The AI Stack Weekly

Issue 09 · Week 25 of 2026.

June 20, 2026/Industry brief · ~7 min read/Public sources onlyDownload brief

The Bottom Line

Power policy and open weights became the week's leverage while the closed frontier stalled.

Flywheel arcAll three lenses

W25 moved at the two ends of the stack that the AI factory actually depends on — the grid and the open model layer — while the closed frontier marked time. The single most consequential event was regulatory: on Jun 18 FERC issued six simultaneous Section 206 show-cause orders telling PJM, MISO, SPP, CAISO, ISO-NE, and NYISO their tariffs 'appear unjust and unreasonable' for large data-center loads, forcing co-location/behind-the-meter rules and cost-shift transparency within 60 days. Developers kept routing around the interconnection queue entirely (Cummins–Circe's 2GW behind-the-meter gas) while hyperscalers added grid-connected capacity explicitly paying 100% of interconnection cost (Amazon's $10B Missouri campus, Google's $1.5B Alabama expansion). At the model layer the story was openness and stasis: Z.ai shipped GLM-5.2 under a genuine MIT license, which independent testing says beats GPT-5.5 on several long-horizon coding benchmarks at roughly one-sixth the cost; MiniMax-M3's sparse-attention weights matured; and Artificial Analysis rebased its Intelligence Index to v4.1 around agentic tasks — all while Anthropic's Claude Fable 5 stayed government-suspended the entire week (Opus 4.8 remained the working leader) and Gemini 3.5 Pro slipped to July. Hardware and networking were quiet (Supermicro opened Vera Rubin NVL72 order availability; Coherent expanded indium-phosphide laser capacity). Net/net: boards should treat power policy as a board-level siting and cost risk now being repriced by FERC; investors should watch the open-weight cost curve collapsing frontier-adjacent capability toward commodity inference; architects should pilot MIT-licensed GLM-5.2 for sovereign and self-host coding while keeping multi-model fallbacks given the closed frontier's demonstrated fragility; operators should lock generation and long-lead transformers before GPUs.

The three lenses

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

10 events across the flywheel — 4 software, 3 hardware, 3 networking.

Software.

Jun 16
Z.ai released GLM-5.2 under an MIT license (~744B / ~40B-active sparse-attention MoE, 1M context); independent testing reports it beats GPT-5.5 on several long-horizon coding benchmarks at ~1/6 the cost, and Artificial Analysis cites it as the leading open-weight model
Z.ai blog; VentureBeat; Hugging Face
Jun 15
Artificial Analysis rebased its Intelligence Index to v4.1, re-weighting the industry's headline benchmark around agentic tasks (GDPval-AA v2, Terminal-Bench, banking agents); scores are not back-comparable to v4.0
Artificial Analysis
Jun 15-21
Anthropic's Claude Fable 5 and Mythos 5 remained government-suspended for the full week with no restoration; Opus 4.8 stayed the working closed-frontier leader as queries continued to fall back to it
Anthropic
Jun 18
Gemini 3.5 Pro slipped its GA from June to July: still a limited Vertex preview with no public model card, pricing, or independent benchmark vs Fable 5/Opus 4.8
Business Insider; Google

What this means

The closed frontier paused — no GA from OpenAI, Google, or a restored Anthropic — and the momentum shifted to open weights and to how 'best' is measured. Architects should pilot MIT-licensed GLM-5.2 for self-host/sovereign coding now and re-baseline on the agentic AA v4.1 index; see The Model Pulse for the full architecture read.

Hardware.

