Daily AI / Tech / VC Briefing: Sunday 25 October 2026

Good evening, here's today's briefing. It's a quiet Sunday, so there are six items. A top Wall Street economist says AI has paid nothing so far outside Big Tech. SpaceX is in the news twice, over US military tracking data and over Ukraine. The UK is putting OpenAI and Anthropic credits into a new biosecurity lab. Sweden's government is switching off ChatGPT and Claude. And the coding-agent race keeps getting more aggressive.
LEAD: Apollo's chief economist says AI has delivered "zero financial payoff" outside Big Tech, as AI credit stress grows
Background: AI stocks, especially chipmakers, fell this week as investors began to doubt how long the spending boom can last. A handful of hyperscalers are expected to spend about $800 billion on capex this year, and more than $1 trillion next year. What happened: Torsten Slok, chief economist at Apollo, ended a week of warnings with a Saturday note. He said AI has "delivered zero financial payoff" to US companies outside the big hyperscalers. Margins and revenue per worker are rising at the Magnificent Seven, but not across the rest of the S&P 500. Earlier in the week he called it "the core economic flaw in the AI bull case" that compute costs more to produce than it is worth to users. He also said 80% of the top 1% of AI spenders are VC-backed, meaning compute bills are being paid with venture money, not earnings. Credit markets are showing the same strain. On Thursday, Meta's five-year credit default swaps hit a record 114.79 basis points. CoreWeave's stock is down more than 30% this month, and its CDS have risen to 1,145 basis points. Why it matters: Lab valuations, neocloud debt and the data-centre buildout all assume that ordinary companies will pay for AI. If that demand is slow to arrive, the funding loop that many AI start-ups depend on gets tighter. This lands in the same weeks that Anthropic is lining up its IPO. Sources: Fortune · Bloomberg (paywalled)
2. SpaceX gets the US military's most precise space-tracking data, and Musk says SpaceX closed a Ukraine drone "loophole"
Background: Starlink has about 11,000 satellites in orbit, and the fleet could grow to around 20,000. The civil traffic-coordination system TraCSS costs about $65 million. It helps avoid up to 50,000 close approaches a day, and the administration tried to defund it in two budgets before Congress restored the money. What happened: The Guardian reports that SpaceX has gained enhanced access to US Space Command's High Accuracy Catalog. This includes detailed messages that track objects down to about 10cm. In return, the military gets access to SpaceX's own tracking system, Stargaze. The Space Force said it is "engaging with several companies" to swap commercial data for parts of the catalogue "where it provides benefit to the US government." One critic, Cole Donovan, said: "It's like giving a trucking company control of all the traffic lights." Separately, Musk posted on X on Saturday that Ukraine's military had "violated Starlink terms of service and used it for offensive drone strikes," and that "that loophole was closed last week." Ukrainian adviser Serhiy Beskrestnov called the post "very strange." The BBC notes that service had been restored after President Trump said he told Musk to "put it back." Why it matters: SpaceX is becoming part of the infrastructure that governments rely on, which brings more power and more political risk. This comes as SpaceX's AI business and a possible Tesla merger are under scrutiny. (This is an update to the Starlink and Ukraine story from 14 to 16 October.) Sources: The Guardian · BBC
3. UK sets up a £300m AI Biosecurity Lab with OpenAI and Anthropic, plus £120m more for the AI Security Institute
Background: The UK is announcing AI and security spending ahead of Wednesday's Budget. This week AI-discovered "php2shell" flaws in PHP pushed some government sites offline as a precaution. What happened: The Cabinet Office announced the AI Biosecurity Lab, an independent non-profit it calls the world's first dedicated AI biosecurity organisation. It is backed by £100 million of government money and up to £200 million in commitments from companies. OpenAI will provide £100 million in compute and model credits, plus access to its models through its Rosalind Biodefense programme. Anthropic will provide up to £100 million in credits, model access and technical support. Google DeepMind and Basecamp Research will work with the lab as research partners. The BBC adds an extra £120 million for the AI Security Institute, £60 million for an agentic-AI "National Cyber Shield", and £70 million for a vulnerability-detection centre called NERV. Why it matters: Both leading labs are paying for government safety work in kind, with credits rather than cash. That builds goodwill with regulators while placing their models inside national infrastructure, a pattern to watch for Cursor and other model buyers in the public sector. Sources: GOV.UK · BBC
