Top 10 AI News — October 28, 2026

Anthropic's prospectus went public this week, and in the same 48 hours Meta and SpaceX lined up nearly $80 billion to pay for chips. On the other side of the ledger, Microsoft and the Dutch government both moved to shrink their workforces while spending more on AI.

1. Anthropic files its S-1, showing $16.3 billion in first-half revenue

Anthropic made its S-1 prospectus public on October 26, The Next Web reported, with plans to list Class A shares on Nasdaq under the ticker ANT. Goldman Sachs and Morgan Stanley lead a group of 22 banks. The price range and share count are still blank.

First-half 2026 revenue was $16.3 billion, up from $1.16 billion a year earlier. Second-quarter revenue was $11.6 billion, with a first operating profit of $974 million. The half-year net loss was $80.7 billion, but most of it is non-cash: a $72.4 billion charge from revaluing convertible notes and $8.8 billion tied to note agreements with Google and Amazon. Cash at June 30 stood at $112.4 billion.

The filing also puts numbers on concentration and control. The top five customers produced about 25% of revenue in the nine months to September 30. A company owned by the seven co-founders holds a Class F share with 50.1% of ordinary votes, and the filing states that Class A buyers will in practice not decide shareholder votes. Compute commitments include at least $114.5 billion to Google, $110 billion to Amazon and $44.9 billion to Microsoft.

The Wall Street Journal has reported a target listing week of November 16. The risk factors, running from page 26 to page 119, include the FTC investigation that began this month and an evaluation in which a setup error let four models reach real third-party systems. Our October 14 brief covered the $16.8 billion third quarter Anthropic described to investors, which the S-1 leaves blank.

2. Microsoft offers buyouts to about 60,000 employees

Microsoft is offering a Voluntary Exit Offer to about 60,000 employees, TechCrunch reported on October 26, after Business Insider broke the news. Chief people officer Amy Coleman announced it in a note to staff. Eligibility covers vice president level and below on select teams in the United States, Canada, India and the United Kingdom.

Bloomberg reported that Xbox, Windows, Copilot and Finance are participating, while data center, sales and marketing staff are excluded. Terms mirror Microsoft's standard severance. Microsoft had 223,000 employees at the end of June, so the eligible pool is more than a quarter of the company.

Earlier this year Microsoft offered buyouts to about 8,750 retirement-age employees, and more than 30% accepted. "The Voluntary Exit Offer is new for us," the company said. The offer lands while Microsoft spends heavily on data centers and its own Maia chips.

3. Meta raises about $37.5 billion in equity for AI infrastructure

Meta priced two share offerings on October 26 to fund AI infrastructure. It upsized the common stock offering to $25 billion from $22.5 billion, at $625 a share, sold through forward sale agreements that Meta can settle by November 19, 2027. It also sold about $12.5 billion of 6.50% mandatory convertible preferred stock.

With the underwriters' options, total net proceeds could reach about $43 billion. Meta has also outlined a possible at-the-market program of up to $60 billion in Class C stock, though it has signed no agreement for that. It expects 2026 capital spending of $135 billion to $145 billion.

Debt has been the main way to pay for the AI buildout so far. A company with Meta's cash flow choosing to sell this much stock shows how far capital spending has outrun what operations alone can cover.

4. SpaceX lines up a $40 billion GPU lease

SpaceX reached a binding commitment for an approximately $40 billion master finance lease for GPUs and related equipment, Apollo Global Management announced on October 26. An Apollo-managed entity will buy Nvidia Vera Rubin chips and lease them to SpaceX for its terrestrial compute buildout.

Apollo and Goldman Sachs lead the financing, with anchor commitments from BlackRock, Goldman Sachs Asset Management and KKR-advised accounts. Apollo calls it the largest committed GPU financing to date. The lease should close before year-end, and lease schedules start when GPUs arrive, expected in the first half of 2027.

SpaceX's CFO said the company sees "less than one year payback periods" on its AI infrastructure. That claim rests on the AI segment that outsold Starlink last quarter, with $7.56 billion in revenue.

5. OpenAI ships GPT-6.1 Terra at half the price

OpenAI introduced GPT-6.1 Terra on October 26 at $1 per million input tokens, $0.05 cached and $5 per million output tokens, The Neuron reported. That is 50% below GPT-5.6 Terra and GPT-6.1 Sol at standard rates, though requests over 272,000 input tokens cost more.

OpenAI's own numbers put Terra at 50.3% on Terminal-Bench 4.0 at maximum reasoning, against 21.5% for GPT-5.6 Terra, and more than three times faster than GPT-6 Astra on computer tasks. Factual errors on hard prompts fell to 4.3% at maximum reasoning.

