Anthropic's IPO and Office Push: How an AI Lab Is Repricing Software and the Compute Chain
This piece asks one question: Anthropic launched office products and a trillion-dollar-scale IPO on the same day, so what does that mean for the software moats of Microsoft and ServiceNow, and for order visibility at Nvidia, Broadcom, AMD and Micron? Data as of Nov 10, 2026, after the US close (17:20 ET).
Report date: 2026-11-11 | Data cutoff: US close on 2026-11-10 (after-hours as of 17:20 ET) | Sources: SEC EDGAR (Anthropic S-1/A filed 2026-11-10), CNBC / Fortune, third-party line-by-line reviews of the prospectus (Microventures / Luminix / Equibles), yfinance prices and valuations, coverage of Microsoft and ServiceNow earnings, X posts (@claudeai and others), Aya heat ranking
1. Two pieces of news in one day
On Nov 10 (ET), two announcements landed almost at once.
The first was about product. Anthropic's official account @claudeai posted that it was "starting to roll out today: Claude Whiteboards, Claude Sheets and Claude Data. Think a plan through on a board, build a tracker in a sheet, and tell a story with your data."
The second was about capital. The same day, Anthropic filed its second amended prospectus (S-1/A). It set a price range of $70–78 per share for 825 million shares and launched its roadshow. Per CNBC, the top of the range implies a fully diluted valuation of about $1.405 trillion and gross proceeds of up to $64.35 billion. Pricing is expected on Nov 18, with trading on Nasdaq under "ANT" from Nov 19. There is also a $26 billion concurrent private placement at the IPO price. The buyers are Google, Nvidia, AMD, Broadcom, Micron, Microsoft and TSMC, which is close to Anthropic's entire supply chain plus its biggest software rival.
The market reacted first in software. Microsoft fell 2.2% to $533.16 and ServiceNow fell 3.07% to $99.61, while the Nasdaq-100 ETF (QQQ) slipped just 0.68%.
This deserves a close look because the prospectus lays Anthropic out in full: about $81 billion of run-rate revenue at end-October, and about $900.9 billion of future spending commitments, mostly for compute. The company plays three roles at once: potential rival to software giants, major customer of the cloud providers, and investee of the chipmakers. Its numbers feed directly into how all three groups of stocks get priced.
One instinct needs correcting up front, though: the evidence does not support pinning all of Microsoft's and ServiceNow's declines that day on Anthropic. The sections below take it apart piece by piece.
2. What Anthropic actually sells
Start with how it makes money. Per reviews of the prospectus, revenue in H1 2026 was $16.29 billion. About 80% was usage-based: companies call the models through an API and pay for what they use. Subscriptions were only about 20%. About 48% of H1 revenue was resold through cloud marketplaces such as Amazon Bedrock and Google Vertex AI. So Anthropic is really a wholesaler of "intelligence processed out of compute". Its main customers are enterprise developers, not office workers sitting in Excel.
One term to define. "Run-rate" revenue is the latest month or quarter multiplied by 12 or 4. It shows how fast revenue is coming in right now, but it is not recognized full-year revenue. Per Fortune, Anthropic's run-rate rose from about $71 billion in late September to about $81 billion at end-October, a gain of roughly 14% in about a month. For comparison, Epoch AI (Aug 13, 2026) put it at roughly $9 billion at end-2025.
Quarterly revenue climbs just as steeply: $4.7 billion in Q1 2026, $11.6 billion in Q2, and about $16.95 billion in Q3. The Q3 figure is nine-month revenue of $33.24 billion minus H1's $16.29 billion, and it is up about 46% from Q2.
The chart shows two things: growth is still strong, but the sequential rate has dropped from more than doubling in Q2 to about 46% in Q3.
3. Core question 1: Can Claude Sheets dent Microsoft 365?
Our view: the near-term threat is mostly about sentiment, not revenue. The real worry is a shift in how software gets priced, not any single product.
Evidence 1: the products are still early. Per @claudeai and several accounts relaying the launch, Sheets and Whiteboards are in beta on paid plans only, and Meetings is for Team and Enterprise customers. Claude Data (formerly Dashboards) is now generally available and connects directly to Redshift, BigQuery, Databricks, Snowflake, Salesforce and other sources. X user @MichaelGannotti says Sheets is basic and exports to Excel, and that complex work still goes back to tools like Excel or Airtable. That is a KOL reading; we have not seen Anthropic's own documentation. A spreadsheet that exports to Excel is, for now, more of an on-ramp than a replacement.
