Skip to content
Sign in
↑Part of: Anthropic warns its insurance may not cover AI-related liabilities
Technology · Global

Anthropic’s IPO filing details investor exposure and major AI infrastructure costs

Anthropic’s IPO filing discusses risks its AI models could pose to investors and reports substantial spending on computing and infrastructure. The filing says Anthropic spent $8.9 billion on compute in the first half, with less than a third used to serve customers.

AI-assisted synthesis · Based on 4 outlets' reporting · Article text read from 2 of 4 outletsHow it's made

View timeline · 9 entries
4 outletsLatest reporting 3 min read
Coverage: Center OnlyL 0 · C 4 · R 0
Updat3 graphic: Anthropic
Updat3 graphic

What happened

Key facts

  • Anthropic’s IPO filing devotes 69,800 words to AI risks to investors and says little about risks to the rest of the public (per fortune.com).
  • The filing says Anthropic’s large language models could pose “catastrophic or existential risks to humanity.” (per techbuzz.ai, tomshardware.com).
  • Anthropic’s filing says the company plans to spend $542 billion on data center infrastructure (per techbuzz.ai).
  • Five of Anthropic’s largest stated compute deals are estimated to cost $103 billion annually on average once delivered (per tomshardware.com).
  • Anthropic’s revenue was running at about $46 billion annually at Q2’s pace, while its five largest stated compute deals are estimated to cost more than twice that amount annually once delivered (per tomshardware.com).

The brief

Anthropic’s IPO filing sets out both a stark warning about AI risks and a large infrastructure spending plan, while a separate breakdown of its financial disclosures shows that most of its first-half compute spending did not go toward serving customers. The filing says Anthropic’s language models could pose “catastrophic or existential risks to humanity”; it also describes a possible “age of miracles.”

Techbuzz.ai characterizes the 335-page S-1 as a philosophical argument about humanity’s future and says the filing presents the company’s technology as capable of producing transformative benefits as well as severe harm. Fortune’s headline focuses on the balance of disclosure: it says the filing devotes 69,800 words to risks for investors while offering little about risks to the broader public. Those descriptions point to a distinction between warnings about the technology’s potential and the filing’s attention to investor-facing disclosure. The company submitted the S-1 to the SEC on Monday, according to Techbuzz.ai. Its excerpt says the filing opens with a letter acknowledging the tension in asking investors to support a technology that the company also says could pose catastrophic or existential risks. The filing’s stated infrastructure plan is $542 billion for data centers. Tom’s Hardware details the compute costs behind that plan. Anthropic spent $8.9 billion on compute in the first half, with less than one-third going to customer service; most of the rest went to training, research and running Claude internally. The outlet says cost of revenue rose about 2.3 times compared with the first half of the previous year, while revenue grew 14 times. On that basis, it estimates Anthropic retained about 85 cents per dollar after serving costs, up from 12 cents. The commitments also indicate how large future computing needs could be. Tom’s Hardware estimates that five of Anthropic’s biggest stated compute deals would average $103 billion a year once delivered, more than twice the company’s revenue at Q2’s annualized pace of about $46 billion. The filing lists at least $114.5 billion in Google commitments between April 2026 and July 2033, with additional Google and other TPU-provider commitments estimated at $125 billion to $150 billion from 2028 to 2031. The accounts emphasize different parts of the filing rather than contradicting one another: Techbuzz.ai foregrounds the company’s description of AI’s potential extremes, Fortune’s headline highlights the imbalance between investor risk discussion and risks to the public, and Tom’s Hardware examines compute expenditure and contractual commitments. Yahoo Finance’s headline also says the filing left key questions unanswered, without identifying those questions. Together, the disclosures put the filing’s public-risk language alongside large projected infrastructure costs and a compute budget largely directed to development and internal operations. The reported figures describe planned spending and estimated commitments; they do not establish that every projected contract cost has already been incurred.

Why it matters

  • Anthropic’s filing says its models could pose catastrophic or existential risks, while Fortune’s headline says the filing gives little attention to risks to the broader public compared with its 69,800 words on investor risks.
  • Anthropic’s $8.9 billion first-half compute spending largely went to training, research and internal Claude operations rather than customer service, concentrating the disclosed costs in developing and running the technology.
  • Estimated annual costs of $103 billion for five major compute deals exceed twice Anthropic’s reported $46 billion annualized revenue pace at Q2, highlighting the scale of the company’s infrastructure commitments.
  • The filing lists at least $114.5 billion in Google commitments through July 2033, alongside a further estimated $125 billion to $150 billion from Google and other TPU providers for 2028–2031.

