Anthropic began trading on Nasdaq on Thursday, November 19, 2026 under the symbol ANT. The first trade was $111.50, the close was $97.17, and the final prospectus contains a sentence that matters to anyone who builds on Claude: the company says it has changed usage limits and rate limits before and may change them again. This article covers what the day-one numbers show, who was allowed to sell shares, what the Anthropic IPO prospectus says about rate limits and compute, and one contract in the filing that either side can already cancel. It accompanies today's Daily Diff episode.
TL;DR
- Anthropic priced its IPO at $83.00 on November 18, sold 980 million Class A shares (Anthropic), and opened trading at $111.50 on Thursday. The close was $97.17, 17 percent above the offer price. The lowest price of the session was $92.72.
- Valuation figures differ by share count: CNBC says about $1.8 trillion at the close and Forbes says $1.27 trillion.
- The prospectus lets non-founder directors, officers and employees sell up to 315 million shares from the first trading day. It says they may. It does not say anyone did.
- In its risk factors Anthropic writes that it has changed, and may in the future change, usage limits, rate limits, or model and product capabilities.
- Anthropic says it has about $900.9 billion of compute and infrastructure commitments, $70.7 billion of them due within twelve months. One $34.9 billion Lambda agreement is "currently terminable by either party".
- My verdict is NEEDS REVIEW: keep building on Claude and keep a second model wired in.
What did the Anthropic IPO do on day one?
The offer price was $83.00, above the $70 to $78 range the company had filed on November 10, according to the prospectus history that CNBC and Reuters summarise.
Yahoo Finance, Nasdaq, CNBC, Forbes and Reuters agree on these numbers:
| Item | Price |
|---|---|
| Offer | $83.00 |
| First trade | $111.50 (+34 % over the offer) |
| Day high | $115.10 |
| Day low | $92.72 |
| Close | $97.17 (+17 % over the offer) |
Two readings follow from the table. The stock never traded below the offer price during the regular session, so anyone who bought at $83 was ahead all day. And it closed about 13 percent below the first trade, so anyone who bought at the open was behind. Reuters describes the early exuberance as fading. I am quoting day one only. The second session was already trading while this article was written, and a price here would be stale.
Which market value is right?
Three outlets give three figures for the close. CNBC says Anthropic closed with a market cap of about $1.8 trillion. Forbes lists $1.27 trillion at the close. Reuters writes that its $2 trillion figure at the debut is fully diluted and includes options and restricted stock units that have not vested, and that the value on outstanding shares reached as much as $1.5 trillion. The prospectus counts 13,219,642,274 Class A and Class N shares after the offering and the concurrent private placement, and 13.22 billion shares times $97.17 is about $1.28 trillion.
Bessemer's Byron Deeter, whose firm is an Anthropic investor, told CNBC: "They could have priced the IPO higher. At every financing they have chosen to leave some money on the table." Bessemer is an Anthropic investor, so read the comment as an insider's view of the pricing.
Who was allowed to sell Anthropic shares on day one?
The prospectus describes lock-up and market standoff agreements covering about 15.5 billion shares. Most existing holders are restricted for up to 180 days, and the co-founders and certain holders of more than 0.5 percent of fully diluted shares for up to 365 days. The release schedule has exceptions. On the first trading day, non-founder directors, officers and current and former employees and consultants are eligible to sell up to 315 million Class A shares. That is about a third of the 980 million shares the IPO sold. A second tranche of up to 303 million shares is scheduled for the 27th day after the prospectus date, December 15, 2026, and larger waves follow around its earnings releases and at the 180th and 365th days. The concurrent private placement shares, about $26 billion of Class N stock, are not subject to the lock-up.
"Eligible to sell" is the filing's wording. The filing does not say that anyone sold, and I did not find a source that does.
What does the Anthropic prospectus say about rate limits?
If you build on Claude, this is the part of the filing I would read twice. In the risk factors, Anthropic writes: "We have also changed, and may in the future change, usage limits, rate limits, or model and product capabilities, which could adversely affect our reputation and the perceived value of our offerings. Some of these changes have resulted, and could in the future result, in claims that our disclosures regarding model access or capabilities were inadequate or misleading."
Two points on reading it. First, this is standard risk-factor language: a company lists what could go wrong, and the sentence is not an announcement of cuts. Second, it is also a plain statement of fact about the past: limits have changed before. The same section says that concerns about model safety, interpretability, alignment and welfare may lead Anthropic to limit or delay the deprecation of older models and to keep supporting legacy versions, which creates tradeoffs in "compute distribution".
The practical consequence is architectural. If your product depends on one model at one limit, a limit change is an outage you did not plan for. Queue and retry with backoff, keep the model name in configuration, and test a second provider path before you need it. That advice is mine; the filing does not give it.
Why does the compute bill matter for developers?
