Anthropic's IPO (ANT) Through Its Prospectus | What's Behind the Loss, the Risks, and the US Stocks It Touches
Anthropic, the developer of the conversational AI Claude, has filed to list on Nasdaq. The prospectus shows a bright side, with rapidly growing revenue and the core business turning profitable, alongside points that need care: an $80.7 billion net loss, the need to keep releasing new AI models, and an unusual voting structure. We walk through what the numbers mean and which US stocks are connected.
30-second read
The key points in 7 slides — enough to get the whole picture.
Introduction
Anthropic, which develops the conversational AI Claude, has filed to list on the US Nasdaq market. The (Form S-1) made public on October 26, 2026 shows two sides: a bright side of rapidly growing revenue and a core business that has turned profitable, and points that need careful reading, such as a huge $80.7 billion net loss and an unusual voting structure.
Based on the figures in the prospectus, this column covers the company, its results, what lies behind the loss, voting rights, its huge contracts and its risks, then looks at the US stocks connected to the IPO and what to watch after the listing. Unless otherwise noted, figures come from the prospectus (preliminary prospectus); figures from news reports are marked "according to reports."
What kind of company is Anthropic?
Anthropic (Anthropic, PBC) is a US AI company that develops the conversational AI Claude. It was founded in January 2021 by seven people, including researchers who came from OpenAI. It is headquartered in San Francisco and had about 5,300 employees as of the end of September 2026.
The prospectus describes the company as "first and foremost a research lab." A distinguishing feature is that it does not focus on image or video generation, concentrating instead on text and programming for businesses.
Main products
| Category | Products | Description |
|---|---|---|
| Mythos (including Fable), Opus, Sonnet, Haiku | Four model families at different levels of capability and price. Fable is a general-release version with added safeguards; Mythos is for vetted users | |
| Apps | Claude.ai, Claude Code, Claude Cowork and others | Chat, programming, and carrying out office tasks |
| For developers | API (Claude Platform) | Lets other companies build Claude into their own services |
| Pricing model | (API and via cloud providers) and subscription plans (Pro, Max, Team, Enterprise) | About 80% of revenue is usage-based (pay for what you use). Subscription plans, from individual to large-enterprise, are about 20% (1H 2026) |
Leadership, safety and ties to Japan
- Leadership: The CEO is co-founder Dario Amodei (formerly VP of Research at OpenAI), and the President and Chair of the Board is his sister, Daniela Amodei. The prospectus states that "all seven co-founders remain with the company."
- Approach to safety: In September 2023 the company published its "Responsible Scaling Policy," which strengthens safeguards each time AI capabilities pass certain thresholds. Its top model, Mythos, is offered only to vetted users because it could be misused for cyberattacks.
- Ties to Japan: In October 2025 the company opened a Tokyo office, its first location in the Asia-Pacific region. In 2026, NEC (April) and Fujitsu (May) each announced strategic partnerships with it.
Overview of the IPO
On October 26, 2026, Anthropic made its listing application (Form S-1) public with the US Securities and Exchange Commission (). Because it is a preliminary prospectus, the number of shares and the price are still blank.
What the prospectus tells us
| Item | Details |
|---|---|
| Exchange and ticker | Nasdaq Global Select Market, ANT |
| Shares to be listed | Class A common stock (Chapter 5) |
| Lead underwriters | Goldman Sachs and Morgan Stanley. Morgan Stanley handles price stabilization after the listing |
| Underwriters | 22 firms in total, including Japan's Mizuho and MUFG |
| Shares, price and amount raised | Not yet determined (blank) |
| Public offering in Japan | Yes. However, there are no plans to list on a Japanese exchange |
| Additional investment by Google | Has the right to buy up to $10 billion of shares at the IPO price at the time of the listing |
| Use of proceeds | Investment in compute and infrastructure, hiring and retaining talent, and other purposes |
| Dividends | Has never paid one and has no plans to for the foreseeable future |
What has been reported (not in the prospectus)
| Item | What was reported | Source |
|---|---|---|
| Timing of the listing | The week of November 16, or Thanksgiving week | The Wall Street Journal, Bloomberg |
| Valuation | $1.5–2 trillion | The Wall Street Journal, Reuters |
| Amount raised | Possibly more than $100 billion | The Wall Street Journal |
| Allocation to individual investors | Possibly 15–20% of shares allocated to individuals | Bloomberg |
If the deal is as large as reported, it would be one of the largest ever, alongside SpaceX in June 2026 (reportedly valued at about $1.77 trillion). The indicative price range is expected to appear in an amended prospectus.
