Anthropic sets its IPO range and Wall Street promptly argues about whether it's too high, too low, or a referendum on the whole AI trade, while a new Claude productivity suite knocks Microsoft's stock. Meanwhile Google's Gemini 4 Carbon finally ships, Nvidia's spy scandal deepens, and OpenAI posts a record quarter. The honest takeaway from all the IPO numbers: nobody actually knows yet.
Heading into Anthropic's IPO, we know a lot of numbers and almost nothing at all.
Anthropic is pricing at $70–78 a share, implying $1.25–1.4 trillion, and its banks are already mulling a bump to $1.5 trillion on strong demand. Bears read the drift down from the splashy $2 trillion chatter as a warning for the AI trade. Contrarians point to $81 billion in run-rate revenue and retail desperate for frontier-lab exposure. The valuation targets were never Anthropic's own; they came from anonymous investors and stuck. Microsoft dipping on a Claude spreadsheet launch is more evidence of a market flailing until the stock actually trades. Buckle up: the next phase starts next week.
Some speculated that the midterm wave bringing data-center opponents to power might deter Anthropic. Instead it has kicked off its roadshow and is aiming to list next week.
Nothing in Dario's roadshow pitch, an "age of miracles" with risks carefully addressed, is new. What's new is the balance. The miracles are the point, and the risks are what must be overcome to get there. That ordering has eluded the company in the past.
— Dario Amodei, Anthropic CEO, to Bloomberg's Emily Chang. Emily Chang asked whether he was comfortable with Claude helping the military kill more people more quickly. Dario reframed it as whether he wants a stronger country, then said decisions about military use aren't the technology provider's to make beyond some very high-level boundaries. After the clashes with the Department of Defense/War, NLW notes he has never been quite this crisp.
Anthropic seeks to raise $57–64 billion in the IPO, plus $26 billion from industry partners Nvidia, Microsoft, Google, and Broadcom, with Nvidia and Google in for up to $10 billion each. With the greenshoe option, the total float comes to around $100 billion.
The $2 trillion figure came from an anonymous investor source of unknown proximity to the company, and it was splashy enough to stick. Anthropic couldn't respond during its quiet period. The slide from $2T to $1.6T to $1.5T to talk of $1.1T says something about shifting investor mood, but it isn't a goal Anthropic set and missed.
The Financial Times reports Anthropic's bankers plan to raise the share price to a top-end valuation of $1.5 trillion after "strong indications of demand." The FT adds that Anthropic cares far more about raising cash than about valuation, given its large compute commitments.
Investors have no clear idea what to expect, what's reasonable, or what other investors think is reasonable. Bears say the lower valuation is bearish and that a weak debut will trigger another leg down in the AI trade. Until the stock trades, everyone is flailing.
Artemis's John Ma says the consensus view is now bearish: overvalued, revenue flattening, tokens commoditizing. The counter: run-rate revenue rose $10 billion in October alone to $81 billion, still driven mainly by coding agents. Finance, legal, and life sciences are barely tapped, and retail investors may pile in regardless and push it toward $3 trillion.
iCapital's Sonali Basak notes that $1.4 trillion is around 40× trailing 12-month revenue. Palantir trades at roughly 55×, which puts Anthropic closer to the multiples of CrowdStrike and Cloudflare.
Anthropic disclosed $10 billion in new seven-year commitments with Microsoft and Google. It also said its $35 billion Lambda deal is cancelable after Lambda missed a deadline, though it doesn't currently intend to cancel. That reframes the worry about hundreds of billions in non-cancellable commitments: which fails first, AI demand or Neoclouds' ability to finance and deliver data centers on time?
Claude Sheets is an Excel competitor built for collaboration between humans and agents. Whiteboards is a team planning surface. Claude Data, which evolved from Dashboards, pulls live CRM and database feeds and competes with Power BI, with model-generated visual data stories and narrated videos added on top.
