OpenAI Chief Revenue Officer Denise Dresser sent a four-page internal memo on April 12, 2026 — subsequently leaked to CNBC, The Verge, and Axios — accusing competitor Anthropic of overstating its annual revenue run rate by approximately $8 billion. The OpenAI-Anthropic revenue memo’s central claim: Anthropic’s widely-cited $30 billion ARR figure includes cloud provider marketplace fees that OpenAI excludes from its equivalent Microsoft-channel numbers, making the two figures structurally incomparable.
The memo landed as both companies sit in active pre-IPO positioning. Within 48 hours it had reshaped how institutional investors and enterprise sales teams were interpreting the competitive standings between the two most closely watched AI companies of 2026.
The Billion Accounting Gap: Gross Revenue vs. Net Revenue
The gross-versus-net revenue distinction is routine accounting — until it becomes an $8 billion line in a pre-IPO narrative. OpenAI’s memo argues that Anthropic routes substantial revenue through AWS Marketplace, Azure AI, and Google Cloud, booking the full transaction value as topline revenue — including the cut that Amazon, Microsoft, and Google retain. OpenAI, by contrast, reports its Microsoft-facilitated revenue net of the partner share.
The practical result: two companies serving comparable actual customer spend can report materially different “revenue” figures. Dresser’s memo, per CNBC’s reporting, places Anthropic’s adjusted run rate at approximately $22 billion — a 27% reduction from the $30 billion headline.
Neither figure has been independently audited. Anthropic is privately held with no obligation to publish GAAP financials. OpenAI’s own financials remain internal. Both companies are running on investor-grade ARR estimates, not financial-statement-grade revenue — a distinction that becomes unavoidable when S-1 filings reach the SEC.
What the OpenAI Anthropic Revenue Memo Actually Says
Beyond the accounting argument, Dresser’s memo is ideological. She characterizes Anthropic’s market positioning as “built on fear, restriction, and the idea that a small group of elites should control AI” — a direct assault on Anthropic’s safety-first brand identity, the same identity that secured Google’s multi-billion-dollar backing and drives procurement decisions in regulated industries wary of frontier model risk.
The memo also labels Anthropic’s compute strategy a “misstep.” The target is Anthropic’s reliance on cloud-distributed inference rather than owned compute infrastructure — a structural bet that trades capital efficiency for scale dependency. OpenAI, with its Microsoft Azure relationship and ongoing discussions about dedicated data center capacity, positions its infrastructure approach as the superior long-term play.
This isn’t the first time Anthropic’s operational decisions have drawn outside scrutiny. Earlier this year, Anthropic accidentally released source code for a Claude AI agent, raising questions about internal controls at a company that markets itself primarily on safety and reliability.
Anthropic’s GAAP Defense
Anthropic has not publicly confirmed the $30 billion figure, but sources close to the company indicated its revenue accounting follows standard GAAP principal-agent guidance. Under ASC 606, a company recognizes gross revenue when it controls the service, sets pricing, and bears primary delivery risk — and net revenue when acting as an agent. If Anthropic’s cloud marketplace arrangements qualify it as principal, gross revenue recognition is technically defensible.
The question isn’t which company is fabricating numbers. It’s whether their accounting choices make the headline figures comparable for investors evaluating both businesses. They don’t — and PitchBook senior analyst Harrison Rolfes made that explicit: “Both companies are in an ARR accounting arms race ahead of their IPOs, and neither is reporting on a basis that would survive a Big Four audit.”
The ASC 606 principal-agent distinction has produced restatements and investor disputes at major SaaS platforms before. At Anthropic’s scale, the $8 billion delta is not a rounding-error accounting choice. It is a choice that changes the valuation math by hundreds of billions of dollars at standard revenue multiples.