Jun 17
Supermicro opened customer engagement / order availability for Vera Rubin NVL72 and HGX Rubin NVL8 reference designs (1,152-GPU scalable unit, 331TB HBM4), though deployments remain H2 2026 at GA — order availability, not shipments
The Data Center Engineer (Supermicro DCBBS)
Jun 16
HPE Discover expanded Private Cloud AI with NVIDIA Vera-CPU servers (ProLiant DL394) and the NVIDIA Agent Toolkit — on-prem agentic-AI enablement rather than new silicon or shipments
SiliconANGLE
Jun 18
A J.P. Morgan note pegged the 2026 custom-ASIC market at ~$60-70B with Broadcom at 80-85% and Marvell 10-12% share, reinforcing the ASIC-overtakes-merchant-GPU-units thesis for 2027
J.P. Morgan note (Finvaulta summary)

What this means

Hardware was a follow-through week: the marquee catalysts (Micron earnings, SK hynix HBM4 dynamics, NVIDIA ISC, Groq's raise) all landed Jun 22-24, just outside the window. In-window the signal was OEM/system enablement (Supermicro order availability, HPE Vera CPU) and the custom-silicon thesis hardening. Operators should treat Vera Rubin as orderable but not shippable until H2; investors should keep diversifying toward the Broadcom/Marvell ASIC duopoly.

Networking.

Jun 16
Coherent broke ground on a Sherman, TX indium-phosphide fab expansion and signed a $50M CHIPS LOI to roughly quadruple InP wafer output — the lasers feeding 800G-to-1.6T transceivers and NVIDIA CPO external laser sources
Coherent; NVIDIA blog; The Register
Jun 17
HPE Juniper began shipping the QFX5250-64OE-L, an industry-first 1.6T, fully liquid-cooled 102.4T switch (64x 1.6T ports, ORV3-compliant) built on Broadcom-class silicon
Juniper community blog
Jun 17
Marvell published a plasmonics optical-roadmap (with Polariton/ETH Zurich) citing a 1.1THz modulator lab record and 400G/lane SiPho devices — a forward R&D signal, not a commercial design win
Marvell blog

What this means

Networking was quiet between the March (OFC) and June-1 (earnings) news clusters: the week's signal was supply-chain capacity (Coherent's InP expansion addressing the laser chokepoint) plus a liquid-cooled 1.6T system, not a new chip design win. Critically, no second non-Broadcom vendor (Marvell or Credo) cited a fresh 1.6T/CPO production win — architects pressing for a second silicon source still have to wait.

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.

What’s real, what’s noise.

5 claims this week — 3 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
FERC issued six simultaneous Section 206 show-cause orders rewriting large-load/data-center interconnection across all US RTOs/ISOs.
Sources: FERC press release and orders; Day Pitney and Akin/JDSupra legal analyses. Primary regulatory action, multi-source.
SIGNAL. The rules for how AI campuses connect and who pays are being rewritten nationally on a 60-day clock. This is a board-level siting and cost variable — pro-speed medium-term, but it injects near-term regulatory uncertainty into every US data-center plan.
4 / 5
Z.ai's GLM-5.2 shipped MIT open weights that beat GPT-5.5 on several long-horizon coding benchmarks at ~1/6 the cost and lead the open-weight field.
Sources: Z.ai release and Hugging Face card (primary); VentureBeat independent testing; Artificial Analysis placement. Caveat: SWE-Bench/Aider numbers not yet fully settled.
SIGNAL. A genuinely permissive (MIT, no regional limits) frontier-adjacent open model is now self-hostable at a fraction of closed-API cost. Architects should pilot it for sovereign/self-host coding; it pressures closed-flagship pricing.
4 / 5
Anthropic's Claude Fable 5 and Mythos 5 remained globally suspended for the entire week following the Jun 12 government order.
Sources: Anthropic news page (primary) plus secondary trackers; no restoration date announced.
SIGNAL. The sovereign/regulatory single-point-of-failure on a top-tier closed model is not a one-day outage — it persisted a full week. Keep Opus 4.8/Sonnet fallbacks wired and do not single-source the frontier for production-critical paths.
2 / 5 — noise
GPT-5.6 (OpenAI) and Grok V9 (xAI) are launching imminently and will reset the frontier.
Sources: Backend codenames, transient routing strings, and prediction-market odds only; no system card or API contract. GPT-5.5 remains the documented flagship; 'Grok V9' does not exist.
NOISE. Rumor with no primary artifact. Keep current baselines and do not delay procurement on an unreleased SKU; any GPT-5.6 was tracking to land after this window.
2 / 5 — noise
Meta has signed a ~1.6GW compute deal with Crusoe.
Sources: Bloomberg-sourced reporting (Jun 18); both Meta and Crusoe declined to comment, terms undisclosed.
NOISE / unconfirmed. Treat as reported, not closed. The structural pattern (Meta adding external behind-the-meter compute) is real, but the specific deal is not yet verifiable for a procurement or investment decision.