Results: rapid revenue growth and a business built on new models
Revenue for 1H 2026 was $16.29 billion, about 14 times the same period a year earlier (about $1.2 billion). In just six months, the company brought in about 3.5 times its full-year 2025 revenue ($4.59 billion).
Unit: $B
Source: Anthropic prospectus (Form S-1, October 26, 2026)
Results by year
Tap a dotted term to see what it means
| Period | Revenue | Compute costs | Operating income (loss) |
|---|---|---|---|
| 2023 | $41M | $433M | ($551M) |
| 2024 | $386M | $2,522M | ($2,980M) |
| 2025 | $4,585M | $7,326M | ($8,060M) |
| 1H 2026 | $16,286M | $8,945M | ($693M) |
Figures in parentheses are losses. Operating income (loss) is revenue minus the costs of the core business. refers to the computer infrastructure, equipped with AI chips, used to develop and run AI.
Why it turned profitable: In Q2 2026, was about $970 million, the first profitable quarter among those in the prospectus. Revenue rose about 14.7 times from the same quarter a year earlier, while compute costs rose only about 4 times. Compute costs as a share of revenue fell from 251% in Q1 2025 (costs of 2.5 times revenue) to 48%.
What about Q3?: The prospectus leaves it blank. According to reports, the company told investors at private meetings that revenue was about $16.8 billion.
Customers: rapid growth among large enterprises
- Large customers: Customers that paid $100,000 or more over the past 12 months grew about 12.5 times, from over 950 a year earlier to over 11,900 (as of the end of September 2026).
- Revenue from existing customers up about 9x in a year: The metric that shows how much more existing customers are paying than a year earlier () was about 900% (excluding users of the individual Pro and Max subscription plans, comparing the latest three months with the same period a year earlier). For a typical enterprise software company, anything above 100% is considered good.
- Penetration among large enterprises: More than half of the top 500 US companies pay $100,000 or more a year. However, for these companies, payments to Anthropic average less than 0.1% of their total costs.
Revenue breakdown
| Category | 2025 | 1H 2026 |
|---|---|---|
| (pay for what you use) | $3,796M (83%) | $13,064M (80%) |
| Flat-rate | $789M (17%) | $3,222M (20%) |
| (Reference) Share of total revenue through cloud providers | About 47% | About 48% |
"Through cloud providers" means customers buying Claude on the Amazon, Google or Microsoft clouds. Usage-based revenue includes this cloud-based usage as well as businesses using the API directly.
Over 80% of revenue comes from models released within the past six months
The prospectus states that "models released from April to September 2026 accounted for more than 80% of revenue in the third quarter." A "" here is the core AI system that answers questions and writes text. Anthropic releases new, more capable models every few months. In other words, most of its revenue comes from models released within the past six months. To use a smartphone analogy, it is as if more than 80% of sales came from handsets launched in the last six months.
Other points of note
- Customer concentration: In the first nine months of 2026, no single customer accounted for more than 10% of total revenue. However, the top five customers made up about 25% of total revenue.
- Revenue from users violating the terms: The company estimates that about 5–7% of revenue in the first nine months of 2026 came from users who violated its usage policies. If that usage is stopped, this revenue could fall away.
- : The difference between the cash that actually came in and went out of the core business. It was an outflow of $2.64 billion in 2025 but turned into an inflow of $3.93 billion in 1H 2026.
- : $482 million in 1H 2026, small relative to the scale of revenue. This is because the company has not bought AI chips or data centers itself; it has mainly rented them from the large cloud providers and paid usage fees. Those payments are recorded not as capital expenditure but as an expense each period (the "compute costs" in the table above; $8.95 billion in 1H 2026). Going forward, it will also expand an approach (Anthropic Direct) in which it leases AI chips directly and places them in dedicated data centers run by third parties. Because this approach requires large payments before the compute becomes available, the company expects capital expenditure to rise significantly.