OpenAI’s Microsoft Problem, Admitted Out Loud
The memo’s most strategically revealing line was its most overlooked. Dresser acknowledged that Microsoft has “limited our ability” to reach enterprise customers directly — an internal admission that OpenAI’s foundational infrastructure partnership carries structural costs to its go-to-market independence.
OpenAI’s reliance on Azure for compute and distribution means that in enterprise sales contexts, Microsoft’s priorities and channel agreements shape what OpenAI can and cannot offer. OpenAI’s reported $1 billion Disney deal navigated around standard Microsoft enterprise channels — an arrangement that illustrated both the scale of the opportunity and the friction embedded in the partnership.
For investors evaluating an OpenAI IPO, the admission matters. A company that cannot fully control its enterprise go-to-market because of a partner agreement carries a different risk profile than one with direct customer ownership. Dresser aimed to attack Anthropic’s financials; she also disclosed a structural constraint that analysts will price into OpenAI’s offering.
The Enterprise Battlefield: What Ramp Data Shows
Spending data from Ramp — the corporate card platform tracking real-time B2B software adoption across thousands of companies — shows Anthropic closing the enterprise customer gap with OpenAI within a two-month window in early 2026. That ground-level signal is what makes Dresser’s memo urgent rather than defensive theater.
Enterprise AI spending is consolidating faster than most predicted. CIOs who defaulted to OpenAI in 2023–2024 are now running structured vendor evaluations, and Claude models are consistently appearing on shortlists in healthcare, financial services, and legal — sectors where safety positioning converts directly into procurement outcomes. Cost-per-token convergence between both companies’ most-deployed models has effectively eliminated pricing as a differentiator in most enterprise segments.
MegaOne AI tracks 139+ AI tools across 17 categories. In 2026, Anthropic’s enterprise offering — Claude for Work, the API, and dedicated compliance tooling — has shifted from “credible OpenAI alternative” to “preferred vendor for regulated industries” across multiple verticals. That trajectory is precisely what OpenAI is attempting to disrupt.
IPO Stakes: Why This Leak Was Not an Accident
OpenAI’s most recent funding valued it at $157 billion. Anthropic raised at a $61.5 billion valuation in its last round. For both companies, the revenue narrative directly affects IPO pricing mechanics: an $8 billion difference in claimed ARR at a 30x revenue multiple implies a $240 billion difference in implied market capitalization.
Leaks of this nature do not emerge from sophisticated communications operations by accident. A four-page memo from a CRO reaching three major outlets within days of internal circulation follows the playbook of a deliberate narrative plant — seeding doubt about a competitor’s financials before institutional investors finalize their positions ahead of upcoming offerings.
The timing makes the strategy explicit. April 2026, with IPO windows opening for both companies, means the intended audience is not OpenAI’s sales force. It is the analysts, fund managers, and financial journalists who will price and cover these offerings. The memo is investor relations conducted through a different channel.
OpenAI’s competitive maneuvering operates across multiple fronts simultaneously. OpenAI’s strategic positioning and acquisition activity reflects a company that treats narrative control as a core operating discipline, not an afterthought.
The Real Signal in the Billion Dispute
Whether Anthropic’s actual run rate is $22 billion or $30 billion is less important than what either number signals: two AI companies are competing for a market that barely existed at scale three years ago, and that market is now large enough to generate genuine revenue accounting wars in the pre-IPO phase.
The dispute will not resolve before either company files. Investors will receive adjusted figures, non-GAAP disclosures, and growth rate comparisons designed to resist direct comparison. What they will actually be evaluating is trajectory — and on that metric, Ramp’s real-time spending data suggests the race is considerably closer than either company’s investor communications would prefer to acknowledge.
The memo landed a punch. It also revealed that OpenAI is monitoring Anthropic’s enterprise traction closely enough to produce a four-page internal document in response. That level of competitive surveillance is its own data point about where the real threat is coming from. Anthropic’s reply, when it arrives, will be equally tactical — and aimed at the same audience that just read Dresser’s memo.