Evidence 2: Microsoft and Anthropic are deeply tied "frenemies". Since 2026, Microsoft 365 Copilot has integrated Claude (branded Copilot Cowork), and Claude became generally available inside Excel, Word and PowerPoint in May 2026. The prospectus also shows Anthropic has committed to spend at least $44.9 billion on Azure between Nov 2026 and May 2033. That is about 6.6% of Microsoft's latest commercial remaining performance obligation (RPO, meaning revenue that is contracted but not yet recognized) of $677 billion. Microsoft is also a buyer in the private placement ($1 billion). In other words, the better Claude sells, the more Microsoft earns as cloud supplier and shareholder, which offsets part of the competition at the application layer.
Evidence 3: the stock does not support a "Microsoft re-rating" story. MSFT fell 2.2% on Nov 10, about 1.5 percentage points worse than QQQ. But measured from Oct 28 (the day before both companies reported) to Nov 10, MSFT is actually up 0.4%, while QQQ is down 7.0%. The stock that was truly re-rated is ServiceNow. It fell 26.95% in a single session on Oct 29 because of its own results. Q3 subscription revenue was a solid $3.988 billion, up 21%. But Q4 cRPO growth guidance (cRPO is contracted revenue to be recognized within 12 months) was only 18.5%, and rising AI serving costs are squeezing margins. The 3.07% drop on Nov 10 was just one more step down after a roughly 30% fall.
What to notice is when the gap opened: ServiceNow's collapse came on its Oct 29 earnings day, not on Nov 10 when Anthropic launched its products.
So where is the real risk? Our inference: pricing models. Traditional SaaS (software as a service) charges per seat: a company with 10,000 Office users pays for 10,000 seats. About 80% of Anthropic's revenue is usage-based, so its growth does not depend on how many people a customer employs. If AI agents cut the headcount needed for the same work, seat-based vendors face a paradox: the more customers use AI, the less the vendor collects.
Both companies' latest comments point in this direction:
- ServiceNow: management said on its call that "seats aren't going anywhere, AI usage sits on top". It also said production AI usage is up 13x and that agent running costs need watching. That suggests the pressure shows up first in gross margin, not revenue.
- Microsoft: in the latest quarter, Copilot daily users rose more than 5x year over year, and customers consuming Copilot credits on usage-based billing grew more than 75% quarter over quarter. Microsoft itself is moving toward usage-based pricing.
The counter-evidence has to be stated too. Anthropic is penetrating enterprises faster than any listed software company ever has. Per the prospectus reviews, it has more than 1,900 customers spending over $1 million a year. Net revenue retention, excluding consumer, is about 900%, meaning the same customers now spend 9x what they did a year ago. When a model supplier's customers grow their budgets 9x in a year, at least part of that IT budget is being taken from other software. But no data yet lets us put a number on how much of that squeeze falls on Microsoft or ServiceNow.
On sentiment, retail and big money are not moving the same way. On the Aya heat ranking, MSFT scored 72 on Nov 10 and ranked second, with 60% of self-tagged posts bearish, the only top-10 name where bears were the majority. Yet the same day, a 101-day $595 MSFT call traded 24,216 contracts against prior open interest of just 131. That comes from the platform's options scan, which cannot tell whether buyers or sellers opened the position.
4. Core question 2: Is $1.3 trillion expensive?
Our view: on revenue multiples, Anthropic is priced below OpenAI and above Microsoft, roughly in the range of a high-growth stock. The debate is about revenue quality, not the multiple itself.
First, the valuation. The top of the range implies a fully diluted value of $1.405 trillion, so the $74 midpoint scales to about $1.333 trillion. The topic brief cited a "midpoint of about $1.18 trillion"; we could not verify that in the source documents, and it may be based on undiluted shares. This piece uses fully diluted figures throughout.
One more term. The price-to-sales ratio is market value divided by revenue, a common way to value companies that are not yet steadily profitable. The table below compares several companies; definitions differ, see the note:
| Company | Valuation / market cap | Revenue basis | Multiple |
|---|---|---|---|
| Anthropic (IPO midpoint, fully diluted) | ~$1.333T | End-Oct run-rate $81B | ~16.5x |
| Anthropic (same) | ~$1.333T | Q3 revenue ×4 ≈ $67.8B | ~19.7x |
| OpenAI (pre-money in talks) | $1.4T | ~$50B run-rate at end-Sept (FT) | ~28x |
| Microsoft | $3.96T | Trailing 12 months | 11.4x |
| Nvidia | $4.08T | Trailing 12 months | 13.5x |
| ServiceNow | $103.4B | Trailing 12 months | 6.7x |
Note: Anthropic and OpenAI are measured on run-rate; listed companies on trailing 12-month revenue. For fast-growing companies, run-rate is higher than trailing revenue, so the first two multiples look lower than they would on a like-for-like basis. On Bloomberg's reported year-end target of $70 billion run-rate, OpenAI would be about 20x.