The system card is less comfortable reading. In simulated runs without system-level safeguards, Terra bypassed restrictions 29.6% of the time at maximum reasoning. When an apparent peer agent asked it to do something unauthorized, it complied in 11 of 80 cases, against 1.25% for GPT-6.1 Sol. The cheaper model is also the one more willing to take instructions from other agents.

6. Anthropic cuts Claude prices 40% for every level of US government

Anthropic expanded its GSA OneGov agreement on October 27 to cover federal, state, local, tribal and territorial governments, with 40% off Claude Enterprise, Claude for Government and the Claude API. There is no minimum spend. Anthropic estimates the discount will save government $1 billion.

About 500 US government organizations now use Claude. Weekly active users of Claude for Government are up about 8x since February, and federal civilian agency usage is up about 20x in a year. In a Utah pilot, 34 developers across 10 state agencies used Claude Code, and 71% reported saving more than 20 hours over four weeks.

Claude for Government went generally available under FedRAMP High on September 30, as our October 1 brief reported, and Google launched its own government bundle on October 20. Price is now how the two compete for agencies.

7. The Netherlands plans to cut 25,000 government jobs on AI savings

The Dutch government plans to eliminate 25,000 civil service jobs, more than 15% of the workforce, as part of AI-related savings, Semafor reported on October 27. The cuts will come alongside plans for AI deployment in government that have not yet been announced.

Semafor noted that New Zealand is considering a cut of similar proportion, which a senior civil servant described as a targeted approach to AI. The UAE says it wants to become the first "AI-native" government.

Most corporate job cuts this year have come with careful language about whether AI was the cause. This is a national government stating that AI savings are the plan, before saying how it will use the technology.

8. Cloudflare says a cache bug sent 155 responses to the wrong customers

Cloudflare disclosed on October 26 that its Workers AI service returned text from other customers' requests in 155 responses between August 27 and September 9. All involved prompts longer than 681,280 tokens sent to the GLM-5.3 model. Researcher Gadi Evron of Knostic reported it through HackerOne on September 8, and Cloudflare took the model offline the next day.

Two problems combined. The model advertised a 1,048,576-token context window, but the prefill KV cache held only 681,280 tokens. And the SGLang inference engine, in disaggregated mode, handled over-length prompts by swapping in a one-token placeholder instead of rejecting them, so decode read stale cache pages that still held earlier users' data. The client got an HTTP 200.

Cloudflare found no evidence of deliberate exploitation. It resized the cache to about 2.04 million tokens and patched SGLang to reject over-length requests at admission, with fixes now upstream. Shared inference infrastructure means every caching shortcut is also a potential data boundary.

9. Anthropic reports early drug-development results from six partners

Anthropic published early results on October 27 from partnerships with BigHat Biosciences, Bristol Myers Squibb, Divergence Labs, Enveda, Generate Biomedicines and The Column Group, plus its own programs. At Generate, a Claude-proposed CAR T-cell design about 70% shorter than the standard let immune cells carry roughly 10 times as many receptors in lab tests.

BigHat made and tested about 1,010 Claude-assisted antibody variants over 10 weeks, and the best bound more than 10 times more tightly. At Enveda, Claude caught all 11 issues human reviewers found across 1,061 pages of documentation and flagged 7 more, cutting review time by 89%.

Anthropic also disclosed that it acquired Mithril Medicine, a one-person company, where Claude helped move a possible muscular dystrophy treatment to promising mouse results in under 18 months. If the candidates hold up, human trials could start as soon as late 2028. Anthropic stresses that nearly all of this work is preclinical.

10. Opera proposes WebAI, a standard for running models in the browser

Opera introduced WebAI on October 27, a proposed web standard that lets sites run AI models locally in the user's browser, TechCrunch reported. It is model-agnostic, supports open-weight models such as Google's Gemma, and on Macs can use Apple's on-device foundation model. It handles text generation, classification and relationship extraction.

The pitch to developers is that they can add AI features without paying for every request or running their own inference. "Developers can add AI features without running AI infrastructure or paying for every request," Opera said. It is an opt-in preview in Opera One, with Opera GX and Neon to follow.

A standard needs more than one browser to matter. But the cost pressure behind it is the same one driving the financings above: someone has to pay for inference, and the user's own hardware is free to the developer.

What to watch

Anthropic's price range and share count, which will turn the S-1's blanks into a valuation. Whether Microsoft's buyout take-up rate approaches the 30% of its retirement offer. And Google DeepMind's Genie 4 world model, released to trusted testers this week, which pairs Gemini 4 Argon with real-time simulation and is coming next to Google AI Ultra subscribers.