Daily AI / Tech / VC Briefing: Tuesday 3 November 2026

Good afternoon, here's today's briefing. The AI memory crunch reached Apple's results, Nvidia is reported to be capping how much of its partners' compute goes to Chinese customers, and Anthropic's IPO investors are pushing for a lower price. It is also US midterm election day, the first big test of AI as a political issue.
LEAD: Apple beats on sales but warns memory costs will squeeze margins
Background: AI server builders such as Nvidia and the hyperscalers are buying up memory chips at premium prices, leaving less supply for phone and PC makers. John Ternus took over from Tim Cook as Apple CEO on 1 September, and Monday's results were his first.
What happened: Apple's September-quarter revenue rose 12% to $114.48 billion, ahead of estimates, and iPhone revenue rose 17% to a record $57.58 billion. But Apple guided December-quarter gross margin to 45–46%, below the 46.3% analysts expected, blaming mainly higher memory costs plus spending on its Siri AI. Ternus called memory pricing "unprecedented" and said "looking beyond the December quarter, we see the pricing for memory purchases continuing to increase." Apple expects iPhone and Mac supply limits over the holidays, and COO Sabih Khan said purchase obligations rose by $33 billion in a quarter as Apple locks in supply. Shares fell about 3–4% on Tuesday.
Why it matters: Even the world's strongest hardware company cannot fully shield itself from the AI boom's demand for memory. Expect higher device prices across the industry, though analysts think Apple may win market share from rivals with thinner margins.
Sources: CNBC · Reuters via Investing.com · CNBC live updates · Fortune
Nvidia reportedly tells cloud partners to sell most compute outside China
Background: US export controls bar Nvidia's top chips from China, so Chinese firms have rented AI computing from data centres abroad instead.
What happened: The Financial Times reported, citing five people, that Nvidia has told cloud and data-centre customers worldwide that at least 51% of their services must be sold to non-Chinese companies. The rule was delivered verbally by sales staff in recent months and applies to future contracts, not existing ones. The FT said a rack of Nvidia B300 chips selling for about $600,000 internationally traded in China in October for about $2.8 million. Reuters said it could not immediately verify the report, and Nvidia neither confirmed nor denied it to AFP.
Why it matters: If accurate, this closes a major workaround Chinese AI labs have used, and some overseas cloud providers say finding enough non-Chinese customers will be hard.
Sources: Reuters via MarketScreener · Silicon Report · The Standard
Anthropic: IPO investors push for a lower valuation
Background: Yesterday's briefing covered Anthropic's amended filing with third-quarter results. What's new today is how investors are reacting to the price.
What happened: The Wall Street Journal reported that institutional investors have told Anthropic and its advisers to price the IPO at $1.5 trillion to $1.6 trillion, 20–25% below the roughly $2 trillion talked about in recent months, citing market volatility, AI overinvestment worries and political pushback against data centres. The roadshow could start next week, with a listing targeted for the week before Thanksgiving. Fortune's analysis found compute and infrastructure costs rose about 77% in the quarter to $9.9 billion, against revenue growth of nearly 47%, and adjusted operating margin slipped from 19% to 15%. University of Florida professor Jay Ritter said only 14 of 4,110 IPOs from 1980 to 2024 priced above $40 per dollar of past-year revenue, the level Anthropic would hit above about $1.41 trillion.
Why it matters: The biggest tech IPO ever is now a test of how much public investors will pay for AI growth when costs are rising faster than sales.
Sources: WSJ · Fortune · Chosun Biz
SpaceX/xAI: Grok Bot tests extra usage in exchange for training data
Background: Grok Bot users ran into weekly usage caps last week, and Elon Musk said serving costs were too high relative to revenue (covered yesterday). SpaceX has said paying weekly users reached 800,000 in early October, and its Super Intelligence segment booked $7.6 billion of third-quarter revenue, including compute rented to outside customers.
