AI Forecast Tracker
Week 26 · June 22 – June 28, 2026
high signal

The Bet Comes Due

The bet that the AI boom bursts by 2026 comes due on Tuesday, and the market spent its final week trying to make it true. A semiconductor selloff wiped more than a trillion dollars in three days — Korea's exchange tripped its circuit breaker twice, Micron fell 13 percent intraday — and a cluster of Chinese hedge funds that called the 2007 top declared this one a "super bubble" ready to break. Then it all reversed: Micron's revenue quadrupled with a hundred billion dollars of contracts already signed, data-center spending stayed north of a trillion, and the two labs everyone watches both have IPO paperwork on file. None of the five things that would count as a burst happened, so the prediction resolves refuted — on the same week the most capable coding model on Earth entered its fifteenth day switched off by the US government, restored to exactly a hundred hand-picked organizations and no one else. The frontier is booming and a government holds the switch; both are true at once.

4 predictions updated3 milestones25 companies refreshed

Key Developments

1

The bet that the AI boom bursts by 2026 comes due Tuesday — the market threw its hardest punch of the year and still could not land it

P-012 — the prediction that the AI investment boom bursts by 2026 — resolves June 30, and its final full week tried hard to make it true. A semiconductor selloff wiped roughly $1.3 trillion in three days: Korea's exchange tripped its circuit breaker twice on June 23, Micron fell 13 percent intraday, the Nasdaq dropped 2.21 percent, and a cluster of Chinese hedge funds — one founded by the man who called the 2007 top — declared a "super bubble." Then it reversed inside a week: Micron's revenue quadrupled to $41.46 billion with roughly $100 billion of contracts already signed (the stock closed up 15 percent), Dell'Oro reaffirmed data-center spending above a trillion dollars for 2026, and OpenAI and Anthropic both have IPO paperwork on file. None of the five things that would count as a burst — a top-ten collapse, a 25 percent capex cut, venture funding down by half, a sustained 40 percent index drawdown, a mass enterprise retreat — happened, so the prediction resolves refuted.

Challenges P-012 AI investment boom is a bubble likely to burst by ...Confirms P-017 Annual data center CapEx will grow from $500B (202...Challenges
Counterpoint

This was the closest the market came to two of those criteria all year, and the fragility under it is real: 63 percent of Q1's $300 billion in venture funding went to four companies, Broadcom already missed in early June, and the data-center debt stack — GPUs that depreciate in a year financed by loans that run seven to fifteen — is the channel a real correction would travel. But that is a 2027-2028 story; the June-2026 burst this bet asked about did not arrive.

Source →
2

Two weeks after the government switched off the best coding model on Earth, it is still off — for everyone but a hundred hand-picked organizations

As of June 27, Day 15 of the export-control suspension, Claude Fable 5 remains fully offline for every commercial user. On June 26 a letter from Commerce Secretary Lutnick restored the companion Mythos 5 to roughly 100 vetted US critical-infrastructure organizations — a whitelist, not a reopening — while bipartisan House members demanded written justification for the takedown. Prediction markets moved up to 68-71 percent odds of full Fable restoration before July 1, and Anthropic's new July 8 ID-verification policy looks like the mechanism for a US-citizen-gated return. The most capable model anyone could rent is now something the government decides who gets back first.

Challenges Confirms P-034 AI capabilities are advancing at a rate that outst...Challenges P-001 AI models will handle most aspects of software eng...
Counterpoint

This is still an access event, not a capability regression — Fable's 95 percent SWE-bench Verified score is intact, just unreachable, and rising restoration odds plus the partial Mythos reopening say the market reads the outage as temporary. The durable change is not lost capability; it is that frontier availability is now a switch in Washington, and a whitelist restoration means the government is deciding not just whether the frontier ships but who stands at the front of the line.

Source →
3

With Anthropic's frontier model dark, OpenAI walked into the opening — a three-tier GPT-5.6 and a 1.5-million-token memory

OpenAI previewed the GPT-5.6 "Sol" family on June 26: Sol as the flagship, Terra at roughly half the price of GPT-5.5, and Luna as the fast lane, all sharing a 1.5-million-token context window — enough to hold a small codebase and its full history in a single prompt. OpenAI is pitching it as much as an alignment fix as a capability jump, claiming gains on cybersecurity and biology workflows. The timing is the real signal: it ships while Fable 5 is suspended and Google's Gemini 3.5 Pro has slipped to July for the second time, leaving OpenAI briefly alone at the frontier customers can actually reach.

Confirms P-001 AI models will handle most aspects of software eng...Confirms P-018 AI inference costs have fallen 99%+ in a single ye...
Counterpoint

Every number here is OpenAI's own — Benchmark Theater, and the clearest case of it this week. GeneBench is a brand-new evaluation OpenAI authored, the cybersecurity and biology claims have no independent replication yet, and a limited preview is not general availability. Treat the capability story as directional until a third party scores it; the load-bearing fact is the timing, not the benchmarks.

