OpenAI Hits $25 Billion in Revenue and Eyes an IPO — Then Britannica and Merriam-Webster Sue It

OpenAI crossed $25 billion in annualized revenue this month, a figure that puts it in a category shared by fewer than 200 companies in the world — and it took OpenAI just under four years of commercial operation to get there. The milestone, confirmed by sources familiar with the company's financials speaking to the Financial Times, is the clearest financial validation yet of CEO Sam Altman's argument that the company can justify its $300 billion private valuation. The next logical step, according to three people briefed on internal discussions, is an initial public offering in the second half of 2026 — a move that would represent the largest tech IPO since Arm Holdings in 2023.

The revenue figure reflects a product suite that has matured considerably since GPT-3. OpenAI's current flagship model, GPT-5.4 Thinking, features a one-million-token context window and supports the kind of multi-step agentic workflows — autonomous task execution over hours or days — that enterprise customers have been demanding. The enterprise segment now accounts for roughly 62 percent of total revenue, up from 41 percent two years ago, as companies from law firms to pharmaceutical research groups have embedded OpenAI's API into core business processes. Consumer ChatGPT subscriptions, priced at $20 to $200 per month depending on tier, contribute the remaining 38 percent.

The IPO preparation is well underway in practice, even if no formal prospectus has been filed. OpenAI has retained Goldman Sachs and Morgan Stanley as lead underwriters, according to sources familiar with the mandate cited by Bloomberg. The company is also planning to nearly double its headcount to approximately 8,000 employees over the next 18 months — a hiring surge that will require the kind of human resources infrastructure typical of a pre-IPO company. An outside general counsel was hired in February; a chief financial officer with public-company experience was hired in January.

“OpenAI has retained Goldman Sachs and Morgan Stanley as lead underwriters, according to sources familiar with the mandate cited by Bloomberg.”

Into this picture of momentum arrived, on Monday, a federal lawsuit filed in the Southern District of New York by Encyclopaedia Britannica and Merriam-Webster. The complaint alleges that OpenAI systematically scraped approximately 100,000 copyrighted articles, reference entries, and dictionary definitions from both companies' websites and databases without authorization, license, or compensation, and used that material to train its large language models. The plaintiffs are seeking statutory damages under the Copyright Act, which allows awards of up to $150,000 per work for willful infringement — a theoretical exposure, if the court sides with the plaintiffs on willfulness, that runs into the billions of dollars across 100,000 works.

The lawsuit is significant not just for its financial stakes but for who is filing it. Encyclopaedia Britannica and Merriam-Webster are not fringe plaintiffs; they are two of the most authoritative reference institutions in the English language, and their participation adds a gravitas to the AI copyright litigation wave that previous plaintiffs — individual authors, small publishers — did not carry. The New York Times sued OpenAI and Microsoft in late 2023 and that case is still in discovery. The Authors Guild has an ongoing class action. But Britannica and Merriam-Webster explicitly position their complaint as a defense of "the knowledge commons" — the shared corpus of vetted, expert-reviewed information that AI models disproportionately rely on for accuracy.

OpenAI's standard legal response in such cases has been to argue that training on publicly accessible text constitutes fair use under US copyright law. Several law professors cited by The Atlantic have assessed that the fair use argument is plausible but not settled — no appellate court has yet ruled on it directly. The Second Circuit, which covers the Southern District of New York, is considered the most copyright-protective federal circuit in the country, which is why plaintiffs consistently choose to file there.

The broader AI investment environment remains extraordinarily active despite the legal clouds. Total AI capital expenditure across the industry is projected at $115 to $135 billion in 2026, according to Bernstein Research — essentially double the 2025 level. Microsoft alone has committed $80 billion in AI infrastructure investment through fiscal year 2026. The energy implications are attracting serious attention: Morgan Stanley published a research note estimating a 9-to-18 gigawatt shortfall in US power supply through 2028 driven by AI data center demand, a figure that has accelerated utility capital investment plans across the country and pushed power stocks to multi-year highs.

Google contributed its own chapter to the week's AI news. The company released Gemini 3.1 Flash-Lite on Tuesday, a model the company claims is 2.5 times faster than its predecessor and priced at $0.25 per million input tokens — a pricing level that makes it competitive with the cheapest models from Anthropic and Mistral. Flash-Lite is aimed explicitly at high-volume, latency-sensitive enterprise applications where inference cost per query matters more than maximum capability. The release continues a pattern of Google using Gemini's lower-cost tiers to compete on price while reserving its frontier models (Gemini 3.1 Ultra) for capability comparisons against OpenAI's GPT-5.4 Thinking.

**What this means for you**

For investors, OpenAI's $25 billion revenue figure and IPO trajectory create a decision point: whether to seek pre-IPO exposure through secondary market shares (currently trading at implied valuations between $290 and $320 billion on private secondary platforms) or wait for the public offering. The risk in the IPO itself is the copyright litigation — a ruling against OpenAI on willfulness could create a disclosure obligation that complicates the prospectus. For enterprise customers, the rapid decrease in AI inference costs (Gemini Flash-Lite at $0.25/million tokens is approximately 80 percent cheaper than GPT-4o was at launch) means the economics of integrating AI into workflows are improving faster than most CFO projections from two years ago anticipated. For content creators and publishers, the Britannica-Merriam-Webster lawsuit represents the strongest institutional challenge yet to the fair use theory that currently underpins the entire AI training data ecosystem.