What happened: A prerelease Grok Bot build spotted by leaker blankspeaker shows a prompt, shown when a user hits 100% of their allowance, offering to reset the limit if they allow their interactions to be used to train SpaceX's AI models. The leaker says it is in testing and may not ship; SpaceX has not commented.
Why it matters: It would turn user data into a currency for usage, a new way to manage heavy agent costs, but it raises privacy questions for a product that can read email, calendars and connected apps.
Sources: Tech City Authority · Quartz
AI on the ballot: the midterms' first test of tech politics
Background: AI regulation has become a political issue, with rival industry-funded super PACs on both sides.
What happened: Axios reports that AI groups Public First and Leading the Future spent more than $50 million through affiliated super PACs in nearly 50 races, though their ads largely avoided AI itself. Elon Musk was by far the biggest midterm donor, at more than $400 million. Races being watched include Michigan's Senate contest, where Democrat Abdul El-Sayed wants a new federal AI agency and a ban on superintelligence, and California's 11th District, where Scott Wiener's AI transparency law split the industry.
Why it matters: The results will shape whether candidates campaign on AI rules in 2028 and how strong the pro-safety bloc is in the next Congress.
Palantir's business customers overtake government
Background: Palantir has long been seen as a government contractor.
What happened: Third-quarter revenue rose 96% to $2.31 billion, against $2.18 billion expected. US commercial revenue jumped 148% to $984 million, passing US government revenue of $953 million for the first time. Palantir now expects about $8.53 billion of revenue for 2026. Shares still dipped about 3% after hours.
Why it matters: It is one of the clearest signs yet that companies are paying real money for AI software, not just experimenting.
Competitor moves
Background: Coding-agent makers are competing on cost controls and guardrails as well as model quality.
What happened: Anthropic added two sentences to Claude Code's prompt on Opus 5.5 to stop it recreating "someone else's specific game", saying IP infringement breaks its usage policy; original and genre games are meant to keep working. Cognition launched hierarchical budgets for Devin Enterprise, letting admins split AI spend across departments, projects and users up to three levels deep, with temporary top-ups. SpaceX's Grok Bot usage-reset test (above) is another attempt to manage agent costs.
Why it matters: As agents run longer and cost more, buyers want spending controls and legal safety. Expect Cursor, Codex and Copilot to face the same pressure.
Sources: The Verge · Ars Technica · Devin blog · The New Stack
VC and deals
What happened: Salesforce agreed to buy Cresta, which makes AI voice agents for customer service, in a deal reportedly worth about $2.7 billion, expected to close in its first quarter of fiscal 2028. CrowdStrike agreed to buy AI agent monitoring firm Fiddler and AI security tester Dreadnode, reportedly for about $400 million and $95 million (WSJ). Drax, Brookfield and Brookfield's Radiant plan an AI data centre at Drax Power Station in North Yorkshire: up to 100MW from early 2028, more than 1GW eventually, with total investment that could exceed £20 billion; Drax shares rose over 8%. In venture, Trajectory raised $45 million led by Sequoia at a $300 million valuation for continual post-training of models; Tutor Intelligence raised a $75 million Series B led by Andreessen Horowitz to rent out robot workers by the hour; and UK-German defence start-up Hypersonica raised €100 million from Lightspeed and General Catalyst.
Why it matters: Big incumbents are buying AI agent capabilities rather than building them, and the UK is finally getting a gigawatt-scale AI data centre plan, though it still needs planning approval and a tenant.
Sources: Salesforce · Silicon Republic · SecurityWeek · CNBC · TechCrunch · Business Wire · Sifted
Hires
OpenAI hired former US Army Secretary Dan Driscoll as senior counselor for national security modernisation, a new role advising leadership on AI for national security and cyber security, with a focus on helping US allies adopt AI. Driscoll stepped down from the Army earlier this year after reported tensions with Defence Secretary Pete Hegseth. No notable hires at the coding-agent companies surfaced in the last day.