What lies behind the $80.7 billion loss
The net loss for 1H 2026 was $80.75 billion, about five times revenue. However, about 90% of it, $72.39 billion, is an accounting valuation loss that involves no cash outlay. Most of the loss is not the result of ballooning AI investment. The core business () lost about $700 million over the half, combining a Q1 loss (about $1.66 billion) and a Q2 profit (about $970 million). Looking at Q2 alone, the core business was profitable.
| Period | Operating income (loss) | Valuation losses on convertible notes, etc. | Net loss |
|---|---|---|---|
| 2023 | ($551M) | ($1,045M) | ($1,531M) |
| 2024 | ($2,980M) | ($5,434M) | ($8,309M) |
| 2025 | ($8,060M) | ($34,308M) | ($41,965M) |
| 1H 2026 | ($693M) | ($72,388M) | ($80,748M) |
The gap between the three columns is mainly "other income and expense." For 1H 2026 it includes an expense of about $8.8 billion from marking share purchase agreements with Google and Amazon and similar items to fair value (also non-cash), and about $1.1 billion of interest income.
Why the valuation losses arise
Anthropic has raised money from Amazon and others in the form of . A convertible note is a loan that can later be converted into shares at a set price.
Suppose the company agrees that a note can be converted at $2 per share, and the shares then rise in value to $60 each. The lender now holds "the right to get something worth $60 for $2." From the company's point of view, the value it will hand over in future (a liability) has increased, and it records that increase as a loss.
The prospectus likewise explains that the loss is "primarily due to an increase in the of the convertible notes as a result of the increase in the company's valuation." It is a counterintuitive mechanism: the more highly the company is valued, the larger the loss.
As of the end of June 2026, the principal of the convertible notes held by Amazon was $6.20 billion (non-interest-bearing). Their fair value on the balance sheet, meanwhile, had swelled to $108.57 billion. Anthropic has no bank borrowings.
Will the valuation losses continue after the IPO?
- The convertible notes automatically convert into shares (non-voting Class N shares) at the IPO. Valuation losses stop being recorded on the converted portion.
- However, there is a cap that keeps Amazon's ownership from exceeding 12% (Amazon may choose to waive it), and any amount above the cap is exchanged into new notes that remain outstanding.
- The remaining notes convert quarterly after the IPO, and valuation gains and losses continue to be recorded until then. As a result, net income could swing sharply with the share price after the listing.
Don't look only at "adjusted profit"
The company reports "adjusted net income," which excludes valuation losses, stock-based compensation and other items, of $2.67 billion for 1H 2026. However, this is a measure the company defines itself ().
The prospectus states plainly that the company "expects expenses to continue to increase and losses to continue."
An unusual voting structure
Even as Anthropic shareholders, ordinary investors will have little influence over the company's direction. An entity owned by the seven founders holds 50.1% of total voting power, and a majority of the board is chosen by a trust made up of outside experts.
Share classes and voting rights
| Share class | Voting rights | Main holders |
|---|---|---|
| Class A | One vote per share (all votes other than Class F total about 49.9%) | Public investors, employees, founders (the only class to be listed) |
| Class N | None (except as required by law) | Strategic partners such as Amazon and Google |
| Class F (1 share) | 50.1% of total voting power | The founders' entity (Founder LLC) |
| Class T (1 share) | Elects 4 of the 7 directors | Long-Term Benefit Trust (LTBT) |
Four of the seven directors are elected by the (LTBT), and the remaining three are elected jointly by Class A and Class F shareholders. The single Class F share carries votes equal to 50.1% of total voting power, and its vote count rises automatically when new shares are issued, so its share of voting power is never diluted. In practice, this means the founders' wishes carry strong weight. In addition, Class E shares for employees (excluding the founders) will be issued after the IPO, but they can vote only on limited matters such as changes to the trust arrangements.
The Long-Term Benefit Trust (LTBT)
Established in 2023, it is a body independent of the company. Its trustees are meant to have no financial interest in the company and are experts in AI safety, national security, public policy and related fields. Former Fed Chair Ben Bernanke became a trustee in June 2026. The trust can elect and remove directors but has no authority over day-to-day management.
What is a public benefit corporation (PBC)?
