OpenAI $25B, Anthropic $30B: The AI Revenue Race in Numbers

What $25 Billion in Annualized Revenue Actually Means

When The Information reported on March 4–6, 2026 that OpenAI had crossed $25 billion in annualized revenue — up from $21.4 billion at end-2025 — it was not reporting an audited GAAP figure. It was reporting an annualized run rate (ARR): take the most recent month's revenue, multiply by twelve. That distinction matters more than it might seem, especially for a company that disclosed it will lose roughly $14 billion in 2026 against that same revenue base.

ARR is a forward projection, not an income statement line. It captures momentum brilliantly and hides structural costs just as well. OpenAI's $25 billion ARR represents roughly $2 billion a month in bookings across ChatGPT subscriptions, the API, and enterprise contracts — a real number, but one that Bloomberg confirmed sits alongside first-quarter 2026 losses where the company was spending $1.22 for every $1 it earned. Keep that arithmetic close at hand as you read what follows.

Still, the scale is historically unusual. Snowflake, the largest pure-software IPO ever when it listed in September 2020, carried an ARR of roughly $500 million at the time of its debut. Databricks, expected to go public in late 2026, is approaching a $5.4 billion ARR. OpenAI at $25 billion would list as something with no modern precedent in enterprise software — closer in scale to a mid-sized bank than a SaaS startup.

---

The Valuation Stack: Two Companies, Very Different Stories

Before comparing revenue trajectories, it helps to see where both companies sit in terms of capital structure and investor pricing.

Company	Reported ARR (latest)	Last Disclosed Valuation	Ownership Structure	IPO Status
OpenAI	~$25B (The Information, Mar 2026)	$852B (Bloomberg, Mar 2026 round)	Public Benefit Corporation; nonprofit foundation holds 26%; Microsoft ~27%	Confidential S-1 filed May 22, 2026 (CNBC); September 2026 target, per reports
Anthropic	~$30B (Bloomberg / VentureBeat, Apr 2026)	$965B (Axios, May 28, 2026 round)	Private; Amazon $13B+ committed; Google/Alphabet ~14% stake	No public filing; reports point to late 2026 or early 2027
Google DeepMind	Not separately disclosed	Subsidiary of Alphabet (~$2T market cap)	Wholly owned by Alphabet	N/A — no separate listing
Meta AI	Not separately disclosed	Subsidiary of Meta Platforms (~$1.5T market cap)	Wholly owned by Meta	N/A — no separate listing

A few important caveats on that table. The ARR figures for OpenAI and Anthropic come from media reporting — The Information, Bloomberg, VentureBeat — not from audited filings or official press releases. OpenAI has not published GAAP revenue for 2026; its S-1, filed confidentially with the SEC, will remain sealed until roughly 15 days before any roadshow. Anthropic is private and has made no public disclosure of financial statements. Google DeepMind and Meta AI do not break out standalone AI-product revenue, though both companies have disclosed sharply rising AI-related capital expenditure: Alphabet guided to $75 billion in capex for 2026 in its Q1 earnings, and Meta raised its 2026 capex forecast to as much as $145 billion.

---

Anthropic's Velocity Problem (for OpenAI)

At the time The Information published its March 2026 OpenAI ARR figure, Anthropic was approaching $19 billion — a number Bloomberg attributed to CEO Dario Amodei at a Morgan Stanley TMT conference. That gap of roughly $6 billion looked stable enough. It did not remain so.

By April 2026, Bloomberg and VentureBeat reported Anthropic had crossed $30 billion in run-rate revenue. The company's own internal forecasts, which initially projected $9 billion for all of 2025, were off by a factor of eight according to Amodei. The driver was Claude Code: Reuters reported that the agentic coding product's run-rate had climbed past $2.5 billion by February 2026, and business subscriptions quadrupled between January and April. SaaStr noted in April that Anthropic had more than 1,000 business customers each spending over $1 million annually, a figure that had doubled since February.

What drove the acceleration

Three factors compound each other. First, the shift from consumer chatbot to developer infrastructure: when a software team embeds Claude in a CI/CD pipeline or an IDE, usage scales with code commits rather than with marketing budgets. Second, Amazon Bedrock: with Amazon having committed over $13 billion to Anthropic and Bedrock serving as a distribution channel, more than 100,000 companies were running Claude on AWS as of April 2026. Third, enterprise pricing tolerance: API calls are priced per token, and production workloads generate token volumes that make consumer subscriptions look rounding errors.

OpenAI is not standing still — enterprise contracts now represent over 40% of its revenue, on track to reach parity with consumer by end-2026. But the velocity gap is real. Anthropic went from $9 billion to $30 billion ARR in roughly four months. That is the growth rate that has rattled the market assumptions that OpenAI is the unchallenged monetization leader.

---

The IPO Equation: OpenAI's Structural Peculiarities

OpenAI filed its confidential S-1 with the SEC on May 22, 2026, according to CNBC, with Goldman Sachs and Morgan Stanley leading the deal. The company is reportedly targeting a listing in September 2026 at a valuation between $852 billion and $1 trillion. None of that is confirmed by OpenAI itself; the company has not commented publicly on the filing or timeline. Treat "September listing" as analyst and press conjecture, not corporate guidance.

The corporate structure question

What makes an OpenAI IPO genuinely novel is the governance architecture it completed in October 2025. The company converted from its prior nonprofit-controlled hybrid structure to a Public Benefit Corporation (PBC), with the OpenAI Foundation retaining a 26% equity stake valued at approximately $130 billion, plus warrants tied to valuation milestones. Microsoft holds roughly 27%. Sam Altman is reportedly not taking personal equity in the new structure, an unusual arrangement that investors and proxy advisers will scrutinize closely once the S-1 is public.

