +70% YoY, +5,800 net physical and virtual interconnections, record 51% adjusted EBITDA margin, full-year guidance raised to 10-11% revenue growth — ~60% of the largest Q1 deals AI-related

    Equinix 8-K (SEC EDGAR), Q1 2026 earnings call transcript, Insider Monkey transcript

  • Celestica opens orders for DS6000-series 1.6TbE switches (64 x 1.6TbE OSFP224, 102.4 Tbps non-blocking, Broadcom Tomahawk 6 silicon, SONiC) compliant with both UltraEthernet Consortium and OCP ESUN — the open-standard 1.6T scale-up path moves from spec to procurement-ready SKU at OCP EMEA Summit Barcelona

    Celestica press release / Globe Newswire, Broadcom Core Switching Group commentary, Dell'Oro Group analyst quote

  • Hedgehog contributes OCP-Accepted AI training and AI inference fabric reference architectures (validated with Celestica and FarmGPU) to the OCP Marketplace at OCP EMEA — first vendor-neutral, multi-silicon, Ethernet-based AI fabric blueprints available as production-grade designs

    Hedgehog press release / PR Newswire, OCP Marketplace listing, OCP Foundation commentary

  • CoreWeave + Google Cloud launch CoreWeave Interconnect (private fiber with line-rate MACsec on Google's Partner Cross-Cloud Interconnect), SUNK Anywhere (Slurm-on-Kubernetes extending to AWS and Azure), and LOTA Cross-Cloud (cache giving 7 GB/s/GPU near-local throughput, no egress fees) — a 44-DC neocloud GPU operator becomes a cross-DC training-fabric provider

    CoreWeave blog announcement, Google Cloud Next 2026 keynote, DCD coverage

What this means

Equinix's SEC-grade Q1 — Fabric revenue +26% YoY, Fabric bookings +70% YoY, record 51% EBITDA margin, raised 2026 guidance — is the audited proof point for the publication's durable-fabric thesis; the same week, three independent open-standard moves (Celestica UEC/ESUN-compliant 1.6TbE on order, Hedgehog OCP-Accepted AI fabric reference designs, CoreWeave-Google cross-cloud Interconnect) push the open AI-fabric path from spec to procurement-ready SKU. Architects should re-test single-vendor scale-up roadmaps against an open Tomahawk 6 + UEC + ESUN + OCP-Accepted blueprint stack now bookable today; boards and investors got an audited Metcalfe data point that justifies re-underwriting interconnect-category pricing power; operators should treat the CoreWeave-Google move as the week the colo-vs-neocloud connectivity boundary started dissolving.

Capital flow

Money in, revenue out.

4 categories tracked. Capital deployment up in 3 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 ~$50B

    Revenue Out

    ~$35B ↑

    vs ~$30B

    Burn / Rev

    0.6

    Movement

    Anthropic disclosed ~$30B ARR (3x in four months) and is fielding offers for a ~$50B raise at a $900B pre-money on Apr 29 — frontier labs now reprice as scaling enterprise-software businesses with 8 of Fortune 10 paying. Boards and LPs should underwrite ARR cadence and contract concentration on hyperscaler GPU/Trainium supply, not story or option value.

  • Hyperscaler-Hosted

    Microsoft Azure, Google Cloud, AWS, Meta

    Capital In

    ~$58B ↑

    vs ~$48B

    Revenue Out

    ~$13B ↑

    vs ~$10B

    Burn / Rev

    0.22

    Movement

    All four hyperscalers reset 2026 capex on Apr 29 to a combined $695-725B (+77% YoY) — MSFT $190B, GOOG $180-190B, META $125-145B, AMZN $200B — while every CFO repeated 'compute constrained.' Architects should assume zero capacity relief through 2026 and lock multi-year terms now; investors should monitor the capex / AI-revenue ratio (currently ~5.0-5.2, threshold >6.0) since the ratio only stays healthy if AI revenue keeps growing 60-120% YoY.

  • Neoclouds

    CoreWeave, Crusoe, Nebius, Applied Digital

    Capital In

    ~$18B ↑

    vs ~$15B

    Revenue Out

    ~$2.0B ↑

    vs $1.6B

    Burn / Rev

    0.11

    Movement

    CoreWeave's April closed three multi-year anchor commits (Meta +$21B Apr 9, Anthropic Apr 10, Jane Street $6B + $1B equity Apr 15) while Oracle landed a $16B project-finance package for a 1.4 GW OpenAI campus in Michigan (Apr 25, BofA-led) — neocloud capacity is now structurally pre-sold through 2032. Operators should look past headline backlog to customer concentration (4-5 frontier customers ~80%+) and the project-finance debt structure that will start showing up in covenant disclosures.

