Anthropic's IPO range puts a price on frontier AI
What moved on our finance boards: Anthropic's filing prices frontier AI, Nvidia's CEO resigns, AI cloud demand holds and a humanoid maker lists.
For years the frontier AI labs have been valued in private rounds. This week one of them filed the numbers that let the public market set the price, and the rest of the AI market started repricing around it. Here is what changed on our finance boards this week, why it matters, and what we expect next.
Anthropic's filing puts a public price on frontier AI
Anthropic's amended S-1, filed on November 10, states run-rate revenue, its own measure, of about $81 billion at October 31, up from more than $47 billion in May, and $17.0 billion of revenue in the third quarter alone. It set a range of $70 to $78 a share. At the top of that range, CNBC puts the fully diluted value at up to $1.4 trillion, against a last private mark of about $965 billion. Pricing is expected on November 18 and trading on November 19.
That moves the reference point for every private lab. OpenAI told investors its run-rate was close to $50 billion at the end of September, counted more narrowly than Anthropic counts its own, and it has said it will not list in 2026; its CFO told staff to expect 2027. Cohere's growth is real, with annual recurring revenue up 165% on a year earlier, but its Series E prices it at C$35.7 billion, so the discount that made it the cheap option in the group is gone. Our State of Frontier AI board cut Cohere to low conviction on price, not on the business.
The board's read on Anthropic itself is balanced. On a fully diluted basis the whole range sits above the private mark, so much of the upside may already be priced in. Its adjusted operating margin fell to 15% in the third quarter from 19% in the second as short-term compute charges recurred, and on a GAAP basis the quarter was an operating loss of $567 million.
| Company | Latest price | Set by | What comes next |
|---|---|---|---|
| Anthropic | $70–78 a share; up to ~$1.4T fully diluted | IPO range, Nov 10 | Pricing expected Nov 18, trading Nov 19 |
| OpenAI | ~$852B | March round, reused in an August share tender | Not listing in 2026; its CFO points to 2027 |
| Databricks | $190B | $5B round, Aug 13 | No public S-1 as of Nov 14 |
| Cohere | C$35.7B | Series E first close, Oct 29 | Aleph Alpha deal agreed Sep 16; no close announced by Nov 13 |
| Mistral | Above €21B | €3B Series D, announced Sep 8 | Targets $1B of annual recurring revenue by the end of 2026 |
| SSI | ~$32B, as reported | Round reported in April 2025 | Nvidia committed about $5B in July, terms undisclosed |
Our call: the window is open. We expect at least one more company in that table to file publicly for an IPO by June 30, 2027. We are wrong if none has by then.
Nvidia's risk moved from demand to governance
Demand is not Nvidia's problem. Data-center revenue was $89.0 billion last quarter, up 117% on a year earlier, at a 75.0% gross margin, and Nvidia guides $108.0 billion of revenue for the quarter it reports on November 17.
Governance is. On November 8 Nvidia's board opened an independent investigation into reporting about its founder and CEO, Jensen Huang, saying it had no reason to believe national security or the company's secrets were compromised. On November 11 it accepted his resignation as CEO and director. CFO Colette Kress is interim CEO, and Harvey Jones was named chair.
Our State of Frontier AI and State of Physical AI boards cut Nvidia from high to medium conviction while the board's decision was pending. They hold it at medium until a permanent CEO is named. Tuesday's results will be the first Nvidia reports under an interim CEO, and our State of Nvidia board tracks both.
The second source of AI chips became real at the same time. AMD said on its earnings call that it began volume shipments of its MI455X in the third quarter, Anthropic, Meta, Microsoft, OpenAI and Oracle are running workloads on its first Helios racks, and its data-center revenue rose 97% to $8.5 billion. Our State of AI Compute board raised AMD from low to medium. What AMD has not shown is pricing power.
Show as table
| Setup | % | Source |
|---|---|---|
| Nvidia, Q2 FY2027, to late July 2026 | 75.0 | Nvidia results, Aug 26 2026 |
| AMD, Q3 2026, to late September | 54.0 | AMD results, Nov 3 2026 |
Our call: AMD delivers its fourth-quarter guide of about $17 billion at a non-GAAP gross margin near 55%, so the volume keeps coming and the pricing gap stays. We are wrong if fourth-quarter revenue misses that guide or that margin falls below 55%.
For Nvidia, Tuesday tests pricing power as well as leadership. A gross margin at or above the 74.0% midpoint keeps the pricing case intact, and a print below 73.5%, the bottom of the guide, would be the first sign it is fading.
AI cloud: demand held, the dates moved
The quarter's results from the AI clouds point one way. CoreWeave, the most leveraged link in the chain, reported a backlog of about $132.5 billion, raised its 2026 guidance and raised about $4.2 billion of convertible notes, so the financing strain our boards watch for is not showing in its numbers. Nebius grew revenue 571% with a 47% group adjusted EBITDA margin. Google Cloud grew 103%, with a backlog of $560.8 billion.
