According to Anthropic's ledger, its annual revenue is $30 billion, while OpenAI estimates the same batch of sales figures to be worth only $22 billion. Neither figure is fabricated. This was the first jab thrown by OpenAI's Chief Revenue Officer, Denise Dresser, in the four-page internal letter that was exposed by the media on April 13.

The starting point of the matter was a memo obtained by The Information. In the letter, Dresser did three things: praised Amazon for its new collaboration which has "incredible demand," acknowledged that the Microsoft partnership "limited our reach to customers," and then spent considerable time dissecting Anthropic's revenue figures. The timing of the letter's leak was precisely one week after Anthropic announced it had crossed the $30 billion annual revenue milestone.

What appears to be an internal company communication is essentially a carefully constructed information war. To understand it, one should directly approach it from three dimensions: revenue recognition criteria, competitive landscape on the enterprise side, and the arms race for computing power, and then place all of them into a single cloud collaboration structure diagram.

Where the $8 billion Accounting Discrepancy Comes From

Anthropic reported $30 billion in annual revenue, while OpenAI claims the actual figure is $22 billion. The difference of $8 billion arises from the starkly different choices the two companies have made in their revenue accounting metrics.

Anthropic uses a gross revenue accounting method: when a business purchases usage credits for Claude through AWS, Anthropic accounts for the total amount of money as top-line revenue, treating the platform fees paid to Amazon as costs. Conversely, OpenAI only records the net amount actually received from Microsoft, and Microsoft's share does not impact the top line.

Both methods comply with U.S. Generally Accepted Accounting Principles (GAAP). Anthropic's rationale is that it is the "principal" party in customer transactions, while the cloud provider is merely a distribution channel. OpenAI's logic views Microsoft as the "agent," counting only the actual amounts received. The root of the disagreement lies not in who is fabricating numbers but in who more aggressively asserts their dominant position in the sales chain.

Dresser writes in the memo that Anthropic "employs accounting methods that make the revenue figures appear larger," including incorporating the gross amounts of AWS and Google platform payouts into top-line revenue. The subtext of this statement is not difficult to understand; when Anthropic submits its S-1 filing to the SEC, auditors will adjudicate this metric, possibly necessitating unified disclosure adjustments. When calculated on the same basis, Anthropic would be at $22 billion, and OpenAI would be at $24 billion, changing the leading position.

It should be noted that Anthropic's revenue growth rate is already at a historic level. According to data from Bloomberg and Sacra, its annual revenue soared from approximately $9 billion at the end of Q4 2025 to now $30 billion, more than threefold in less than five months, driven mainly by actual client purchases, not explainable through mere accounting adjustments. The crux of this accounting controversy is not that Anthropic is shrinking, but that OpenAI is using the "metrics" knife to redraw the boundaries.

The Speed of the Enterprise Side's Catch-up is Faster than Most Anticipated

The Ramp tracking platform, which monitors the actual AI spending behaviors of thousands of businesses, provides firsthand data for assessing true enterprise choices.

According to the Ramp AI Index for April: Anthropic's share among enterprise paying clients has risen to 30.6%, while OpenAI is at 35.2%, narrowing the gap from 11 percentage points in February to 4.6 percentage points. Based on Anthropic's average monthly growth rate of +6.3 percentage points over the past two months (already the largest single-month increase on record for this metric), it is on track to overtake OpenAI in this metric in about two months.

More noteworthy are the structural signals. In three high-spending sectors, Anthropic's lead has become a reality: information technology/software (63% vs. 54%), financial services (52% vs. 46%), professional services (47% vs. 44%), all exceeding OpenAI. These three industries just happen to be where AI budgets are most concentrated and procurement decisions are most professional. This indicates that the companies with the largest voice in the AI purchasing chain have begun to collectively lean towards Anthropic.

Dresser unusually acknowledges in the memo that Anthropic "has a significant lead among enterprise clients," attributing it to programming capability. This statement coming from inside OpenAI carries a weight completely different from external evaluations; it conveys to its employees that the competitor has won on the core battleground. She also added a warning: "You do not want to be a single-product company in a platform war." This serves as a reminder to employees that if Claude's programming advantage cannot extend to the platform level, it will ultimately be just a ticket rather than a boarding pass.

Computing Power Disparity: Close Today, Four Times by 2030

Computing capacity is the competitive dimension among AI companies that is the hardest to bridge in the short term, as its construction cycle is measured in years and the funding threshold is in the tens of billions.

Currently, the numbers don't look too disparate: OpenAI at around 1.9 gigawatts, Anthropic at around 1.4 gigawatts, a difference of about 35%. Dresser describes Anthropic in the memo as "operating on a meaningfully smaller curve," but this statement is not an exaggeration given the current capacity comparison; the disparity is real, but it has not reached a decisive extent yet.

The real divergence occurs after 2027. OpenAI plans to reach 30 gigawatts of computing power by 2030, backed by a $30 billion five-year cloud computing contract with Oracle, the entire Stargate infrastructure project, and a total construction commitment of $1.4 trillion.

Anothropic's path dependency relies on a Broadcom custom chip agreement, with a capacity of 3.5 gigawatts, to be deployed through Google Cloud, effective from 2027, in addition to AWS’s existing training clusters, targeting 7-8 gigawatts by the end of 2027.

Even if Anthropic fully meets its 2027 targets, there would still be a fourfold gap between it and OpenAI's 2030 plans. This gap is not technically insurmountable; if model efficiency improves enough to yield higher returns per unit of computing power, Anthropic could produce sufficiently good products with less computing power.

However, it must maintain the momentum of Claude on the enterprise side and leverage recurring subscription revenue to support its computing power procurement costs: according to Sacra, it is estimated that Anthropic will pay about $1.9 billion to cloud partners this year, rising to about $6.4 billion by 2027.

Amazon: Betting on Two Competitors at Once

The most intriguing line in this memo is Dresser's direct characterization of the partnership with Microsoft, stating that this collaboration "also limits our ability to reach them in the places where enterprises are located."

The movement of OpenAI toward Amazon has become very clear: according to CNBC, in February this year, Amazon announced a $50 billion investment in OpenAI, while acquiring exclusive third-party cloud distribution rights for OpenAI's enterprise Agent management platform, Frontier.

This marks a proactive switch from the Microsoft track to the Amazon track, and the underlying logic is straightforward: many enterprise clients' AI infrastructures have already been built on AWS's Bedrock platform, and Microsoft's exclusivity terms make it difficult for OpenAI to sell directly there.

However, Amazon's role in this competition is equally worth noting; it is currently Anthropic's largest cloud infrastructure partner and strategic investor, having invested a total of $8 billion, with their Project Rainier cluster deploying approximately 500,000 Trainium 2 chips. Amazon's total bet in the entire AI race sums up to $58 billion, simultaneously flowing to two competitors that are directly engaged in the enterprise market.

This is not a diversified bet from a hyper-scale cloud provider; it is a more precise structure: Amazon is both Anthropic's "strategic ally and largest investor" and the new cloud foundation OpenAI is using to "replace Microsoft."

As the two companies vie for the same batch of enterprise clients, the channels they are competing for happen to be Amazon's Bedrock platform, which simultaneously distributes models from both companies. Whichever performs better on Bedrock, Amazon profits, but OpenAI and Anthropic suffer losses against each other.

Under the pressure of continuously losing enterprise market share and structural fractures emerging in the Microsoft partnership, OpenAI chooses to rebuild its narrative through a meticulously calculated digital war while simultaneously leveraging Amazon for a restructured distribution channel. Dissecting the three sets of numbers reveals that this competition is more complex than either side would like you to see.

免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。