Capacity is what rate limits are made of. The prospectus says that as of and after September 30, 2026, Anthropic has entered into purchase commitments and lease agreements, including short-term commitments and leases not yet commenced, with an aggregate committed spend of approximately $900.9 billion. About $451.6 billion is expected before the end of 2030, and about $70.7 billion is due in the twelve months after September 30. For comparison, the filing reports revenue of $33.2 billion for the nine months to September 30, 2026, up from $2.5 billion a year earlier, and an operating loss of $1.3 billion.
A Futurum analyst told CNBC that demand is running ahead of supply, so missing capacity turns into lost revenue very quickly. He also said a serious shortage is unlikely, because Anthropic's biggest suppliers are also its biggest investors. CNBC adds that Anthropic has said long-term capacity is significantly cheaper than short-term commitments, which it would need when demand exceeds its long-term capacity.
Which contract can already be cancelled?
The Lambda agreements are the open item. Anthropic agreed in May, June and August 2026 to spend about $41.3 billion with Lambda for access to NVIDIA GPU compute. About $34.9 billion of that is the August agreement, and the filing says: "either party may terminate if Lambda does not secure the facility lease by a specific date. Lambda did not secure the facility lease by such date and, as a result, the August 2026 agreement is currently terminable by either party. Neither party has delivered notice of termination as of the date of this prospectus, and we do not currently intend to terminate the agreement."
That contract is about four percent of the total commitments ($34.9 billion of $900.9 billion). It does not show that capacity is short. It shows one deal where a supplier missed a condition, and a company that says it does not plan to walk away.
Also today: did Claude solve two black-hole problems?
Anthropic published a research post on November 17 saying Claude solved two open problems in the physics of black holes. According to the post, a researcher asked Claude for open problems, it returned 22 candidates, and separate agents worked on them in the background for about six weeks, reading each other's progress posts. One result completes a three-loop calculation set up in 2003 and finds that the transition is abrupt at weak coupling in N=4 super Yang-Mills theory. The other finds the exact limit of the simplest modular bootstrap, about 9.13, where the best earlier proof was 8.5.
Anthropic says almost every piece of the first calculation was done twice by agents acting independently, that the results were formally verified in the Lean proof checker in September, and that a physicist checked every main result. The two preprints are on Hexagon, a repository that says it does not perform peer review, and Hexagon labels them "Primarily AI-generated text" with "Human understanding: some parts". Anthropic says neither result describes our universe directly or has practical applications. I could not find an independent expert review yet. For developers the reusable part is the harness: many agents, shared notes, every number computed twice and a checker at the end.
Also today: OpenSSL CVE-2026-121481 scores 9.8, and OpenSSL says Low
CVE-2026-121481 is an integer overflow in EVP_EncodeBlock(), OpenSSL's base64 encoder: with an input length close to the largest signed integer, the output-size calculation overflows and the function writes past the caller's buffer. The 9.8 CRITICAL score on the NVD record comes from CISA's enrichment container. OpenSSL's own CVE record rates the issue Low, says "No path by which an attacker can trigger this has been identified", and explains that its TLS, QUIC and CMS code never encode buffers that large.
The reporter says a command-line route exists. In the review of the fix pull request, reviewer Bob Beck wrote that the report's reproducer is "completely synthetic, and probably should not be a CVE without evidence it really happens". On November 20, OpenSSL team member Matt Caswell wrote that using OPENSSL_assert in the library is "a DoS waiting to happen". The pull request is open. The record lists versions before 4.0.4 and 4.1.1 as affected, and neither release has been published. CISA's own exploitation tag reads "none". If your code passes an externally controlled length to that function, bound the length before the call.
Verdict: NEEDS REVIEW
The market paid up on day one. The filing also says the capacity behind the product is still being built, that limits have moved before, and that one large supply contract can already be cancelled. I would keep building on Claude and keep a second model wired in. I gave the S-1 the same stamp in October.
FAQ
What did Anthropic's stock do on its first day?
It priced at $83.00, traded first at $111.50, reached $115.10, fell to a low of $92.72 and closed at $97.17, 17 percent above the offer price.
Can Anthropic employees sell shares right away?
The prospectus lets non-founder directors, officers and current and former employees and consultants sell up to 315 million shares from the first trading day. Most other holders are locked up for 180 or 365 days.
Does the Anthropic prospectus say rate limits will change?
It says Anthropic has changed usage limits and rate limits before and may change them in the future, and that some changes led to claims about its disclosures.
Is OpenSSL CVE-2026-121481 critical?
CISA's enrichment scores it 9.8, and OpenSSL's own record rates it Low because no attacker-controlled path to the overflow is known.
Sources
- Anthropic prices initial public offering
- Anthropic final prospectus (424B4), SEC EDGAR
- CNBC: Anthropic defies the odds
- CNBC Tech Download: compute access
- Forbes live blog
- Reuters
- Anthropic research post on black-hole physics
- Hexagon preprint 2611.00175v1 and 2611.00174v1
- CVE-2026-121481 record, NVD
- OpenSSL pull request 33652
This article expands on an episode of **The Daily Diff, a five-minute daily video on what shipped and what broke in tech.
Watch the episode · Subscribe on YouTube · the written diff lands in your inbox every morning at thedailydiff.dev.
Top comments (0)