The PBC structure requires the company to consider "broader interests of all stakeholders" — a clause that gives the board legal grounds to weigh factors beyond shareholder returns. Whether public markets will price that feature or discount it remains genuinely unclear.

The loss problem

OpenAI's own internal projections, reported by Fortune in late 2025, showed the company expecting $44 billion in cumulative losses through 2028, before turning profitable around 2029–2030. In Q1 2026, it was losing $1.22 per dollar of revenue earned. The $122 billion funding round closed in March 2026, anchored by $50 billion from Amazon, $30 billion from Nvidia, and $30 billion from SoftBank at an $852 billion valuation (Bloomberg). That capital is intended to fund the compute buildout, but with an estimated $207 billion in additional capital required through 2030 to honor existing infrastructure commitments, the IPO is partly a capital-raising exercise as much as a liquidity event.

CFO Sarah Friar has reportedly flagged internally that the timeline between revenue growth and capital requirements is tight. That tension will be the central question on the roadshow.

---

The India Angle: Partner, Competitor, Dependent

Indian IT services firms are in an awkward position that the AI revenue race makes more acute. TCS, Infosys, and Wipro are simultaneously partners of and competitors to OpenAI and Anthropic.

TCS has formal agreements with OpenAI, Google Cloud, Nvidia, and Microsoft. Infosys has tied up with both Anthropic and OpenAI, embedding their models in client AI implementations. TCS reported over $2.3 billion in annualized AI services revenue in Q1 2026 — roughly 7.5% of its total — up from $1.8 billion the previous quarter. Infosys told investors that AI work now accounts for 5.5% of total revenue and touches 90% of its large corporate clients.

The complication: Anthropic is now expanding into enterprise AI implementation services, not just foundation model APIs. A May 2026 analysis in Business Standard and People Matters noted that as Anthropic moves downstream into delivery, it edges into territory that Infosys and TCS currently occupy. The scenario is not hypothetical — Amazon's $13 billion-plus Anthropic investment includes commercial milestones tied to AWS-delivered AI services, meaning Amazon and Anthropic together have an incentive to compete with the Indian IT integrator layer.

Equity exposure

Indian retail and institutional investors have no direct way to own OpenAI or Anthropic equity today. The indirect routes — Infosys and TCS as partners; Alphabet (Google) listed on NSE as GDR; SoftBank Vision Fund, which backs OpenAI, listed in Tokyo — are all proxies, not direct exposure. If OpenAI lists in September 2026, it would become available through US brokerage accounts and potentially through LRS (Liberalised Remittance Scheme) for Indian investors, within the $250,000 annual limit. Whether the IPO price at listing reflects or discounts the current $852 billion private valuation is a separate question entirely.

---

Comparing Scales: Why These Numbers Have No Historical Peer

To understand why the AI revenue race is reshaping IPO expectations, the comparison to prior tech listings is useful.

Snowflake went public in September 2020 at a market cap of roughly $33 billion with an ARR around $500 million and a revenue multiple of about 66x. Databricks, still private, carries a $5.4 billion ARR and a $134 billion private valuation — roughly 25x ARR. These are high-growth SaaS multiples by any historical standard.

OpenAI at a $1 trillion IPO valuation against $25 billion ARR would imply a 40x multiple. Anthropic at $965 billion against $30 billion ARR implies roughly 32x. Neither is absurd given the growth rates involved, but both depend on a credible path to gross margin expansion and eventual profitability that neither company has yet demonstrated at the reported scale. OpenAI's compute costs are structurally heavy; Anthropic's hardware agreements with Broadcom and Google TPUs are designed to reduce per-inference costs but have not yet shown up as margin improvement at this revenue level.

The relevant comparison is not Snowflake or Databricks. It may be Amazon in 2001: a company with enormous revenue, real losses, and a long horizon to unit-economic sanity. The outcome there was eventual dominance, but the path required investors to hold through years of negative returns.

---

What to Watch

• OpenAI S-1 publication. The confidential filing becomes public approximately 15 days before any roadshow. When that document drops, it will contain the first GAAP revenue disclosure, gross margin data, and actual loss figures. Everything before that is third-party reporting, including this article.

• Anthropic's revenue velocity in Q2 2026. The company reportedly crossed $44 billion ARR by May 2026 according to some sources, and reports suggest it may post its first operating profit in Q2 2026. If confirmed, that changes the profitability narrative for the entire AI sector.

• Amazon's $35 billion contingency. Of its $50 billion commitment to OpenAI, $35 billion is contingent on OpenAI going public or achieving AGI milestones. The IPO is partly structured to unlock that capital.

• Microsoft's stake dilution mechanics. Microsoft holds roughly 27% of the PBC. How that stake behaves through an IPO lockup and secondary market will matter to anyone modeling MSFT.

• Indian IT competitive pressure. Watch TCS and Infosys quarterly earnings calls through 2026 for language around "AI model providers entering services." Any shift in tone there signals that the competitive overlap with Anthropic and OpenAI is moving from theoretical to contractual.

• The valuation reset risk. Both companies are priced at valuations implying that generative AI remains the primary infrastructure layer for enterprise software indefinitely. Any credible shift — open-source models compressing API pricing, a regulatory intervention in the US or EU, or a major model failure in production — would reprice both quickly. The IPO window in late 2026 is, in part, a race against that uncertainty.