Anthropic's legal name is "Anthropic, PBC." PBC stands for "," a corporate form set out in the Delaware General Corporation Law (DGCL). A PBC is a for-profit company that issues shares, much like an ordinary corporation. It can make a profit and go public, and it is different from a nonprofit. The difference is that the law requires its directors to consider not only shareholders' interests but also the public benefit purpose written in its certificate of incorporation (the company's founding charter).
| Item | Ordinary corporation | Public benefit corporation (PBC) |
|---|---|---|
| Issuing shares and listing | Allowed | Allowed |
| What directors must weigh | Mainly shareholders' interests | A balance of three things: (1) shareholders' financial interests, (2) the interests of those materially affected by the company's conduct, and (3) the public benefit purpose in its charter |
| Public benefit reporting | Not required | Must report to shareholders on progress toward its public benefit purpose at least every two years (DGCL Section 366) |
- Anthropic's public benefit purpose: The purpose stated in its charter is "the responsible development and maintenance of advanced AI for the long-term benefit of humanity."
- What it means for investors: The prospectus says that so far, attention to safety and society has largely aligned with long-term commercial success. At the same time, it states that decisions striking this balance may sometimes hurt results or the Class A share price. For example, its usage policies prohibit use for fully autonomous weapons or mass surveillance, even where that would generate revenue.
- How it differs from other tech giants: Structures that give founders strong voting rights also exist at Meta and Alphabet. What sets Anthropic apart is that, in addition, an outside trust that holds no equity elects a majority of the board, and a public benefit purpose is written into its charter.
Huge contracts and the circular flow of money
Developing and running AI requires enormous amounts of . To secure it, Anthropic has signed long-term contracts with the large cloud providers and others.
Major contracts (per the prospectus)
| Counterparty | Committed payments | Term | Can it be cancelled? |
|---|---|---|---|
| About $114.5 billion or more (shortfall paid even if usage falls short) | 2026–2033 | Generally no | |
| Amazon | $110 billion or more (shortfall paid even if usage falls short) | 2026–2036 | Generally no |
| Microsoft | $44.9 billion or more | 2026–2033 | Generally no |
| Broadcom | Lease payments for TPU systems of about $161.2 billion | 5 years | Generally no |
| SpaceXAI | Up to $120.5 billion (about $3.1 billion a month once fully operational) | Through 2029 | From January/March 2027, on 90 days' notice |
| AMD | Expected to exceed $20 billion | From 2027 | Amount is an estimate |
The amounts committed to Google, Amazon and Microsoft alone total about $269.4 billion. That is about 16 times Anthropic's 1H 2026 revenue (about $16.3 billion) and about twice its cash and short-term investments (about $131.7 billion at the end of June 2026). However, the payments do not fall due at once; they are spread over a long period running to 2033–2036.
The prospectus lists several figures, including contractual commitments (about $265.6 billion at the end of June 2026), leases that have not yet commenced (about $321.4 billion) and new contracts signed from July onward (about $245 billion). These may overlap, however, and the "total" line in the prospectus is blank. Note that the "$X billion in total" figures seen in news reports can vary widely depending on how they are calculated.
Using four kinds of chips
| Chip | Developer | Main sources |
|---|---|---|
| Trainium | Amazon | Amazon (AWS) |
| TPU | Google (with Broadcom) | Google; leased systems from Broadcom |
| GPU | NVIDIA | Microsoft, SpaceXAI, Nscale, Lambda |
| GPU (MI450) | AMD | AMD (from 2027) |
By not relying on any single supplier, the company limits the risk of shortages or a weak position in price negotiations.
When funders are also suppliers
The prospectus states explicitly that Amazon, Google and Microsoft are each Anthropic's "investor, customer, distribution channel and competitor." "Distribution channel" refers to the three companies selling Claude to businesses through their own clouds. As noted in Chapter 3, about half of revenue comes through these cloud providers.
These are not improper transactions, and the contract terms are disclosed in the prospectus. However, if growth in the AI market slows, it could simultaneously affect both Anthropic's payment burden and the cloud providers' revenue and investment gains.
US stocks affected by the IPO and their latest results
Anthropic shares cannot be bought until the listing, but the companies on the receiving end of its huge spending are already public. The prospectus lists investment in compute and infrastructure as one use of the IPO proceeds (Chapter 2), and more cash on hand after the IPO strengthens its ability to keep paying these companies. On the other hand, once Anthropic's shares trade daily, the profits of companies that hold its stock will be affected by moves in its share price in both directions.