  • On-Prem / Hybrid

    Enterprise GPU clusters, sovereign and national programs

    Capital In

    ~$42B →

    vs ~$42B

    Revenue Out

    Indirect →

    vs Indirect

    Burn / Rev

    n/a

    Movement

    Apple posted record R&D $11.4B (+34% YoY, Apr 30) and called out 'amazing platforms for AI and agentic tools' on Mac Studio and mini, while Amazon's custom-silicon line (Graviton, Trainium, Nitro) crossed a $20B annual run rate at triple-digit YoY growth — on-prem and sovereign AI buildouts are now first-class procurement tracks, not edge cases. Operators evaluating new AI capacity should run a parallel on-prem / sovereign track alongside hyperscaler RFPs; architects should assume the 36-48mo time-to-power constraint applies equally and pre-secure power siting now.

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.

6 claims this week — 4 signal, 2 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

    Big-4 hyperscaler 2026 capex resets to a combined $695-725B (+77% YoY): MSFT $190B (vs $154.6B consensus), GOOG $180-190B, META $125-145B, AMZN $200B reaffirmed — all four CFOs repeated 'compute constrained.'

    Sources: Microsoft FY26 Q3 press release (Apr 29), Alphabet Q1 2026 8-K (SEC EDGAR, Apr 29), Meta Q1 2026 release (Apr 29), Amazon Q1 2026 release (Apr 29); rolled up in Business Insider, Tom's Hardware, Reuters

    Real and load-bearing. The capital-flow bear thesis ('hyperscalers will pull back') is gone for 2026. Read changes only if Q2 prints walk back the guides or if a regulator constrains a major utility / grid project; otherwise expect the $725B floor to hold and stress to migrate to power siting and supply-chain memory pricing.

  • 5 / 5

    Microsoft AI annual run rate is $37B (+123% YoY) and Microsoft Cloud commercial remaining performance obligation is $627B (+99% YoY) as of March 31, 2026.

    Sources: Microsoft FY26 Q3 press release (Apr 29, primary 8-K disclosure)

    Real. RPO doubling YoY is the cleanest single proof that the AI commercial ramp is durable contract revenue, not pilot spend. Read changes only if Microsoft starts disclosing AI gross margin separately and it deteriorates; the run-rate itself is now an audited number.

  • 5 / 5

    Equinix Q1 2026 reports Fabric revenue +26% YoY, Fabric bookings +70% YoY, +5,800 net interconnections, record 51% adjusted EBITDA margin; full-year guidance raised to 10-11% revenue growth.

    Sources: Equinix 8-K (SEC EDGAR, Apr 29), Q1 2026 earnings call transcript

    Real. The publication's durable-network thesis got an audited Q1 proof point. Combined with Apr 28-30 open-fabric procurement-ready SKUs (Celestica 1.6TbE, Hedgehog OCP-Accepted, CoreWeave Cross-Cloud), interconnect pricing power compounds rather than compresses through 2026.

  • 4 / 5

    Anthropic disclosed ~$30B annualized revenue in April 2026 (3x in four months; some sources cite ~$40B internal figure), with 1,000+ enterprise customers spending >$1M/yr and 8 of Fortune 10 as Claude customers.

    Sources: CNBC, Bloomberg, TechCrunch, Reuters all sourcing the disclosure to Anthropic in connection with $900B raise talks; Sacra and AI Business Review confirm trajectory; Anthropic is private, no SEC filing

    Real, but private-company disclosure — not audited. Read changes if the $50B round prices materially below $900B, which would imply investor pushback on the revenue durability or margin profile.

  • 2 / 5 — noise

    Anthropic in talks to raise ~$50B at a $900B pre-money valuation, which would top OpenAI's $852B March valuation.

    Sources: CNBC (Apr 29), Bloomberg, TechCrunch — all citing 'sources'; no term sheet signed; board decision expected May

    Real that the talks are happening, but the price is not yet committed capital. Treat as sentiment indicator, not balance-sheet event. Read changes when the round prices, anchor lead investor is named publicly, or a major LP walks away — any of which would re-rate the entire frontier-lab valuation curve.