Early warning panel

The levers we monitor.

10 metrics tracked — 2 rising, 2 falling, 6 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.

Metric

Current

Prior

Dir

Threshold

Frontier lab cash position (avg months runway, top 3)
~34-37 mo (flat; no new in-window round)
~34-37 mo (xAI now public via SpaceX IPO)
→
<18 mo triggers re-rating risk
What this measures
Top 3 frontier labs (OpenAI, Anthropic, Google DeepMind) by disclosed runway, with xAI now public inside SpaceX. No new financing event in-window; both OpenAI and Anthropic remain at the confidential-draft-S-1 stage. Capital access stays wide; boards should not assume funding pressure forces near-term commercial concessions.
Hyperscaler capex / AI revenue ratio (top 4 weighted)
~5.0-5.3 (Amazon $10B + Google $1.5B added; top-4 guides flat)
~5.0-5.3 (Oracle FY27 ~$70B guide; top-4 flat)
→
>6.0 invites investor pushback at next earnings
What this measures
Top 4 hyperscalers (MSFT, GOOG, META, AMZN) weighted aggregate of capex divided by AI-attributable revenue. No new top-4 earnings print in-window; the incremental Amazon and Google campus commitments fit prior ~$725-805B 2026 capex guides. Investors should keep the bubble hypothesis on funding/conversion and now on FERC-driven siting cost, not demand.
CoreWeave revenue backlog
~$100B (Jun 15); Cantor estimate ~$131B by end-Q2
$99.4B (next print Q2, early Aug)
↑
Conversion velocity matters more than gross figure
What this measures
Booked but unrecognized revenue. Reporting puts backlog at ~$100B as of mid-June (vs the $99.4B Q1 figure), with Cantor Fitzgerald modeling ~$131B by end-Q2 — analyst-estimated, not company-guided. The official next print is Q2 in early August. Operators should keep watching conversion velocity over the headline figure.
NVIDIA Q-over-Q data center revenue
$75.2B (Q1 FY27); Q2 guide $91B, reports Aug 26
$75.2B (Q1 FY27); Q2 guide $91B, reports Aug 26
→
Q2 FY27 guide $91B implies further +21% QoQ
What this measures
No within-window NVIDIA event; the next earnings print is Aug 26. The Q2 guide is $91B. In-window context (Supermicro Vera Rubin order availability, all three HBM makers volume-shipping HBM4 12-Hi) supports the ramp; packaging and power remain the binding constraints, not demand.
Open vs closed gap on coding (SWE-Bench / agentic)
Open closing fast: GLM-5.2 (MIT) reportedly beats GPT-5.5 on long-horizon coding at ~1/6 cost; AA rebased to v4.1
Closed lead widened on paper (Fable 5 80.3% vendor scaffold) vs open ~58.6%; contested
↓
Sustained open lead reshapes enterprise procurement
What this measures
The gap narrowed sharply in-window: GLM-5.2's MIT weights reportedly beat GPT-5.5 on several long-horizon coding benchmarks at ~1/6 the cost, and Artificial Analysis rebased its Intelligence Index to v4.1 (agentic), where open leaders sit ~44 and GLM-5.2 took the open lead. With Fable 5 suspended, the available closed leader is Opus 4.8 (AA v4.1 56). Architects should treat open self-host as a live procurement option, not a hedge.