Key stocks and recent results
| Company | Relationship with Anthropic | Latest results |
|---|---|---|
| Alphabet (GOOGL) | Major shareholder, provider of cloud and TPUs, competitor | Q3 2026: revenue $127.6 billion (+25%), operating income +30% |
| Amazon (AMZN) | Major shareholder, convertible note holder, cloud provider, competitor | Q3 2026: revenue $201.9 billion (+12%), AWS $46.0 billion (+39%), operating income $28.1 billion |
| Microsoft (MSFT) | Shareholder and cloud provider. Sells Claude to businesses through its AI service on Azure (Microsoft Foundry) and builds it into its own products such as Microsoft 365 Copilot | July–September 2026 quarter: revenue $92.3 billion (+19%), Azure +48%, net income +32% |
| Broadcom (AVGO) | Lessor of TPU systems, provider of funding | May–July 2026 quarter: revenue $29.6 billion (+86%) |
| AMD (AMD) | Provider of AI systems (from 2027) | Q2 2026: revenue $11.5 billion (+50%) |
| NVIDIA (NVDA) | GPU developer. Anthropic mainly rents compute that Microsoft, SpaceXAI and others have built with NVIDIA GPUs (direct payments to NVIDIA are not stated in the prospectus) | May–July 2026 quarter: revenue $96.2 billion (+106%) |
| Micron (MU) | One of the suppliers of high-speed memory for AI | June–August 2026 quarter: revenue $54.2 billion (about 4.8x the prior year) |
Figures in parentheses are year-over-year changes.
Alphabet: the core business is strong, but cash is flowing out to capex
- Q3 net income fell 24.3% year over year. The cause, however, was valuation losses on equity holdings (about $8.7 billion, mainly SpaceX shares). Operating income from the core business grew 30%.
- Google Cloud revenue was $30.7 billion (+103%), with a backlog of $561 billion.
- On the other hand, (cash generated by the business minus capital expenditure) was about –$9.1 billion, negative for a second straight quarter. Capital spending on AI is exceeding the cash the company generates.
- On October 27, Oracle announced that it will offer Google TPUs on its cloud from 2027. TPUs are starting to spread beyond Google.
- Its own AI app, Gemini, has reached 1 billion users, making Alphabet one of Anthropic's biggest competitors as well.
Microsoft: a "multi-model strategy" offering several AIs
- On Azure and Microsoft 365, customers can choose AI from several companies, including OpenAI and Anthropic. Whichever AI wins, Microsoft earns as long as its cloud is used.
- It has kept free cash flow positive while continuing heavy capital spending.
- It is also a major shareholder in OpenAI, giving it business relationships with both companies.
Amazon: the value of its Anthropic stake moves its profit
- Q2 2026 net income was $62.6 billion, of which $53.4 billion in non-operating income was reported to be "primarily from its investment in Anthropic."
- In Q3, AWS revenue grew 39% and operating income rose to $28.1 billion (from $17.4 billion a year earlier). Non-operating items, however, were a loss of about $1.2 billion, and net income was $20.7 billion (down 2% year over year). This is because the prior-year profit included a valuation gain of about $9.5 billion on its Anthropic investment.
- As this shows, the valuation of its Anthropic stake moves Amazon's net income significantly. After the IPO, a fall in Anthropic's share price could also weigh on Amazon's profit.
Points to keep in mind
Annual averages and percentages are estimates by the Beikabuzukan Editorial Team. Actual payments are not necessarily even from year to year, and lease payments begin only once the systems become available.
Key risks described in the prospectus
The "Risk Factors" section of the prospectus runs from page 26 to about page 120, roughly 94 pages. It is where the company itself sets out what could hurt its results or share price, and it includes items characteristic of a pure-play AI company.