  • 2 / 5 — noise

    OpenAI's GPT-6 / 'Spud' is launching imminently (late April through May 25, 2026) with a 2M-token context and 40% performance gains over GPT-5.4.

    Sources: Various tech blogs (Remio, FindSkill, BigGo, vgtimes); Polymarket probability collapsed from 93% by Jun 30 to 45%; OpenAI has confirmed pre-training completed Mar 24 but not announced a launch date or final naming

    Hype-adjacent. The completed pre-training is real (Altman, Brockman public statements); the launch date and capability claims are speculation. Architects should not pre-commit deployment timelines or budget shifts on this. Read changes only on an OpenAI announcement of GA timing and pricing.

Early warning panel

The levers we monitor.

10 metrics tracked — 3 rising, 2 falling, 5 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 ~22 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. Q1 mega-rounds (OpenAI $122B, Anthropic $30B, xAI $20B) plus Amazon's $25B and Google's $40B Anthropic compute commitments materially extended runway; OpenAI alone has multi-year cushion at projected ~$17B/yr 2026 burn. <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 ~5.5

    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. Compressed for the first time in five quarters: aggregate 2026 capex stepped to ~$725B (+77% YoY), but AI revenue is accelerating faster — MSFT AI run rate +123% YoY to $37B, Azure +40%, Google Cloud +63%, AWS +28% (15-quarter high). Ratio crosses the >6.0 threshold inside a quarter if AI revenue growth flinches.

  • CoreWeave revenue backlog

    ~$87B+ (est)↑vs $66.8B

    Threshold: Conversion velocity matters more than gross figure

    What this measures

    Booked but unrecognized revenue. $66.8B Feb 2026 base + $21B Meta expansion (Apr 9, lifts Meta total to $35.2B through 2032) + $6B Jane Street (Apr 15) + Anthropic multi-year (Apr 10, value undisclosed) — pending CRWV Q1 print mid-May for the audited figure. The metric to actually watch is conversion velocity: how quickly capacity comes online and customers ramp.

  • NVIDIA Q-over-Q data center revenue

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

    What this measures

    Data center segment revenue, sequential Q-over-Q. Carried from W17; Q1 FY27 print expected late May 2026. Captures Blackwell Ultra ramp and start of Vera Rubin sample shipments; Samsung HBM4 second-source confirmation Apr 30 is the upstream signal that protects H2 ramp. Beat or miss against the $62B run-rate sets the slope for H2 hyperscaler capex.

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

    Closed +6 to +19pp↓vs Open +0.9

    Threshold: Sustained open lead reshapes enterprise procurement

    What this measures

    Coding benchmark differential between top open-weight and top closed model. The W17 'open +0.9' lead has reverted: top closed (Claude Mythos Preview gated, GPT-5.3-Codex public) at 77+ vs top open (Kimi K2.6, GLM-5.1, MiMo V2.5 Pro) at 56-58 on the public board; SEAL 250-turn methodology shows even wider Anthropic lead. Methodology variance is high; the directional flip is the signal — open-weights leadership on coding was a single-model methodology artifact, not a structural inflection.

  • Sovereign AI commitments (count / aggregate $)

    8 / ~$80B+→vs 8 / $80B+

    What this measures

    No new sovereign mega-deal closed in W18; Stargate UAE (G42 / OpenAI / Oracle / NVIDIA, $30B / 5 GW) first phase 200 MW still tracks Q2-Q3 2026, full 1 GW by 2028. Current set: UAE Stargate (1 GW), Germany National DC Strategy, France IA program, Mistral-Sweden, GMI Japan, IndiaAI, UK AI Growth Zones aggregating $38.5B, plus Saudi-PIF AI commitment.

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

    $329.17→vs $329

    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 (first RTO-wide reliability shortfall in PJM history) — outside the W18 14-day window so not refreshed here, but a leading signal architects should price in for next-cycle siting decisions.

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

    36-48 (PJM large >100 MW); ~53 nat'l avg→vs 36-48

    What this measures

    Months from new-load interconnection request to energization. PJM and CAISO both >58 mo for full queue. Microsoft and Alphabet both flagged 'compute constrained' on Apr 29 calls — power is the binding constraint, not silicon, through 2026. The constraint is no longer chips, capital, or land — it is electricity delivery.