What slipped was timing. IREN moved its goal of more than $4 billion of annualized run-rate revenue, its own measure, from the end of 2026 to January, citing a customer-requested design change and supplier timing, and Nebius moved its run-rate target out to February. Demand is running ahead of what can be built and connected.
| Company | Demand | What moved |
|---|---|---|
| CoreWeave | Backlog ~$132.5B; 2026 guidance raised | ~$4.2B of convertible notes raised; whether they retire its 2026 maturity is not stated |
| Nebius | Revenue up 571% on a year earlier; 47% group adjusted EBITDA margin | Run-rate target moved to above $7B for February 2027 |
| IREN | More than $4B of annualized run-rate revenue, its own measure, still the goal | Moved from the end of 2026 to January, on a customer design change and supplier timing |
| Google Cloud | Revenue up 103% on a year earlier; backlog $560.8B | 2026 capex held at $195–205B |
Our call: through March 2027 the AI cloud's misses stay timing slips, not cancellations, and no hyperscaler cuts its capex guidance. We are wrong if one does, or if a neocloud defaults. Our State of the AI Cloud board tracks both.
Physical AI: a humanoid maker lists, and sensors get cheaper
Agility Robotics, maker of the Digit warehouse humanoid, has traded on Nasdaq as AGLT since November 5, after its merger with Churchill Capital Corp XI closed at a $2.5 billion pre-money value. At its November 12 close of $7.58 a share, its 242 million shares were worth about $1.84 billion. It starts public life priced on its order book and its cash rather than on the promise of humanoids: about $398 million of pro forma cash at September 30, against $131.6 million of operating cash outflow in the first nine months of the year.
Further down the stack, the parts are getting cheaper fast. Hesai shipped 65% more lidar units in the third quarter, but its lidar revenue grew only 23%, so it earned about a quarter less per unit, and it remains on the Pentagon's list of Chinese military companies while it contests the listing in court. Our State of Physical AI board cut Hesai from medium to low. Cheaper sensors help robot makers and squeeze sensor margins.
Show as table
| Setup | % | Source |
|---|---|---|
| Lidar units shipped | 65.3 | Hesai Q3 2026 results |
| Lidar revenue | 22.9 | Hesai Q3 2026 results |
Our call: Hesai's lidar revenue per unit keeps falling as volume rises. We are wrong if a quarter's lidar revenue grows faster than its unit shipments.
The ratings that moved
| Company | Board | Moved | Why |
|---|---|---|---|
| Nvidia | State of Frontier AI; State of Physical AI | High → Medium | Cut while the decision on its CEO was pending; held at medium until a permanent CEO is named |
| AMD | State of AI Compute | Low → Medium | Volume proven: MI455X in volume, five customers on Helios racks; margin ~55% |
| Cohere | State of Frontier AI | Medium → Low | Growth is real, but the Series E price removed the discount |
| Hesai | State of Physical AI | Medium → Low | Lidar revenue per unit down 25.6%; still on the Pentagon’s Chinese military company list |
| Agility Robotics | State of Physical AI | Now listed as AGLT | Trades at ~$1.84B against a $2.5B merger value |
What we're watching
| When | What | Why it matters |
|---|---|---|
| Tue Nov 17 | Nvidia's Q3 FY2027 results | Gross margin against the 74.0% guide, in the first results under an interim CEO |
| Wed Nov 18 – Thu Nov 19 | Anthropic prices and starts trading (expected) | The first public price for a frontier lab, against a ~$965B private mark |
| Pending | Nvidia names a permanent CEO | What our medium rating on Nvidia waits for |
| Early 2027 | AMD's Q4 2026 results | Whether ~$17B of revenue arrives at a ~55% margin |
Our calls, in one place
We grade every call when its date arrives, and the wrong ones stay up with their outcome.
| Call | By | Confidence | Proved wrong if |
|---|---|---|---|
| Another private AI lab or platform from the table above files publicly for an IPO | Jun 30 2027 | Medium | None has filed by then |
| AMD delivers its ~$17B Q4 guide at a non-GAAP gross margin near 55% | Q4 2026 results | Medium | Q4 revenue misses the guide, or that margin falls below 55% |
| AI cloud misses stay timing slips; no hyperscaler cuts its capex guide | Mar 31 2027 | Medium | A hyperscaler cuts its capex guide, or a neocloud defaults |
| Hesai's lidar revenue per unit keeps falling as volume rises | Q4 2026 results | Medium | A quarter in which lidar revenue grows faster than units shipped |
Every rating and call here comes from the live research. Open State of Frontier AI for the full read, or the AI Stack to see where each company sits in the value chain.
Correction: Agility Robotics now appears once, as a listed company. Its earlier private-company profile overstated its 2025 revenue; the listed profile carries the filed figure.
Not financial advice. Ratings and calls are QAI research views, not recommendations or solicitations, and board positions are illustrative. Conflicts of interest: none.