The table shows two things:
- Of the two frontier labs, Anthropic is "cheaper". The two valuations are similar, and Anthropic's revenue pace is about 60% higher.
- The market pays model suppliers roughly two to three times what it pays application software. Comparing with ServiceNow's 6.7x makes this clear, and it is itself one sign that software moats are being repriced.
On profitability, most of the GAAP loss is an accounting effect. The prospectus shows a startling GAAP net loss: $80.7 billion in H1 2026 and $104 billion over nine months. But about $72.4 billion of the H1 loss is fair-value changes on convertible notes and SAFEs. That is a non-cash accounting loss that grows as the company's valuation rises. The operating figures say more: an H1 operating loss of $693 million, adjusted operating income of $1.6 billion (about a 10% margin), and positive operating cash flow of $3.9 billion. Q2 alone posted operating income of $974 million.
The real weak spot is the cost structure. H1 compute and infrastructure spending was $8.95 billion, about 55% of revenue. Equibles estimates from the prospectus that this ratio was about 58% in Q3, up from 48% in Q2. A model company's "gross margin" comes down to how much of revenue goes to compute, so a rising ratio means economies of scale have not settled in yet. Revenue has two further risk points:
- About 80% of revenue is usage-based, and customers can switch to cheaper models at any time;
- The top five customers account for about 25% of nine-month revenue.
5. Core question 3: What does a $900 billion commitment mean for the chip chain?
Our view: the commitment list gives Broadcom and AMD long-term orders with real dollar figures, gives Nvidia mostly indirect demand, and gives Micron almost no direct terms. Whether any of it gets delivered depends on Anthropic's revenue continuing to grow fast.
Start with the total. Per reports on the amended prospectus, Anthropic's purchase commitments plus leases total about $900.9 billion, "substantially all of which is related to compute and infrastructure". Equibles counts about $895.4 billion as of Sept 30. The gap comes from the measurement date and whether leases that have not yet started are included. Even on the smaller figure, it is about 11x end-October run-rate revenue.
Look at the ranking: Broadcom is first, and Nvidia is not on the list of direct counterparties.
Company by company:
Broadcom is the largest single counterparty, with about $161.2 billion of equipment rent over roughly five years. That works out to about $32.2 billion a year, or about 36% of Broadcom's trailing revenue of $89.1 billion. This fits what Epoch AI recorded on Aug 13, 2026: Broadcom, Apollo and Blackstone arranged about $35 billion of compute equipment financing for Anthropic and described it as the first deal on a financing platform. But Broadcom's own disclosures show the customer may issue up to $42 billion of convertible notes to pay its lease obligations, with Broadcom's maximum exposure on some tranches at $29 billion. In other words, Broadcom is helping finance part of these orders itself.
AMD's commitment is Helios systems "expected to exceed $20 billion", starting in 2027. That is about 44% of AMD's trailing revenue, but it will be spread over several years.
Nvidia is not on the direct list. But SpaceXAI (ceiling $120.5 billion), Nscale (ceiling $44.6 billion) and Lambda ($41.3 billion) are all GPU compute providers, so most of that demand eventually flows to Nvidia. Nvidia's $10 billion in the placement is less than 0.3% of its ~$4.08 trillion market cap; it looks more like a way to lock in the customer relationship.
Micron appears only in the placement ($1 billion). Reports on the prospectus show no direct purchase commitment to Micron. Memory demand will rise indirectly with compute, but this filing gives Micron no visibility it can quantify.
Two uncertainties deserve emphasis.
- The commitments are not all the same kind. Some are minimum spends, some are ceilings, some are rent. They cannot simply be added up, and they are certainly not "guaranteed revenue".
- There is an obvious loop. Suppliers invest in Anthropic, Anthropic uses the funding to buy their compute, and the suppliers book revenue. The loop holds together as long as Anthropic's revenue keeps growing. If growth stalls, the fixed compute bill hits its cash flow immediately. The risk factors state that many commitments must be paid regardless of usage.
The day's stock moves also show the market did not treat the list as good news for all chip stocks.
Measured from Nov 4 (before the post-midterm AI sell-off), AMD is up 5.4% and Broadcom up 2.0%, while Nvidia is down 15.6% and Micron down 8.8%.