Sovereign AI commitments (count / aggregate $)
~14 / ~$180B+ (flat; TensorX/Solstice up-to-$1B EU facility is capacity, not drawn)
~14 / ~$180B+ (NAVER+NVIDIA 1GW added)
→
—
What this measures
Analyst-curated count of sovereign/national AI-compute commitments. No major new drawn commitment in-window; the only new item is TensorX/Solstice's up-to-$1B EU GPU/data-center financing facility (capacity, not a drawn commitment). Operators with EU workloads should track the facility but price in multi-year build timelines.
PJM 2026/27 capacity auction price ($/MW-day)
$329.17; 2028/29 BRA results expected ~July 7 (pending)
$329.17 (2028/29 BRA clears Jun 30-Jul 7)
→
11x in 24 months — power is the new binding constraint
What this measures
The 2026/27 BRA cleared at the FERC cap ($329.17); the 2027/28 BRA cleared at $333.44. The 2028/29 auction is still pending, with results expected around July 7, 2026, under a collar floor of $175 and a cap near $325. Architects should not assume near-term price relief; budget capacity at-cap through 2028.
Time-to-power, busiest US markets (months)
60-84; FERC Jun 18 show-cause may compress large-load study timelines
60-84 (new PJM); 2027/28 auction cleared short of target
→
—
What this measures
Months from new-load interconnection request to energization. FERC's Jun 18 show-cause orders aim to speed large-load studies and clarify cost allocation within 60 days — potentially pro-speed medium-term, but no near-term change. Large-power-transformer lead times remain ~128 weeks. Architects should pre-commit power and long-lead grid equipment before GPU SKUs.
Cost-per-task, frontier reasoning model
GLM-5.2 API ~$1.40/$4.40 per MTok; Grok 4.3 on Bedrock $1.25/$2.50
~$0.10-$0.15 (effective); open weights (GLM-5.2) push commodity capability to ~1/6 frontier cost
~$0.10-$0.15 (effective); Fable 5 sets a 2x-Opus top tier ($10/$50 per MTok)
↓
—
What this measures
Median cost across frontier-tier reasoning models for a benchmark complex task. Open weights drove the ceiling down: GLM-5.2 lists at ~$1.40/$4.40 per MTok (~1/6 of comparable frontier) and Grok 4.3 went GA on Bedrock at $1.25/$2.50. Operators running agents at scale should re-benchmark on cost-per-task and pilot open self-host for routine work.
Custom silicon share of incremental AI compute
~33-36%; J.P. Morgan pegs 2026 custom-ASIC TAM ~$60-70B (Broadcom 80-85%)
~33-36%; d-Matrix Corsair inference ASIC in volume production
↑
>35% materially compresses merchant GPU pricing
What this measures
A J.P. Morgan note put the 2026 custom-ASIC market at ~$60-70B with Broadcom at 80-85% and Marvell 10-12%, reinforcing the thesis that ASIC units overtake merchant-GPU units by 2027. Investors with concentrated NVIDIA exposure should keep diversifying into the Broadcom/Marvell co-design duopoly and advanced packaging / power.