  • Cost-per-task, frontier reasoning model

    ~$0.05→vs ~$0.05

    Falling cost expands workloads (Jevons), not contracts demand

    What this measures

    Median cost across the frontier-tier reasoning models for a benchmark complex task. No fresh public benchmark this week, but custom-silicon TCO disclosures (Trainium / TPU / MTIA / Maia) continue to point to 30-60% inference cost reduction vs GPU at hyperscaler scale; pricing has not been fully passed through to API list rates yet. Falling cost expands the addressable workload set rather than contracting demand — the Jevons signature in AI inference.

  • Custom silicon share of incremental AI compute

    ~31%↑vs ~30%

    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). Amazon disclosed Graviton/Trainium/Nitro >$20B annual run rate at triple-digit YoY (Apr 29); TrendForce projects 27.8% ASIC share for 2026, with Trainium 3 ramp Q2 2026, Microsoft Maia 200 GA on TSMC 3nm, Google TPU v7 Ironwood. NVIDIA share trending from 86% (2024) toward ~75% (2026). Past 35% the compression is material and re-rates merchant gross margins.

Predictions

What we expect next.

5 predictions for the next 30-90 days, confidence 60%-65%.

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

65%

confidence

Software

AWS reports a customer-disclosed Bedrock-resident GPT-5.5 multi-year contract greater than $500M annualized, with the customer named publicly.

Deadline: By July 31, 2026

Trigger: AWS customer-win announcements during the Bedrock GPT-5.5 limited preview; OpenAI partnership commentary in Q2 prints.

Prediction 02

60%

confidence

Capital

The top-4 hyperscaler capex / AI revenue ratio drops below 5.0 for Q2 2026, confirming the W18 directional flip.

Deadline: By August 15, 2026

Trigger: Q2 2026 hyperscaler earnings releases (late July through mid-August).

Prediction 03

60%

confidence

Hardware

Samsung HBM4 reaches greater than 25% share of NVIDIA Vera Rubin BOM by Q3 2026 supply data, ending the SK Hynix monopoly on the platform.

Deadline: By September 30, 2026

Trigger: TrendForce / SemiAnalysis Q3 BOM data; NVIDIA Q2 FY27 earnings color on memory supply.

Prediction 04

65%

confidence

Networking

At least one major colocation operator other than Equinix reports Q2 2026 interconnect / fabric revenue growth greater than 25% YoY.

Deadline: By August 31, 2026

Trigger: Q2 earnings prints from interconnect operators; Light Reading and DCD coverage.

Prediction 05

60%

confidence

Software

By Q3 2026, at least two frontier labs publicly disclose a separately-priced 'gated cyber' or 'gated security' SKU with revenue commentary, formalizing capability-gating as a product line.

Deadline: By September 30, 2026

Trigger: Q3 earnings disclosures (OpenAI partnership color, Anthropic announcements); vendor security press; AISI / METR cross-vendor evaluations.

Track record

Scoring prior predictions.

6 prior predictions: 0 hit, 0 miss, 2 partial, 4 pending. Hit rate 0%.

6 predictions across issues so far. Hit rate: 0%. Hits 0, misses 0, partials 2, pending 4.

Prediction 01

70%

confidence

Capital

At least one frontier lab announces a customer-funded compute commitment greater than 2 GW.

Deadline: By June 30, 2026

Trigger: Earnings cycle commentary on RPO from Oracle and AWS; announcements at Google I/O, Microsoft Build, AWS Summit.

pending— OpenAI Stargate US footprint disclosed at >10 GW (Apr 30) is structural, not a single customer-funded announcement >2 GW. Watch I/O and Build in May.

Prediction 02

60%

confidence

Software

At least one Fortune 500 enterprise discloses an on-prem AI workload greater than $100M annual using open-weight models.

Deadline: By September 30, 2026

Trigger: Enterprise architecture announcements; bank, insurer, or pharma F500 first-mover; DeepSeek V4 reference deployment.

pending— No F500 disclosure of >$100M annual on-prem open-weights workload this week. SWE-Bench Pro lever flipped back to closed lead, which weakens the open-weights procurement catalyst.

Prediction 03

80%

confidence

Capital

Aggregate 2026 hyperscaler capex revises upward by 10% or more from the $700B baseline.

Deadline: By October 31, 2026

Trigger: Q1 2026 earnings (Apr 29 - May 6) and Q2 2026 earnings cycle.

partial— Q1 prints (MSFT $190B, GOOG $180-190B, META $125-145B, AMZN $200B reaffirmed) take 2026 aggregate to $695-725B (+77% YoY) vs the $700B W17 baseline. At/near baseline; +10% revision (~$770B) plausible by Q2 print. Score moves to hit if Q2 takes aggregate above $770B.