| Area | Key points in the prospectus |
|---|---|
| AI itself | There are concerns that advanced AI, if deployed without adequate safeguards, "could pose catastrophic or existential risks to humanity." Risk rises if AI automates AI research and development and safety evaluations cannot keep up |
| Insurance | In the event of a major incident, losses "may exceed the capacity not only of a single insurer but of the insurance industry as a whole" |
| Lab equipment and robots | Risk of accidents when AI operates lab equipment or robots. The company also runs its own molecular biology lab |
| Competition | (1) At the frontier, it names Google and OpenAI as its main competitors, saying they exceed it in funding, compute and brand recognition. (2) Competition from low-cost AI: "open-weight" models, whose trained weights are published for anyone to use freely, could win customers in cost-focused uses and put pressure on prices. (3) The performance gap could narrow through "distillation," in which others collect large volumes of Claude's outputs and use them to train their own AI cheaply. The prospectus says such attempts have actually occurred |
| Dependence on others | Smartphone and PC operating systems and app stores are run by other companies such as Apple and Google. If they change their rules or favor their own AI, it could become harder for users to obtain and use the Claude apps |
| Compute | (1) Supply sources are limited. Most of the world's production of leading-edge AI chips is concentrated in Taiwan (mainly TSMC), and a production halt from an earthquake or Taiwan–China tensions could have effects lasting years. High-speed memory for AI is also mostly made by three companies: SK hynix, Samsung Electronics and Micron. (2) Required capacity must be forecast and contracted years in advance and is hard to change later. The prospectus says the company has in the past significantly underestimated its needs, leading to compute shortages that affected product development and delivery |
| Government and regulation | A dispute with the US Department of Defense (which began with the company's refusal to allow use for mass surveillance or fully autonomous weapons). In June 2026 it temporarily suspended its top models due to US Commerce Department export controls |
| Copyright and litigation | Settled a copyright infringement class action for $1.5 billion. It has also been sued by music publishers and by the discussion site Reddit |
| Shareholder rights and dilution | Besides the concentration of voting power, there are many sources of future share issuance that could dilute value, including convertible notes, partners' funding facilities, additional contributions to a public benefit fund and employee stock compensation |
The structure noted in Chapter 3, in which over 80% of revenue comes from models released within the past six months, is also worth considering alongside the competitive risks.
Market backdrop: midterms, the FOMC and the backlash against data centers
Anthropic's IPO is scheduled around the time of the US midterm elections (November 2026). Let's also check the market environment surrounding AI-related stocks.
How stocks have behaved after midterms
Since World War II, the has risen in the 12 months after every midterm election. The usual explanation is that pre-election uncertainty clears. However, this is a past tendency, and there is no guarantee it will repeat this time.
The backlash against data centers
As construction of AI data centers spreads, opposition to new projects is growing, driven by discontent over rising electricity bills and other issues.
| Group surveyed | Opposition to a data center being built nearby |
|---|---|
| All respondents | 71% opposed (48% strongly opposed) |
| Democrats | 56% strongly opposed |
| Republicans | 39% strongly opposed |
Source: Gallup (survey conducted March 2026). Figures by party are the share "strongly opposed"; for independents it was 48%. Gallup notes that a majority of every partisan group is opposed.
Regulatory moves have already appeared (the case of Maine): In April 2026, the legislature of Maine, in the northeastern US, passed a "moratorium bill" that would have barred data center construction for a set period. Concerns about environmental impact and rising electricity bills were behind it. The bill did not become law because Governor Mills vetoed it. The governor did not, however, oppose a moratorium as such. In the town of Jay, there was a plan to build a data center on the site of a closed mill, which the community supported for the jobs it would bring. The governor vetoed the bill because it did not exempt that project, and said she would have signed it with an exemption.
In other words, both the legislature and the governor were open to halting construction. If similar rules spread to other states, data center construction would be delayed. That could affect the large cloud providers' capital spending and Anthropic's plans to secure compute.
Monetary policy (FOMC)
- At the October 28, 2026 meeting, the target range for the policy rate was held at 3.75–4.00%.
- At the press conference the same day, Fed Chair Warsh announced changes to the meeting structure. From next autumn there will be seven rate-setting meetings and one meeting to discuss strategy; in 2028, six rate-setting meetings and two strategy meetings, for eight meetings a year in both cases.
- Rates that stay high are generally seen as a headwind for growth stocks, which are bought in anticipation of future profits.
Japanese suppliers in the AI data-center build-out
Japan has no major listed company at the center of large-scale AI model development. However, when Anthropic and the large cloud providers spend heavily on AI data centers, demand also flows through to the Japanese component and equipment makers that support that infrastructure. None of them is named in the prospectus as a direct counterparty, but within the broader trend of expanding AI data-center investment, the following areas are drawing attention.
