Goldman Sachs analysts, whose bank is leading Anthropic’s I.P.O., have told investors privately that they expect the company to burn through more than $65 billion next year, according to the Times, even though Anthropic’s own prospectus shows it generated cash in the first nine months of 2026 and has barely touched the $141 billion it raised privately. The same week, Apple opened a study for up to 200,000 people to see how they use AI for health questions, and OpenAI gave its shopping agents a wallet. The thread running through today’s stories is who pays for the AI build-out, and what ordinary people get to check before they trust it.
Today’s Headlines
The Money
- Anthropic’s Spending. The Times says Goldman Sachs analysts forecast more than $65 billion of burn next year and another $41 billion in 2028, citing two people familiar with the private discussions. The listing on Nasdaq is expected next Thursday and could value Anthropic at $1.4 trillion while raising an additional $100 billion. The prospectus also anticipates about $40 billion of expenses in the coming years tied to employee stock grants that are not yet on its books. Chief financial officer Krishna Rao’s only comment in the roadshow video was “we are in aggressive investment mode.” Dario Amodei’s pitch in the same video was that the company could help end cancer and mental illness within a decade.
- The Bank of Nvidia. TechCrunch’s Equity hosts say Nvidia has roughly 80 percent of the AI GPU market and cite $99 billion in investments, $25 billion in further commitments and $36 billion in cloud service commitments, including stakes in CoreWeave, Nebius, OpenAI and Anthropic. Anthony Ha points out that Huang is the biggest shareholder, has family at the company, and that there was no succession plan. Tim De Chant says the turning point looked like the November 9 Bloomberg report that Nvidia’s head attorney had a role in inquiries about the Commerce Department’s investigation into Alice Huang. These are the hosts’ readings, not findings, and a lot is still unresolved.
AI Gets Practical
- Apple’s Health Study. Apple is working with OpenEvidence for more than 12 months with up to 200,000 U.S. participants over 21 on an iPhone 15 or later running iOS 26. Answers are grounded in the New England Journal of Medicine, JAMA and Nature, and participants answer monthly surveys. Apple says the data is anonymized, will not train AI models, and is deleted by OpenEvidence after 24 hours. TechCrunch notes it is unclear whether this leads to AI answers in the Health app or Siri, and that OpenAI said earlier this year users ask more than 300 million health questions a week.
- ChatGPT Wallet. OpenAI says each agent purchase requires a Face ID or fingerprint check, and the payment network issues a substitute token so the agent never sees the card. Amex and Mastercard are named as partners in OpenAI’s announcement. It is limited to U.S. Pro subscribers 18 and older, virtual cards are unsupported, and OpenAI’s merchant playbook warns that some sites may block agent purchases and that fraud systems may decline legitimate ones. I read this through a search summary of OpenAI’s pages, and the sources disagree on whether the date is November 12 or 13.
- Grok Bot and Datasette Agent. xAI’s page describes bots that run on their own cloud machine with browser and terminal access, can work in groups, and are included in SuperGrok and Cursor plans. All of that is xAI’s own description. Simon Willison’s datasette-agent 0.6a0 is a smaller step: it puts an agent prompt on the Datasette homepage and supports hash-mentions of tables and queries.
People and Machines
- AI DJs. WIRED’s reporter says Spotify’s DJ arrived in 2023 and was expanded this month with more personalities and controls, while YouTube Music’s Beyond the Beat interjects for about 30 seconds between songs. The breaking point was a paragraph of what he calls platitudes that mispronounced Charli XCX’s name. Both features are optional, so his fix was simply to switch it off.
- OATs. Daehwa Kim of Carnegie Mellon’s Future Interfaces Group presented Object Agent Twins at the UIST conference in Detroit and won two Honorable Mentions. Each is an E Ink display with sensors for moisture, temperature, light and motion, with sprites made using Google’s Nano Banana. WIRED notes the main module costs around $20 and that companies increasingly lean on “cuteness” for AI hardware, citing Meta’s Muse Charm, expected in December.