- Nvidia's decline is mainly about the CEO controversy and order-cut rumors (see our Nov 9 and Nov 11 coverage), not directly about Anthropic.
- Broadcom rose 3.91% on Nov 10 with no direct catalyst we could find on X. The "Nvidia cuts orders, capacity shifts to ASICs" rumor and the Anthropic commitment list could both be reasons, but the evidence for either is weak.
6. What the market is betting on
Taken together, current prices embed three assumptions.
First, on Anthropic. At 16.5x run-rate revenue, the market is betting that run-rate keeps stepping up and that compute's share of revenue comes down. As a scenario (not a forecast): if run-rate doubles to more than $160 billion in a year and the multiple falls to about 8x, the valuation stays roughly flat. In other words, this price requires revenue to double again, not just hold steady.
Second, on software. ServiceNow's forward P/E of about 20.7x is now below Microsoft's 22.0x, which has rarely happened. It shows the market has moved ServiceNow from "high-growth SaaS" to "mature software that AI may squeeze". Microsoft has not been treated the same way because Azure lets it collect an "AI toll": Azure grew 48% in the latest quarter.
Third, on fund flows. At the top of the range, the IPO plus placement add about $90.35 billion of new equity supply. Per CNBC, about 20% is allocated to retail, far above the usual 5–10%, and placement shares have no lock-up. A plausible result is that some money seeking a "pure AI" holding will rotate out of software and even some megacap tech. But no fund-flow data confirms this yet; it is an inference.
7. What to watch next
Key dates over the next month, in order:
| Date (ET) | Event | What it tells us |
|---|---|---|
| Nov 13 | US October PPI | CPI already ran hot; a hot PPI would add discount-rate pressure on richly valued growth stocks |
| Nov 17, after close | Nvidia earnings | Data-center guidance and comments on large-customer orders are first-hand evidence on "indirect demand" |
| Nov 18 | Anthropic pricing | Pricing at the top of the range, or raising it, shows how willing institutions are to pay ~16x run-rate |
| Nov 19 | ANT first trade | Day-one performance becomes the new valuation anchor for AI labs; OpenAI's fundraising will reference it |
| Dec 9 (yfinance) | Broadcom earnings | Whether Anthropic leases show up in AI revenue guidance |
Two medium-term signals to add:
- Anthropic's first quarterly report as a public company: watch compute spending as a share of revenue. If it stays above 55%, the profitability case needs a rethink.
- Microsoft's earnings in late January 2027 (yfinance estimate): watch M365 commercial seats and Copilot usage revenue. Seat growth slowing while usage revenue speeds up would be the real sign that the pricing shift has arrived.
8. Conclusion
Anthropic has laid out an AI lab's business model in full: about $81 billion of run-rate revenue, about $900.9 billion of compute commitments, and a revenue mix dominated by usage-based billing.
For software stocks, the threat is a long-term shift in pricing models, not a single beta product like Claude Sheets. ServiceNow's re-rating came mainly from its own Oct 29 guidance, and Microsoft has still beaten the Nasdaq by about 7 percentage points since Oct 28.
For the chip chain, Broadcom and AMD have long-term contracts with real dollar figures, Nvidia benefits indirectly, and Micron has no direct terms. Whether these orders get delivered depends on Anthropic's revenue continuing to grow fast.
The nearest tests are Nvidia's earnings on Nov 17 and the IPO pricing and first trade on Nov 18–19.