What to watch next

  • Whether Anthropic’s planned $542 billion data-center infrastructure spending appears in subsequent disclosures as projected spending or incurred costs.
  • How Anthropic’s reported $8.9 billion first-half compute spending and its estimated 85-cent post-serving-cost retention change in later financial disclosures.
  • Whether the estimated $103 billion annual average for five large compute deals is reflected in future reported expenses as those deals are delivered.
  • How the listed Google and other TPU-provider commitments develop across the filing’s stated periods: April 2026 to July 2033 and 2028 to 2031.

How it's being covered

Coverage analysisShow details
5 source articles · 4 unique outlets
Outlet political mix: Center Only· 4 Center
Where the outlets are based: US/Canada 75% · Unknown 25%
Outlet political mix
Center Only
Left
Center
Right
0 outlets
4 outlets
0 outlets
The marker shows coverage balance: each unique outlet counts once.
Geography
Where the outlets covering this story are based.
4 unique outletsDominant coverage: US/Canada
Coverage imbalance
  • Coverage is currently concentrated among Center-rated outlets (4 of 4 rated sources); comparatively few Left or Right-rated outlets have published on the event.
  • Coverage is currently concentrated among outlets based in US/Canada (3 of 4 geographically-identified sources).
How we measure coverage

How outlets' accounts compare

3 framing differences · 2 uncorroborated claims · 1 information gap
The reports repeat the same account with different emphasis; its central claims are not independently confirmed in the reporting Updat3 read.Compared across 2 reports whose article text Updat3 read; 2 other outlets were checked by headline only.
Framing / emphasis differences
Sources emphasise different angles, actors or consequences of the same reporting.
  • Techbuzz.ai foregrounds the filing’s warnings about AI’s potential benefits and catastrophic or existential risks; Fortune’s headline emphasizes the disparity between investor-risk discussion and risks to the wider public.
  • Tom’s Hardware focuses on compute expenditures, revenue growth and provider commitments, rather than the filing’s public-risk framing.
  • Yahoo Finance’s headline says the filing left key questions unanswered but does not specify which questions.
Updat3: These are differences of emphasis, not disputes about what is reported. They do not show that the reported claims are true.
One-sided or uncorroborated claims
Claims that come from one party or one source family and are not independently corroborated.
  • Techbuzz.ai quotes the filing describing possible “catastrophic or existential risks to humanity” and an “age of miracles.”
  • Techbuzz.ai says the filing asks investors to “just believe” in AI’s potential.
Updat3: Not independently corroborated yet — treat as a claim, not an established fact.
Not established by the reporting
Information the available reporting does not contain or confirm. This is a gap, not a disagreement.
  • The sources do not specify what risks to the public the filing discusses beyond the headline’s characterization that it says little about them.
Updat3: Updat3 cannot conclude anything about this from the current sources.

Timeline

Developments from this story's sources, dated background we compiled and fact-checked, and our earlier coverage. Each row says which.
  1. The company’s emphasis on safety also had roots in the wider AI field: in May 2020, researchers including Jared Kaplan published work on scaling laws that linked model performance to computing power, data and model size.
    Background · from our historical context
  2. Anthropic was founded in 2021 by former OpenAI employees, including Dario Amodei and Daniela Amodei.
    Background · from our historical context
  3. In December 2022, the company and its researchers published “Constitutional AI,” describing a method for training AI systems using written principles and AI feedback.
    Background · from our historical context
  4. Anthropic released its Claude assistant in March 2023.
    Background · from our historical context
  5. Anthropic made Claude available to the public through Claude.ai.
    Background · compiled and fact-checked by Updat3
  6. Amazon announced an investment of up to $4 billion in Anthropic.
    Background · compiled and fact-checked by Updat3
  7. Google announced a commitment of up to $2 billion to Anthropic.
    Background · compiled and fact-checked by Updat3
  8. Anthropic introduced tool use for Claude models, allowing applications to connect them with external functions.
    Background · compiled and fact-checked by Updat3
  9. Anthropic previewed computer use in Claude 3.5 Sonnet, enabling it to interact with computers through screenshots, mouse actions and keyboard input.
    Background · compiled and fact-checked by Updat3

How we got here

Before Anthropic’s IPO filing, the company had built its business through private financing and large cloud-computing partnerships: Amazon announced a strategic investment of up to $4 billion on September 25, 2023, and Google announced a commitment of up to $2 billion on October 27, 2023. Those arrangements tied Anthropic’s growth to access to substantial computing capacity.

Anthropic was founded in 2021 by former OpenAI employees, including Dario Amodei and Daniela Amodei. In December 2022, the company and its researchers published “Constitutional AI,” describing a method for training AI systems using written principles and AI feedback. Anthropic released its Claude assistant in March 2023.

Original sources

5 articles from 4 outlets

Every story links to the reporting it was built from. How we choose sources

Spotted a factual error, a broken link or a misleading summary? Tell us.