On Sunday April 13, OpenAI chief revenue officer Denise Dresser sent an internal memo to staff making a pointed argument: Anthropic’s widely cited $30 billion revenue run rate is overstated by roughly $8 billion.

The timing was deliberate. Anthropic had announced its $30B milestone just days earlier, framing it as the first time the company had ever outearned OpenAI. Dresser’s memo pushed back on the headline, argued the two companies are not comparing like for like, and made broader criticisms about Anthropic’s compute capacity and strategic position.

The dispute quickly leaked to Fortune, Axios, and several other outlets. It is now public.

The Accounting Argument

The core disagreement is about how to book revenue from cloud partners.

Anthropic distributes its Claude models through AWS, Microsoft Azure, and Google Cloud. When a customer buys Claude access through AWS, Anthropic counts the full billing amount — including the portion that goes to AWS — as top-line revenue. This is called gross revenue recognition.

OpenAI distributes through Microsoft Azure in a similar arrangement. But OpenAI recognises only the net amount it receives after Microsoft’s cut is deducted.

Dresser’s memo argues that applying Anthropic’s gross accounting methodology to OpenAI’s own cloud revenue would make the two companies roughly equivalent in size — not $5 billion apart. Her estimate: Anthropic’s real figure is closer to $22 billion on a comparable basis.

Anthropic pushed back. The company told CNBC that it recognises gross revenue because it is the principal in the transaction and its cloud partners function as distribution channels, not resellers. This is a standard and permissible justification under US GAAP. Anthropic is not doing anything fraudulent.

Both approaches are defensible. But they produce very different headline numbers, and both companies know that those headline numbers will be scrutinised closely as they head into what are expected to be dual IPOs before the end of 2026.

What Else Was in the Memo

The accounting dispute was only part of Dresser’s argument. The memo also made strategic criticisms of Anthropic that reveal what OpenAI sees as its main competitive angles heading into IPO season.

On compute: Dresser claimed Anthropic’s infrastructure constraints are already causing product throttling and weaker availability for enterprise customers. OpenAI, with its deeper relationship with Microsoft’s Azure infrastructure, frames this as a durable reliability advantage.

On enterprise positioning: Dresser acknowledged that Anthropic’s “coding focus gave them an early wedge” with enterprise customers — particularly in developer-heavy organisations. But she argued this narrow positioning becomes a liability as AI expands beyond engineering teams into operations, finance, HR, and customer-facing functions. OpenAI, with its broader product surface across ChatGPT Enterprise, Codex, and Operator, argues it is better positioned for that expansion.

Why This Matters for the IPO Race

Both Anthropic and OpenAI are expected to file for IPOs in late 2026. OpenAI is reportedly targeting a valuation around $852 billion. Anthropic’s most recent funding round valued the company at $61.5 billion.

When two companies are this close to IPO, every revenue metric becomes a talking point for institutional investors. Dresser’s memo was written to shape that narrative internally, and the fact it leaked publicly suggests it may have been intended to shape it externally too.

The memo reads less like an accounting lesson and more like a competitive attack designed to reframe Anthropic’s momentum before the IPO window opens.

Anthropic’s enterprise metrics remain strong regardless of which accounting method you prefer. More than 1,000 enterprise customers are spending over $1 million per year — a figure that doubled in less than two months between February and April 2026. The growth is real. The question is how to measure it.

What This Means for Business

If you are a business owner or technology leader evaluating AI vendors, this dispute is a useful reminder to look past headline numbers.

Revenue figures tell you market traction, not product fit. Whether Anthropic’s run rate is $22 billion or $30 billion on an adjusted basis, both numbers confirm it is a serious, well-resourced vendor. What matters more to your business is whether Claude, GPT-5.x, or another model actually performs better in your specific use case.

Pre-IPO dynamics change vendor behaviour. Both OpenAI and Anthropic are increasingly focused on enterprise revenue because that is what institutional investors want to see. This creates a window for businesses to negotiate better terms — enterprise contracts will be prioritised over consumer deals in the run-up to IPO. But it also means vendor pricing strategies may shift post-IPO as growth imperatives change.

Gross vs net accounting is common in platform businesses. Anthropic is not doing anything unusual. If you are evaluating AI vendors through a due diligence lens — say, for a procurement decision or a vendor risk assessment — ask each vendor which accounting standard they apply to their reported revenue. The answer will help you make like-for-like comparisons.

Compute reliability is a real concern. Dresser’s point about Anthropic’s infrastructure constraints is worth taking seriously, regardless of the competitive motivation behind raising it. For workloads where uptime and throughput guarantees matter — production AI agents, customer-facing automations, internal knowledge tools — asking vendors for specific SLA commitments is more valuable than tracking their revenue milestones.

The two biggest AI labs in the world are fighting over who gets to claim the top spot before their IPOs. That is a testament to how fast enterprise AI has grown. For businesses, the practical takeaway is simpler: evaluate vendors on what they can reliably do for your workflows, not on which accounting method makes their revenue look biggest.

---

If you’re building with Claude or Codex right now, grab the free Working With Claude field guide. Thirty-two pages on the full ecosystem, Claude Code in depth, and how to roll agents out properly. Get the free guide.