4. Sweden's government blocks ChatGPT and Claude from Monday and directs staff to Copilot
Background: European public bodies are tightening control over consumer AI tools. The European Parliament disabled AI features on work devices in February. What happened: From Monday 26 October, the browser versions of ChatGPT and Claude will be blocked on computers at Sweden's Government Offices, which include the Prime Minister's Office and the ministries. Staff are directed to Microsoft Copilot, which the office reviewed and then locked down further. The office's digitalisation director cited information security and "several examples of problematic use." Staff can still reach OpenAI and Anthropic models through Intric, an internal platform. The Swedish Tax Agency and one other agency already have similar blocks. Why it matters: In government, contracts and admin controls matter more than which model is best. That favours Microsoft's distribution. Direct-to-user AI products will need enterprise agreements to win the public sector. Sources: Notebookcheck · SVT
5. Coding-agent price war: OpenAI's discounts pull customers from Anthropic, and Codex resets usage limits again
Background: Claude Code has led enterprise coding, but OpenAI is pushing Codex hard. (This is also covered in today's agents briefing.) What happened: The Information reports that OpenAI is using large discounts to win Anthropic's coding customers. It says OpenAI's share of Rippling's AI spending rose from 16% to 52%, while Anthropic's fell from 57% to 18%. Replit has signed a deal worth more than $100 million a year at more than 30% off, and DataCamp got 40% off in a deal tied to Codex. Codex lead Tibo Sottiaux also posted "Day 20/ Rest and reset" early on Sunday, another usage-limit reset during OpenAI's month of daily Codex improvements. Smaller moves this week: GitHub Copilot now works inside Microsoft Teams, and the legacy Teams app is being retired on 16 November. Why it matters: Coding agents are now competing on price as well as quality. That squeezes Cursor, Replit and others that resell frontier models, and it puts pressure on Anthropic's most important revenue line just before its IPO. Sources: The Decoder (summarising The Information) · The Information · Tibo Sottiaux on X · GitHub changelog
6. OpenAI recorder case: the FBI weighs an economic espionage charge
Background: Former OpenAI engineer Abdulrahman Alfozan, a Saudi national, is charged with planting a hidden recorder in a meeting room used by OpenAI executives. We covered his arrest and detention earlier this month. What happened: The Wall Street Journal reports that the FBI is looking at whether Alfozan planned to share OpenAI information with Saudi officials, which could lead to an extra economic espionage charge. No such charge has been filed, and no completed transfer of information has been reported. OpenAI has said it found no evidence that he accessed model weights, source code or customer data. Why it matters: Frontier labs are now treated as national-security targets. Expect tighter insider controls and vetting across AI companies, including at coding-agent start-ups that hold valuable code and data. Sources: WSJ · EconoTimes · US DOJ
Key hires
- Claude Code departures: Thariq Shihipar, a well-known voice of the Claude Code team, said on Saturday he had left Anthropic after more than a year on the product. Anthony Morris, who created Claude Code on desktop, said Friday was his last day. Both left on good terms. (This is also in today's agents briefing.) Shihipar on X · Morris on X
Coming up
- Monday 26 October: Sweden's Government Offices block on ChatGPT and Claude takes effect.
- Late October: Anthropic's IPO prospectus could become public. Reuters says marketing to investors could start in the week of 9 November, but the dates could change. Nothing was on the SEC's EDGAR system as of today. IFR
- Wednesday 28 October: UK Budget, with the AI and cyber spending above.