| Area | Key Japanese companies (TSE code) | Connection to AI data centers |
|---|---|---|
| Memory | Kioxia Holdings (TSE: 285A) | NAND flash memory for the storage devices (SSDs) that hold data in AI servers |
| Semiconductor test and manufacturing equipment | Advantest (TSE: 6857), Tokyo Electron (TSE: 8035), Disco (TSE: 6146) | Equipment to make and test AI chips and high-speed memory () |
| Semiconductor materials | Ibiden (TSE: 4062) | Substrates on which AI chips are mounted |
| Communication cables | Fujikura (TSE: 5803), Furukawa Electric (TSE: 5801) | Optical fiber connecting equipment inside and between data centers |
| Power equipment | Hitachi (TSE: 6501), Mitsubishi Electric (TSE: 6503) | Power transmission and distribution grids; uninterruptible power supplies (UPS) that keep power flowing during outages |
Note that the three companies the prospectus names as suppliers of high-speed memory (HBM) (SK hynix, Samsung Electronics and Micron) do not include any Japanese company. Japanese companies are involved through the equipment and materials used to make HBM. Their results also depend on businesses outside AI and on the broader economy, so Anthropic's IPO alone will not determine their share prices.
Some argue that ahead of a large IPO, investors sell other stocks to raise money to buy in, but we have not found data showing that such selling is actually occurring in Japanese stocks.
What to watch after the listing
Selling restrictions (lock-ups) lift in many small stages
A "" is an arrangement under which executives, major shareholders and others who held shares before the IPO agree not to sell for a set period after the listing. The usual period is 180 days from the listing, but Anthropic will lift its restrictions in many separate steps.
| Who | How the restrictions lift (summary) |
|---|---|
| Directors, officers and some employees other than the founders | Up to 10% may be sold from the first day of trading; up to 45% cumulatively after full-year 2026 results are released |
| Certain shareholders (180-day group) | A few percent released at a time, such as on days 27, 48, 62 and 76 after the listing. Fully released at the earlier of the release of Q1 2027 results or 180 days |
| Large shareholders (365-day group) | Released in very small portions; fully released at the earlier of the release of Q3 2027 results or 365 days |
| Founders | In principle, may not sell or transfer shares until the earlier of the release of Q3 2027 results or 365 days after the listing. However, from the earlier of the release of Q1 2027 results or 180 days after the listing, they may sell or otherwise dispose of up to 29% of their holdings |
Detailed percentages and dates are based on the "Underwriting" section of the prospectus and may change in an amended prospectus.
Because the releases are split into many steps, a single "wall" of shares hitting the market at once is less likely, but selling could continue throughout the first year after the listing. In addition, the portion of Amazon's convertible notes above its ownership cap will keep converting into shares each quarter.
Why a "10x" is hard
If the valuation at listing is $1.5–2 trillion as reported, a tenfold rise in the share price would require a market capitalization of $15–20 trillion. That is not impossible, but it is far harder than for a small company. When following the share price after the listing, it is important to check whether revenue growth justifies the valuation.
What to check before the IPO price is set
- The "indicative price range" in the amended prospectus, and the valuation implied by it
- Official results for Q3 2026
- The securities registration statement for the public offering in Japan, and which brokers will handle it
Frequently asked questions
Q. When will Anthropic list?
The prospectus does not give a listing date. Reports have pointed to the week of November 16, 2026, or Thanksgiving week (late November).
Q. Can investors in Japan buy the shares?
After the listing, the shares will typically be available through Japanese brokers that handle US stocks. The prospectus says the offering "includes a public offering in Japan," but which brokers will accept applications has not yet been announced.
Q. Why is a loss-making company attracting so much attention?
Because revenue is growing extremely fast and operating income from the core business has turned positive. Most of the huge net loss is an accounting valuation loss that involves no cash outlay (Chapter 4).
Q. Will it pay dividends?
It has never paid a dividend and has no plans to for the foreseeable future.
Q. Which US stocks will the IPO affect?
Alphabet and Amazon, which are both major shareholders and compute providers; Microsoft, which provides cloud services; Broadcom, which leases TPU systems; and others. For Amazon and Alphabet in particular, changes in the valuation of their Anthropic stakes affect their own profits (Chapter 7).
Q. Could the share price move sharply right after the listing?
In general, large IPOs tend to see big price moves right after listing. In Anthropic's case, some executives and employees can sell from the first day of trading, and selling restrictions will continue to lift in stages after that (Chapter 10).
Summary
- The prospectus shows a bright side: rapid revenue growth (about 14x year over year in the first half) and the core business turning profitable in Q2 2026.
- About 90% of the $80.7 billion net loss is a valuation loss on convertible notes that grows as the company's valuation rises, and no cash goes out. What matters is what the number consists of, not its size.
- On the other hand, there are points to watch: a structure in which over 80% of revenue comes from models released within the past six months, huge long-term contracts with the large cloud providers and others, and limited voting rights for ordinary shareholders.