Frontier lab cash position (avg months runway, top 3)
~34-37 mo (flat; no new in-window round)→vs ~34-37 mo (xAI now public via SpaceX IPO)
Threshold: <18 mo triggers re-rating risk
What this measures
Top 3 frontier labs (OpenAI, Anthropic, Google DeepMind) by disclosed runway, with xAI now public inside SpaceX. No new financing event in-window; both OpenAI and Anthropic remain at the confidential-draft-S-1 stage. Capital access stays wide; boards should not assume funding pressure forces near-term commercial concessions.
Hyperscaler capex / AI revenue ratio (top 4 weighted)
~5.0-5.3 (Amazon $10B + Google $1.5B added; top-4 guides flat)→vs ~5.0-5.3 (Oracle FY27 ~$70B guide; top-4 flat)
Threshold: >6.0 invites investor pushback at next earnings
What this measures
Top 4 hyperscalers (MSFT, GOOG, META, AMZN) weighted aggregate of capex divided by AI-attributable revenue. No new top-4 earnings print in-window; the incremental Amazon and Google campus commitments fit prior ~$725-805B 2026 capex guides. Investors should keep the bubble hypothesis on funding/conversion and now on FERC-driven siting cost, not demand.
CoreWeave revenue backlog
~$100B (Jun 15); Cantor estimate ~$131B by end-Q2↑vs $99.4B (next print Q2, early Aug)
Threshold: Conversion velocity matters more than gross figure
What this measures
Booked but unrecognized revenue. Reporting puts backlog at ~$100B as of mid-June (vs the $99.4B Q1 figure), with Cantor Fitzgerald modeling ~$131B by end-Q2 — analyst-estimated, not company-guided. The official next print is Q2 in early August. Operators should keep watching conversion velocity over the headline figure.
NVIDIA Q-over-Q data center revenue
$75.2B (Q1 FY27); Q2 guide $91B, reports Aug 26→vs $75.2B (Q1 FY27); Q2 guide $91B, reports Aug 26
Threshold: Q2 FY27 guide $91B implies further +21% QoQ
What this measures
No within-window NVIDIA event; the next earnings print is Aug 26. The Q2 guide is $91B. In-window context (Supermicro Vera Rubin order availability, all three HBM makers volume-shipping HBM4 12-Hi) supports the ramp; packaging and power remain the binding constraints, not demand.
Open vs closed gap on coding (SWE-Bench / agentic)
Open closing fast: GLM-5.2 (MIT) reportedly beats GPT-5.5 on long-horizon coding at ~1/6 cost; AA rebased to v4.1↓vs Closed lead widened on paper (Fable 5 80.3% vendor scaffold) vs open ~58.6%; contested
Threshold: Sustained open lead reshapes enterprise procurement
What this measures
The gap narrowed sharply in-window: GLM-5.2's MIT weights reportedly beat GPT-5.5 on several long-horizon coding benchmarks at ~1/6 the cost, and Artificial Analysis rebased its Intelligence Index to v4.1 (agentic), where open leaders sit ~44 and GLM-5.2 took the open lead. With Fable 5 suspended, the available closed leader is Opus 4.8 (AA v4.1 56). Architects should treat open self-host as a live procurement option, not a hedge.
Sovereign AI commitments (count / aggregate $)
~14 / ~$180B+ (flat; TensorX/Solstice up-to-$1B EU facility is capacity, not drawn)→vs ~14 / ~$180B+ (NAVER+NVIDIA 1GW added)
What this measures
Analyst-curated count of sovereign/national AI-compute commitments. No major new drawn commitment in-window; the only new item is TensorX/Solstice's up-to-$1B EU GPU/data-center financing facility (capacity, not a drawn commitment). Operators with EU workloads should track the facility but price in multi-year build timelines.
PJM 2026/27 capacity auction price ($/MW-day)
$329.17; 2028/29 BRA results expected ~July 7 (pending)→vs $329.17 (2028/29 BRA clears Jun 30-Jul 7)
Threshold: 11x in 24 months — power is the new binding constraint
What this measures
The 2026/27 BRA cleared at the FERC cap ($329.17); the 2027/28 BRA cleared at $333.44. The 2028/29 auction is still pending, with results expected around July 7, 2026, under a collar floor of $175 and a cap near $325. Architects should not assume near-term price relief; budget capacity at-cap through 2028.
Time-to-power, busiest US markets (months)
60-84; FERC Jun 18 show-cause may compress large-load study timelines→vs 60-84 (new PJM); 2027/28 auction cleared short of target
What this measures
Months from new-load interconnection request to energization. FERC's Jun 18 show-cause orders aim to speed large-load studies and clarify cost allocation within 60 days — potentially pro-speed medium-term, but no near-term change. Large-power-transformer lead times remain ~128 weeks. Architects should pre-commit power and long-lead grid equipment before GPU SKUs.
Cost-per-task, frontier reasoning model
~$0.10-$0.15 (effective); open weights (GLM-5.2) push commodity capability to ~1/6 frontier cost↓vs ~$0.10-$0.15 (effective); Fable 5 sets a 2x-Opus top tier ($10/$50 per MTok)
GLM-5.2 API ~$1.40/$4.40 per MTok; Grok 4.3 on Bedrock $1.25/$2.50
What this measures
Median cost across frontier-tier reasoning models for a benchmark complex task. Open weights drove the ceiling down: GLM-5.2 lists at ~$1.40/$4.40 per MTok (~1/6 of comparable frontier) and Grok 4.3 went GA on Bedrock at $1.25/$2.50. Operators running agents at scale should re-benchmark on cost-per-task and pilot open self-host for routine work.
Custom silicon share of incremental AI compute
~33-36%; J.P. Morgan pegs 2026 custom-ASIC TAM ~$60-70B (Broadcom 80-85%)↑vs ~33-36%; d-Matrix Corsair inference ASIC in volume production
Threshold: >35% materially compresses merchant GPU pricing
What this measures
A J.P. Morgan note put the 2026 custom-ASIC market at ~$60-70B with Broadcom at 80-85% and Marvell 10-12%, reinforcing the thesis that ASIC units overtake merchant-GPU units by 2027. Investors with concentrated NVIDIA exposure should keep diversifying into the Broadcom/Marvell co-design duopoly and advanced packaging / power.