Anthropic framed the suite as where work is heading, not a declaration of war. Its legal team built a Slack-based review tool that drafts redlines from its own playbook. Its finance team turned an executive update into an interactive page the team maintains with Claude. Neither team writes code.
Once Claude becomes the shared work surface, it needs this kind of infrastructure around it. The place where work happens is increasingly inside the LLM harness or app, and this suite is a natural extension of that. Matthew Berman: "If you build software, the labs will compete with you." Microsoft fell 2.2% on the news.
The counterpoint: no enterprise is ditching Excel, and Claude Sheets even exports to it. Microsoft is buying into the IPO, is owed $45 billion in cloud spend, and resells Claude through Copilot. NLW sees the dip as more evidence of a market still working out what this all means.
Google previewed Carbon at the end of October as the top model on the Artificial Analysis Index but held it back for the government's voluntary safety testing. Two weeks later, Fable 5.2 has taken the top spot, but early testers are finding lots to like. Ethan Mollick: it shows "Google is back in the race of frontier models, rapid releases, and rapid improvement."
On Pavel Hurin's bug-hunt test, Carbon found 57 of 105 bugs, one behind Fable 5.2's 58, and completed the run for $31 versus almost $140. The cost savings held across other use cases: one render test matched Fable 5.2 at a third of the cost. One tester: "Nobody should be laughing at Google anymore."
— Pushmeet Kohli, VP at DeepMind, on Carbon's formal verification of Signal. DeepMind showcased Carbon formally verifying Signal's security properties, which required analyzing the underlying machine code. Formal verification is usually dismissed as too complex and time-consuming to be practical. Kohli pitched the result as frontier AI acting as a force multiplier against modern cyber threats.
At a Monday press conference, President Trump said the FBI is looking into the matter but called Jensen Huang "a very honorable man." He acknowledged knowing about the affair before it was reported but was vague on timing. The New York Times reported he was fully briefed last April and chose to do nothing.
Sources who claim to know the alleged spy, Alice Huang, told The Information that a vast smuggling network sat behind Megaspeed, the Neocloud that employed her. They gave a full recap of how the chips flowed through it.
Per The New York Times, Nvidia's general counsel discussed the case directly with investigators, saying their superior had approved the contact. Agents worried that was "potentially tainting the case." Commerce is now working on new export controls, including closing the remote-access loophole, but is delaying an entity-list decision on Megaspeed. House Foreign Affairs Chair Brian Mast says he'll subpoena Jensen if he won't come in.
Jensen runs a famously flat company with 60 direct reports. The Wall Street Journal reports there is no succession plan and a board treated more as a governance formality than a check on the CEO. With an outside law firm investigating, removing a charismatic CEO with no successor looks risky even if the findings are damning. The Journal suggests a "reined-in" Jensen is the best option.
Nvidia is down 11% since the story broke and makes up about 7% of the S&P 500 and 8% of the Nasdaq, so most US investors are heavily exposed. Its success is also now a national-security matter because of the AI race. There is no simple fix. NLW: the entire nation is experiencing "the mother of all keyman risks."
Q3 revenue rose 50% from Q2, after Q2 growth of only 17% compared with Anthropic's 144%. The drivers were API demand for recent models, including GPT-6 Astra, quietly strong image-model demand, and ads running at a $1 billion annualized pace. Operating losses grew to an estimated $10.1–10.9 billion, though operating margin improved from −150% to −109%.
The startup, founded less than a year ago by Sonder founder Francis Davidson, booked $60 million in travel across tens of thousands of users last summer. Davidson says he threw out the old playbook for a lean, AI-first process, including dictating about 1.2 million words a month. The service has shut down, and OpenAI is quiet about its travel plans.
OpenAI's new business hub gives advertisers an analytics-style dashboard and lets brands deploy "business agents" that consumers can talk to after clicking an ad. NLW reads this as the start of ChatGPT competing more directly with Meta. Expect much more treatment of ChatGPT as a commerce platform, since so many of its answers already touch shopping.