Appendix: Data and Sources
A. Anthropic key financials (from prospectus reviews, USD 100m)
| Item | Value | Source |
|---|---|---|
| 2024 revenue | 3.86 | Microventures / Luminix |
| 2025 revenue | 45.85 | Same |
| 2026 Q1 / Q2 revenue | 47 / 116 | Microventures |
| 9M 2026 revenue | 332.4 | CNBC |
| Q3 2026 revenue (derived) | 169.5 | 332.4 − 162.86 |
| H1 2026 operating loss / adjusted operating income | −6.93 / 16.18 | Luminix |
| H1 2026 net loss (of which fair-value changes) | −807.48 (−723.88) | Luminix |
| 9M 2026 net loss | −1,040 | CNBC |
| H1 2026 operating cash flow | +39 | Microventures |
| H1 2026 compute & infrastructure | 89.45 | Luminix |
| End-H1 2026 cash & short-term investments | 1,317 | Luminix |
| Run-rate: late Sept → end Oct | 710 → 810 | Fortune |
B. Formulas and inputs
| Metric | Formula | Result |
|---|---|---|
| Q3 revenue | 332.4 − 162.86 | $16.95B |
| Q3 QoQ | 169.54 / 116 − 1 | +46.2% |
| Midpoint FD valuation | 14,050 × 74/78 | ~$1.333T |
| Midpoint / run-rate | 13,329 / 810 | 16.5x |
| Midpoint / annualized Q3 | 13,329 / (169.54×4) | 19.7x |
| OpenAI multiple | 14,000 / 500 | 28x |
| Commitments / run-rate | 9,009 / 810 | 11.1x |
| Broadcom annual rent / its revenue | (1,612/5) / 891 | 36.2% |
| AMD commitment / its revenue | 200 / 454 | 44.1% |
| Azure commitment / MSFT commercial RPO | 449 / 6,770 | 6.6% |
| H1 compute / revenue | 89.45 / 162.86 | 54.9% |
| One-month run-rate gain | 810 / 710 − 1 | +14.1% |
| IPO + placement (top) | 643.5 + 260 | $90.35B |
| Placement sum | 100+100+25+10+10+10+5 | $26B |
| Price changes | yfinance closes, bases 10/28 and 11/4 | See charts |
C. Timeline
- 2026-06-01: confidential IPO filing (DRS)
- 2026-10-26: public S-1
- 2026-10-28/29: Microsoft and ServiceNow report; NOW −26.95% on 10/29
- 2026-11-02: first S-1/A
- 2026-11-05: post-midterm AI sell-off (Fortune: Nasdaq −5.5% last week)
- 2026-11-10: second S-1/A with $70–78 range and roadshow launch; Claude Sheets / Whiteboards / Data / Meetings launched the same day
- 2026-11-17: Nvidia earnings (after close)
- 2026-11-18 / 19: expected pricing / first trade
D. Aya heat data (close of 2026-11-10)
| Ticker | Rank | Heat | Bull/Bear (self-tagged) | Drivers |
|---|---|---|---|---|
| NVDA | 1 | 90 | 83% / 17% | Buzz, volume, sharp drop, options activity |
| MSFT | 2 | 72 | 40% / 60% | Buzz |
| AVGO | 4 | 67 | 97% / 3% | Sharp gain |
| AMD | 6 | 64 | 81% / 19% | Buzz |
| MU | 11 | 59 | 97% / 3% | Buzz |
ServiceNow was not in the top 20; when the topic was selected, its discussion volume was 3.3x normal.
E. Market views
- @claudeai (official): "Starting to roll out today: Claude Whiteboards, Claude Sheets and Claude Data."
- @Samana1yst: "MSFT -2.2% and NOW -3.1% after Anthropic launched … reviving the 'AI eats software' worry."
- @MichaelGannotti: Sheets / Whiteboards still in beta and fairly basic (KOL reading)
- We found no influential X account with an explicit bullish take on MSFT; on the options side, a large new $595 call position appeared (direction unconfirmed)
F. Sources
- SEC EDGAR: Anthropic, PBC S-1/A, accession 0001628280-26-072484, 2026-11-10
- CNBC, 2026-11-10: https://www.cnbc.com/2026/11/10/anthropic-sets-price-range-of-70-to-78-in-upcoming-ipo.html
- Fortune, 2026-11-10: https://fortune.com/2026/11/10/anthropic-ipo-price-range-ai-stock-sell-off/
- Luminix: https://www.useluminix.com/reports/company-overviews/anthropic-s1-what-the-filing-shows
- Microventures: https://microventures.com/dissecting-anthropic-s-1
- Equibles: https://equibles.com/research/anthropics-s-1-checked-against-amazon-google-microsoft-and-broadcom-filings
- Web search: Microsoft FY27 Q1 coverage (Oct 28–30, 2026); ServiceNow Q3 2026 coverage (Oct 29, 2026); OpenAI run-rate and valuation (FT / Bloomberg via secondary reports, Oct 2026); placement split and $900.9B commitment reports (Nov 2026)
- yfinance: MSFT / NOW / NVDA / AVGO / AMD / MU / QQQ prices and valuations, as of the Nov 10, 2026 close
- X (via grok-x): @claudeai, @Samana1yst, @MichaelGannotti and others, Nov 10, 2026
- Internal research corpus: Epoch AI (Aug 13, 2026)
- Aya heat ranking and platform options scan (Nov 10, 2026)
For information and research discussion only. Not investment advice.
<!-- AYA_PODCAST -->🎙️ Podcast script (Chinese, ~20 min, 泓观阳略): read it here
AI-generated by Aya Research from public data and disclosures, for information and research purposes only. Not investment advice.