- Anthropic shares cannot be bought until the listing, but Alphabet, Amazon, Microsoft, Broadcom and others on the receiving end of its spending are already public. How much these contracts matter to each company's results varies, however.
How to look up related stocks on Beikabuzukan
Glossary
- Prospectus
- The document a listing company files to disclose its business, results and risks to investors. A version with the share count and price still blank is a preliminary prospectus.
- AI model
- The core of an AI system that answers questions and writes text or code. It is built by training on large amounts of data, and improved versions are released every few months.
- Usage-based pricing
- A pricing model where customers pay according to how much they use.
- SEC
- The US government agency that sets and enforces the disclosure rules for publicly listed companies.
- IPO
- When a private company lists its shares on a stock exchange so that anyone can buy and sell them.
- Compute
- The computing infrastructure, built from AI chips, used to develop and run AI models.
- Operating income
- The profit earned from a company's core business. It is calculated by subtracting cost of goods sold, SG&A, and other operating expenses from revenue.
- NRR
- A metric showing what percentage existing customers are paying now compared with one year earlier. An NRR of 110% means revenue grew 10% even without any new customers.
- Subscription
- A fixed-fee pricing model in which customers, once they adopt a product, keep using it by renewing their contract each year. It gives the provider recurring revenue year after year.
- Operating cash flow
- The difference between cash actually coming in and going out from the core business.
- Capital expenditures (CapEx)
- Spending on things like factories, machinery, stores, and IT systems.
- Convertible note
- Debt that can later be converted into shares on preset terms. The higher the company’s share value, the more the lender’s right is worth.
- Fair value
- Restating a balance-sheet item to its current value each period, with gains or losses booked to the income statement.
- Non-GAAP earnings
- A company's own adjusted profit figure that excludes items such as stock-based compensation for employees. Some companies that report a loss under GAAP (the official US accounting standards) show a profit on this basis.
- Stock-based compensation
- Paying part of employees’ pay in company shares instead of cash. No cash leaves, but the share count rises, so it is a real cost to shareholders.
- RSU
- A right for an employee to receive company shares once set conditions, such as a period of service, are met.
- Dilution
- When new shares are issued and existing shareholders’ per-share value or ownership shrinks.
- Long-Term Benefit Trust
- An independent trust that elects a majority of Anthropic’s board. Its trustees have no financial stake in the company.
- Public benefit corporation
- A for-profit company legally required to weigh a public benefit stated in its charter alongside shareholders’ interests, as provided in Delaware law.
- Free cash flow
- Operating cash flow minus capital expenditures — the cash a company is free to use.
- S&P 500
- A stock index made up of about 500 leading large US companies. It is commonly used as a gauge of the overall US stock market.
- FOMC
- The Federal Reserve meeting that sets US policy interest rates.
- HBM
- Fast memory made by stacking memory chips and placing them right next to an AI chip so huge amounts of data can move at once. Essential for AI computing; SK hynix, Samsung Electronics and Micron make most of it.
- Lock-up
- A promise by pre-IPO shareholders not to sell their shares for a set period after the listing.
Sources
- Anthropic, PBC Form S-1 (preliminary prospectus, filed October 26, 2026, SEC EDGAR)
- Anthropic official website: opening of the Tokyo office (October 2025)
- NEC and Fujitsu: announcements of partnerships with Anthropic (April and May 2026)
- Oracle: announcement of Google TPUs on OCI (October 27, 2026)
- Amazon.com: Q3 2026 results (Form 8-K, Exhibit 99.1)
- Earnings releases from Broadcom, AMD, NVIDIA and Micron
- Federal Reserve: FOMC statement of October 28, 2026
- Listing timing, valuation and amount raised: The Wall Street Journal, Reuters, Fortune, Bloomberg
- Coverage of Alphabet, Microsoft and Amazon earnings: 9to5Google, SiliconANGLE, Fortune, The Star, GeekWire
- Public opinion on data centers: Gallup (survey conducted March 2026, as reported by The Washington Times)
- Maine governor's veto: Gizmodo
- Changes to the FOMC meeting schedule: Seoul Economic Daily
This column is based on the preliminary prospectus Anthropic filed on October 26, 2026 and on publicly available news reports. It is not a recommendation to buy or sell any particular security. Please make your own investment decisions after reviewing the latest prospectus and materials from your broker.