Science and Chips
- OpenAI Goes to Science. OpenAI’s new science blog points to a September proof about Navier–Stokes and to GPT-Rosalind agents analyzing data from millions of mouse brain cells in 32 minutes. A guest post by Nathaniel Craig describes a candidate lattice definition of the Standard Model built with GPT-6 Astra and says it relies on OpenAI’s recent Yang-Mills proof. I could not verify any of those claims independently, and they come from OpenAI’s own pages.
- Bharat Semi. The New Delhi startup raised $24 million, is pre-revenue, and its sensors are still in prototyping, with first commercial production targeted for 2027. The co-founder’s claim of at least 70 percent lower system cost is her estimate, and it names air conditioners that sense an empty room and drones as examples.
The Wire (headlines only)
- Sources I could not read in full. Bloomberg, the Wall Street Journal, Reuters, the Washington Post and X were not readable, so these are headlines only. Bloomberg ran “How Dual-Class Stock Entrenches Founder Control at Anthropic, SpaceX, Google” and a video, “Tech Stocks Plunge as Investors Await Anthropic IPO.” The Journal reported an exclusive that the Trump administration will name Nvidia among AI “national champions,” and a piece headlined “Nvidia’s Days as an Open Checkbook for the AI Economy Are Numbered.” Reuters ran Alibaba unveiling a new flagship model with safety upgrades, China’s drive to build an AI-native generation, and the U.S. and Russia opposing talks on limiting autonomous weapons. The Post asked questions about the FAA’s new SMART AI system, with officials saying there is no link to the ground stop. An X post claims a voice mode for Claude Code; its author is unverified.
The Throughline
Several of today’s stories are about money moving ahead of proof. Anthropic is asking public investors to fund a burn the Times says Goldman expects to top $65 billion next year, against a prospectus showing positive cash flow so far. Nvidia, by the figures the TechCrunch hosts cite, has investments and commitments well over $100 billion. In both cases the number that matters is the one that has not happened yet.
The consumer stories are the mirror image, because they are about what a person can check. Apple’s study deliberately grounds answers in named journals and asks participants to rate them. OpenAI’s wallet puts a biometric check on each purchase. The AI DJ column shows the opposite case, a feature that talks confidently, gets a name wrong, and offers no way to verify anything except by being annoyed. The trust mechanism is the product, and the ones that expose it look more durable than the ones that just sound sure of themselves.
The science and hardware items sit in between. OpenAI’s blog hands credibility to scientists by letting them describe how they used the tools, but its headline claims are, for now, OpenAI’s own. OATs and Bharat Semi are reminders that plenty of AI work is small, physical and cheap: a $20 module, an image sensor that does its own inference.
The Bigger Picture
Zoom out and the industry is shifting from raising money on potential to being asked for evidence. A public listing means quarterly numbers, dual-class control questions of the kind Bloomberg’s headline raises, and analysts who forecast losses out loud. The Nvidia story adds the other half: when one company is the financier, the supplier and the bellwether, trouble at the top gets priced across the whole sector.
The caution is how much here is secondhand. The Goldman forecast is relayed by two unnamed people to the Times, the Nvidia timeline is a podcast discussion, and several of the most interesting claims come from the companies selling the product. What looks solid is the direction: the next phase of AI is judged less by what it promises and more by what can be checked.
What to Watch
- The Anthropic listing. The Times says it is expected next Thursday. Watch the opening valuation against the $1.4 trillion figure and how the first public filings treat the stock-grant expenses.
- Nvidia’s permanent CEO. Colette Kress is interim, and the hosts see her as a caretaker. Whether she keeps funding the ecosystem is the real question.
- ChatGPT Wallet rollout. OpenAI says it will be gradual, so watch whether merchants block agent purchases or fraud systems decline them.
- Apple’s study. A 12-month, 200,000-person study will not report soon, but whether this reaches Health or Siri is the signal.
- Meta’s Muse Charm. WIRED says it is expected to launch in December, a test of the cute-AI-gadget thesis.