Predictions

What we expect next.

6 predictions for the next 30-90 days, confidence 55%-80%.

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

58%

confidence

Software

An MIT- or Apache-licensed open-weight model (e.g., GLM-5.2) enters the overall top 5 of the Artificial Analysis Intelligence Index v4.1 — not just the open-weight subset — by August 31, 2026.

Deadline: By August 31, 2026

Trigger: Artificial Analysis Intelligence Index v4.1 leaderboard update.

Prediction 02

75%

confidence

Hardware

By August 31, 2026, all three HBM makers (SK hynix, Samsung, Micron) confirm HBM fully allocated for 2026 and/or 2027 price increases.

Deadline: By August 31, 2026

Trigger: Earnings calls or supply-chain reporting (TrendForce, Bloomberg/Reuters) from the three memory vendors.

Prediction 03

55%

confidence

Networking

A second non-Broadcom vendor (Marvell or Credo) cites a 1.6T or co-packaged-optics production design win by September 30, 2026.

Deadline: By September 30, 2026

Trigger: Earnings call, product release, or customer design-win disclosure.

Prediction 04

80%

confidence

Power

At least one RTO/ISO files a large-load interconnection compliance proposal answering FERC's Jun 18 show-cause orders by the August 17, 2026 deadline.

Deadline: By August 17, 2026

Trigger: FERC docket filings from PJM, MISO, SPP, CAISO, ISO-NE, or NYISO.

Prediction 05

68%

confidence

Power

A named hyperscaler announces a >1GW behind-the-meter or off-grid generation deal for AI data centers by August 31, 2026.

Deadline: By August 31, 2026

Trigger: Hyperscaler energy/data-center announcement; utility or developer disclosure.

Prediction 06

55%

confidence

Capital

At least one major closed lab cuts flagship API prices or ships a cheaper tier by August 31, 2026, in response to open-weight cost pressure.

Deadline: By August 31, 2026

Trigger: Lab pricing page or API changelog (OpenAI, Anthropic, Google).

Track record

Scoring prior predictions.

6 prior predictions: 0 hit, 0 miss, 0 partial, 6 pending. Hit rate —.

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

Prediction 01

58%

confidence

Software

Gemini 3.5 Pro reaches public GA by June 30, 2026 but debuts below Claude Fable 5 on the Artificial Analysis Intelligence Index in its first independent pass.

Deadline: By June 30, 2026

Trigger: Google AI Studio / Vertex GA announcement plus Artificial Analysis leaderboard update.

pending— Tracking toward a miss on the GA leg: as of W25 close, Pro slipped from June to July, still a limited Vertex preview with no public benchmark. Deadline June 30 not yet passed but unlikely to be met.

Prediction 02

60%

confidence

Software

Anthropic restores public (non-government) access to a Fable-5-class model — with geo-gating or KYC controls — by July 31, 2026.