Source →

What the Evidence Moved

P-012AI investment boom is a bubble likely to burst by 2026

Final weekly check, three days to the June 30 resolution. The closest-to-burst week of the cycle — a $1.3T chip selloff (KOSPI circuit breaker twice, Micron -13% intraday) and Chinese hedge funds calling a "super bubble" — still cleared none of the five operational criteria, and the market reversed (Micron +15% on a $41.46B / $100B-contract Q3, >$1T 2026 capex reaffirmed, OpenAI and Anthropic S-1s on file). Resolves REFUTED at month-end.

5%5% 0pp
P-020AI-exposed industries will see job losses of ~20,000 per mon...

Oracle named AI as a layoff cause in a binding SEC 10-K (June 22, ~21,000 jobs) and 2026 tech cuts reached ~185,900 YTD with ~56% citing AI, while Challenger's May report logged 97,006 cuts with AI leading reasons a third straight month — the US AI-exposed run-rate clears the 20K/month bar. Counter: NBER's 2026 review finds near-zero aggregate effect at AI-adopting firms.

74%75% +1pp
P-002AI will disrupt 50% of entry-level white-collar jobs over 1-...

Oracle's binding 10-K AI-causal layoff language is a new grade of disclosure evidence, and a Q2 survey put 41% of firms freezing roles tied to agentic rollouts. Counter strengthened the same week — NBER near-zero aggregate effect and the Wharton / Oxford Economics / Revelio redundancy-washing discount — so a small move up.

73%74% +1pp
P-017Annual data center CapEx will grow from $500B (2025) to $1.4...

Dell'Oro reaffirmed >$1T 2026 data-center capex with 2H acceleration and Micron's Q3 added ~$100B of take-or-pay AI-memory contracts through 2030 — the supplier side is contracting demand at a level that implies a 2030 overshoot of ARK's $1.4T. The June chip selloff threatened but no capex was cut.

60%62% +2pp

Company Impact

Micron Technology

Score change

Q3 FY2026 (June 24) revenue quadrupled YoY to $41.46B with data-center revenue above $25B/quarter, and 16 take-or-pay Strategic Customer Agreements lock ~$100B of contracted revenue through 2030 — the upfront-deposit backlog converts a cyclical memory commodity toward contracted-revenue visibility. md 7->8. Score 7.3->7.55 (stays positive).

Micron Q3 FY2026 earnings and selloff recovery — CNBC · Micron Q3 $100B take-or-pay contracts — TechTimes

7.5
See assessment →

JPMorgan Chase

Score change

June 9 rollout of long-running autonomous AI agents across 450+ production use cases, plus the CIO's May 19 disclosure of up to 30% productivity gains and a 20% private-banking sales lift, moves JPM past the pilot phase on quantified, multi-source evidence. are 6->7. Score 7.5->7.7 (stays positive).

JPMorgan to deploy autonomous AI agents — CNBC

7.7
See assessment →

Mastercard

Score change

Agent Pay for Machines launched June 10 with 30+ partners (Stripe, Adyen, Coinbase, Cloudflare), positioning the network as the settlement rail for agent-to-agent commerce — a category-defining extension of Agent Pay. aam 9->10. Score 8.0->8.05 (stays very_positive).

Mastercard launches Agent Pay for Machines — Mastercard

8.1
See assessment →

Oracle

Data refresh

The FY2026 10-K (June 22) disclosed a 13% workforce cut — about 21,000 jobs — with SEC-binding language attributing the reductions to AI deployment, the first major-tech AI-causal layoff attribution in a federal filing, alongside $55.7B FY26 capex. Score holds 7.8; the RPO surge was already priced at W24.

Oracle cuts 21,000 jobs citing AI in 10-K — Tom's Guide

7.8
See assessment →

Anthropic

Data refresh

Day 15 of the suspension — Fable 5 still fully offline, Mythos 5 partially restored to ~100 US critical-infrastructure orgs (June 26), restoration odds up to 68-71% before July 1. The dr move (2->3) was already taken at W25; held this week as the partial restoration cuts against a further downgrade. Score holds 8.85.

Access suspension of Fable 5 and Mythos 5 — Anthropic · Mythos 5 restored to 100+ US organizations — TechCrunch

8.8
See assessment →

Salesforce

Data refresh

Agentforce Help Agent launched June 25 at $2 per resolution (4.3M inquiries handled on Salesforce's own help desk at 70% autonomous resolution) — a notable outcome-pricing pivot, but the proof point is a dogfooded help desk, not customer revenue, so the score holds 7.0 pending Q2 FY27.

Salesforce launches Agentforce Help Agent at $2/resolution — SiliconAngle

7.0
See assessment →

Sources

Next issue drops Monday

Subscribe to get the briefing before the market opens.