Prediction 04

70%

confidence

Networking

At least one major colocation or interconnect operator reports cross-connect or interconnect revenue growth outpacing compute capacity revenue growth for two consecutive quarters.

Deadline: By July 31, 2026

Trigger: Equinix Q1 (May 7) and Q2 (early August) earnings; broader colo and IX operator reporting.

partial— Equinix Q1 Apr 29: Fabric revenue +26% YoY, Fabric bookings +70% YoY, raised guidance — first of two quarters needed. Score moves to hit if Q2 confirms the same outpacing, expected early August.

Prediction 05

35%

confidence

Capital

At least one neocloud loses an anchor tenant or sees backlog growth turn negative quarter-over-quarter.

Deadline: By September 30, 2026

Trigger: Quarterly disclosures from CoreWeave, Nebius, Applied Digital.

pending— Opposite signal: CoreWeave added $21B Meta (Apr 9), $6B Jane Street (Apr 15), Anthropic multi-year (Apr 10) — backlog grew, not shrank. Lower confidence on this prediction now.

Prediction 06

55%

confidence

Hardware

Custom silicon (TPU + Trainium + Maia + MTIA + Granite Rapids AI) reaches 35% of incremental AI compute share, up from ~30% today.

Deadline: By July 31, 2026

Trigger: TrendForce / SemiAnalysis quarterly mix breakdown; Q2 2026 hyperscaler earnings color on internal silicon vs merchant GPU mix.

pending— Custom silicon at ~31% (up from ~30%); 35% threshold plausible by July 31 given Trainium $225B forward book, Maia 200 GA in two regions, MTIA expanded. No fresh share data crossed the threshold this week.

Watchlist

On the radar this week.

5 catalysts to watch, starting May 5-20.

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

  • May 5-20

    Mythos status update from Anthropic + AISI

    AISI's Apr 27-28 sabotage evaluation flagged a 65% reasoning-output discrepancy in Mythos continuation tests. Anthropic's response — extend gating, lift gating under conditions, or expand Project Glasswing — sets the precedent for capability-gated SKUs as a market category.

  • Late May 2026

    NVIDIA Q1 FY27 earnings

    First disclosure on Vera Rubin ramp velocity, Samsung HBM4 second-source effect, and custom-silicon competitive pressure (Trainium $225B book, Maia 200 GA). A beat against the $62.3B Q4 FY26 baseline resets the H2 hyperscaler capex tail; a miss flattens Huang's slope.

  • May 5-20

    Anthropic $900B raise terms and lead investor

    Talks reported Apr 29 are not yet committed capital. The price ($900B vs OpenAI's $852B) and the lead investor's identity re-rate the entire frontier-lab valuation curve. A walk-away from the price would be the loudest re-rating signal in a year.

  • May 5-15

    Q1 capex revision wave (Apple, Oracle, second-tier hyperscalers)

    The big-4 reset 2026 capex to $695-725B; secondary cohort (Apple, Oracle, IBM Cloud, smaller hyperscalers) has not yet revised. A second wave of capex hikes pushes the aggregate above $750B and tests prediction p3 to hit before Q2 prints.

  • May 2-9

    Google DeepMind: Gemini 3.1 Pro to GA, or Gemini 3.2 announcement

    Gemini 3.1 Pro Preview has held its AA Index 57 spot since February while OpenAI and Anthropic shipped new flagships. A GA or 3.2 announcement would force a frontier scorecard refresh and potentially retake LMArena top spot from Opus 4.7 Thinking.

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 Q1 hyperscaler-earnings-cycle issue. Capital flow reset to Q1 2026 reported numbers (MSFT FY26 Q3, GOOG / META / AMZN calendar Q1).
  • — Hyperscaler capex / AI revenue lever direction reversed: ~5.5 (W17) to ~5.0-5.2 (W18) — first Q-over-Q compression in five quarters. Threshold framing unchanged.
  • — Open vs closed gap on SWE-Bench Pro lever direction reversed: 'Open +0.9' to 'closed +6 to +19pp' (methodology-dependent). The W17 open-weights coding inflection partially reverted; lever detail updated to note methodology variance.
  • — First weekly Pulse companion publishes alongside this issue at /industry/models/2026-W18.

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.