Deadline: By July 31, 2026

Trigger: Anthropic status page, model documentation, or press confirming restored availability.

pending— No restoration in-window — Fable 5 and Mythos 5 stayed government-suspended the full week, with Anthropic saying only that it is 'working to restore access.' Deadline July 31 open.

Prediction 03

65%

confidence

Hardware

At least one more merchant inference-ASIC vendor (Groq, Cerebras, Tenstorrent, or d-Matrix) announces a volume-production or named-hyperscaler milestone before August 31, 2026.

Deadline: By August 31, 2026

Trigger: Vendor press release, earnings disclosure, or customer design-win announcement.

pending— In-window the only item was a Qualcomm-Tenstorrent acquisition rumor (Jun 15). Strong out-of-window evidence followed Jun 22-24 (Cerebras' AWS/OpenAI confirmations, Groq's $650M raise), so this is tracking to hit. Deadline Aug 31 open.

Prediction 04

60%

confidence

Networking

A second of (Marvell, Credo) cites a 1.6T or co-packaged-optics production design win by August 31, 2026, joining Broadcom's Tomahawk 6.

Deadline: By August 31, 2026

Trigger: Earnings call, product release, or customer design-win disclosure.

pending— No new non-Broadcom design win in-window; Marvell's only output was a plasmonics research blog and Coherent's move was laser capacity, not a design win. Deadline Aug 31 open.

Prediction 05

62%

confidence

Capital

Neither OpenAI nor Anthropic converts its confidential draft to a publicly visible S-1 on SEC EDGAR before August 31, 2026.

Deadline: By August 31, 2026

Trigger: SEC EDGAR public filings; Reuters / Bloomberg / The Information confirmation of public S-1 vs confidential DRS.

pending— Holding and on track: both remain confidential drafts with no public EDGAR S-1 in-window; a public filing is expected only days before a roadshow (late Q3/Q4). Deadline Aug 31 open.

Prediction 06

60%

confidence

Power

A hyperscaler announces a >500MW power-first AI campus or behind-the-meter generation deal by September 30, 2026.

Deadline: By September 30, 2026

Trigger: Hyperscaler energy/data-center announcement; utility or developer disclosure.

pending— Strong momentum: in-window power-first activity was developer-led (Cummins-Circe 2GW BTM) with a reported-but-unconfirmed Meta-Crusoe ~1.6GW deal. A qualifying hyperscaler deal (Microsoft-Chevron Project Kilby, 2.67GW) landed Jun 22, just after this window. Deadline Sep 30 open; tracking to hit.

Watchlist

On the radar this week.

5 catalysts to watch, starting Jun 22-Jul 7.

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

Jun 22-Jul 7
PJM 2028/29 base residual capacity auction results
After two near-cap clears and a 2027/28 reliability shortfall, the 2028/29 results (collar near $325/MW-day) are the single most important power-cost signal of the quarter and will reprice US data-center economics.
By Aug 17
RTO/ISO responses to FERC's Section 206 show-cause orders
FERC gave six grid operators 60 days to fix large-load interconnection rules. The compliance filings will define how AI campuses connect and who pays — a board-level siting variable for every US deployment.
Jun 22-24 (W26 catalysts)
Project Kilby, Micron/SK hynix HBM4, Cerebras/Groq
A cluster of marquee items landed just after this window: Microsoft-Chevron's 2.67GW off-grid gas deal, Micron's record HBM-allocated quarter, SK hynix's HBM4-vs-DDR5 pivot, and Cerebras/Groq milestones. They will anchor next week's read.
July
Gemini 3.5 Pro GA and Fable 5 restoration
Both slipped: Pro moved to July and Fable 5 stayed suspended. Their resolution will decide whether the closed frontier re-takes the lead from the surging open-weight field.
Jun-Aug
Open-weight enterprise adoption (GLM-5.2)
Downloads, integrations, and independent SWE-Bench replication for GLM-5.2 and MiniMax-M3 will show whether MIT/permissive open weights become production substrate and force closed-flagship price cuts.

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.