📈 AI accounting 101 | Axios AI+ Summary

1. Key Themes

AI revenue comparisons are distorted by accounting choices, not just growth rates

The headline gap between Anthropic ($65B annualized) and OpenAI ($40B annualized) revenue is significantly inflated by differing revenue recognition methods (gross vs. net), not purely by differing business momentum. "If a customer pays $100 for an AI service through a cloud provider, Anthropic's accounting method allows it to record the full $100 as top-line revenue, per Francine McKenna." Yet even adjusting for this, the gap remains large: "That would still put Anthropic roughly $19 billion to $21 billion ahead of OpenAI, based on the latest reported figures."

AI security is becoming a governance and infrastructure crisis, not just a technical concern

Labs and lawmakers are simultaneously mobilizing around AI-agent and critical-infrastructure risk following the Hugging Face breach and water-system attacks. "OpenAI's accidental hack of Hugging Face and a wave of seemingly AI-assisted cyberattacks on U.S. water systems have raised serious concerns about how vulnerable utilities are to future AI-enabled hacks." In parallel, Congress is moving: "The Stop Rogue AI Act directs Commerce's National Institute of Standards and Technology to develop and publish standards, guidelines and best practices for how organizations can securely deploy AI agents."

The frontier model race has entered a rapid, simultaneous-release phase

Meta, Google, Anthropic, and OpenAI all shipped major updates in the same week, signaling intensifying competitive pressure heading into agentic products. "Meta's release comes amid a flurry of model updates from across the industry," with Google's Gemini 3.8 Flash, Anthropic's Fable/Mythos updates, and OpenAI's upcoming Astra model all launching concurrently.

2. Contrarian Perspectives

The "Anthropic is beating OpenAI" narrative is mostly an accounting artifact, not proof of superior execution. Consensus coverage treats Anthropic's revenue lead as evidence of outright market dominance, but the article shows the accounting methodology itself manufactures much of the gap: "Comparing revenue at the two AI labs isn't apples-to-apples unless you understand how differently they account for some sales through cloud partners." Crucially, "Neither approach is necessarily wrong" — this isn't fraud or manipulation, just different (defensible) principal-vs-agent accounting treatments, meaning investors should discount headline revenue comparisons until IPO disclosures clarify the real picture.

3. Companies Identified

Anthropic — AI lab and Claude developer, preparing for IPO. Why mentioned: Central case study in revenue-accounting comparison with OpenAI; its gross-revenue accounting method is driving perception of runaway growth. Quote: "Anthropic is racing toward an IPO with a headline revenue number that will likely look bigger than that of its rival, OpenAI."

OpenAI — AI lab behind ChatGPT and upcoming Astra model. Why mentioned: Compared against Anthropic on revenue accounting; also central to cybersecurity news given the Hugging Face breach and its cyber summit for critical infrastructure defenders. Quote: "OpenAI's accidental hack of Hugging Face and a wave of seemingly AI-assisted cyberattacks on U.S. water systems have raised serious concerns..."

Nvidia — Chipmaker and AI infrastructure giant. Why mentioned: Announced acquisition of Hugging Face. Quote: "Nvidia CEO Jensen Huang confirmed the company has agreed to acquire Hugging Face for $12.9 billion, emphasizing the importance of the AI library in the open AI ecosystem."

Hugging Face — Open AI model/library ecosystem platform. Why mentioned: Acquisition target of Nvidia; also site of a security breach that's driving policy and industry response. Quote: See above; also referenced repeatedly regarding the "Hugging Face breach."

Meta — Social media and AI company, developer of Muse Spark and Llama-family models. Why mentioned: Released Muse Spark 1.3 update aimed at coding/agentic performance, positioning for "personal agents." Quote: "Meta released Muse Spark 1.3, an update it says significantly improves performance in coding and agentic tasks."

Google — AI lab, developer of Gemini models. Why mentioned: Released Gemini 3.8 Flash and a cybersecurity-focused variant amid the industry's model-release wave. Quote: "Google's new model showed some impressive performance on certain coding benchmarks, and a larger update, Gemini 4, remains in training."

Oath — Financial verification platform. Why mentioned: CEO's perspective used to frame accounting ambiguity and EBITDA-based scrutiny of Anthropic. Quote: "Accounting is 'more of an art than a science,' Lucas Ward, co-founder and CEO of Oath, a financial verification platform, told Axios."

The Information — Tech media/research outlet. Why mentioned: Source of original reporting on the Anthropic/OpenAI revenue-accounting differences and OpenAI's Astra training technique concerns. Quote: "Both companies annualize recent sales in roughly similar ways, The Information reported."

4. People Identified

Francine McKenna — Accounting professor and author of newsletter "The Dig." Why mentioned: Provides the technical explanation of principal-vs-agent accounting driving the Anthropic/OpenAI revenue gap. Quote: "The criteria here has a lot to do with who is the principal and who is the agent in the transaction."

Lucas Ward — Co-founder and CEO of Oath, a financial verification platform. Why mentioned: Offers investor-relevant framing on how to evaluate Anthropic's true financial health ahead of IPO. Quote: "Accounting is 'more of an art than a science.'" Also: "Ward plans to watch Anthropic's EBITDA... for clarity on the firm's true profits."

Greg Brockman — President of OpenAI. Why mentioned: Leading OpenAI's cybersecurity summit and critical-infrastructure access announcement. Quote: "OpenAI president Greg Brockman is expected to make an announcement about expanding access to the company's tools to critical infrastructure and public sector organizations."

Alexandr Wang — Meta's AI chief. Why mentioned: Explains strategic rationale behind Muse Spark 1.3 and Meta's personal-agent roadmap. Quote: "A lot of the usability improvements that we've made will be really helpful for things that Mark [Zuckerberg] has talked about on earnings calls, like personal agents that can work 24/7 on your behalf."

Reps. Josh Gottheimer (D-N.J.) and Mike Lawler (R-N.Y.) — Members of Congress. Why mentioned: Introducing the Stop Rogue AI Act to regulate AI agent security. Quote: "[They] are introducing a new bill today... aimed at securing AI agents amid a growing number of safety incidents involving rogue agents."

5. Operating Insights

Scrutinize revenue recognition methodology when benchmarking AI companies (or your own comparables). Whether a company records gross or net revenue on partner-channel sales can swing reported top-line figures by double digits — "If Anthropic were to pivot and report revenue on a net versus gross basis, that would only amount to about a 6%–10% hit" — but that swing is still material enough to mislead investors relying on press-released ARR figures.
Watch for SEC correspondence in IPO filings as a due-diligence goldmine. Pre-IPO paperwork will reveal how a company justified its accounting to regulators: "As part of the pre-IPO paperwork process, correspondence between the SEC and Anthropic will get released. That should include justifications for Anthropic's accounting practices."
Security readiness lags behind AI capability rollout — a gap operators should proactively close. Even as OpenAI publicizes defender-focused tools, the underlying organizational capacity issue remains unresolved: "it doesn't solve the problems many of these organizations find with funding and hiring security teams."

6. Overlooked Insights

OpenAI's new "Astra" training technique may reduce visibility into model reasoning even as capabilities increase — a potential transparency/oversight risk understated amid the cybersecurity announcements: "An Information report earlier this week about OpenAI pursuing a new training technique also raised concerns about whether security pros will still have visibility into the model's thought process."
Anthropic's Fable and Mythos model updates were framed around loosening safety guardrails for commercial reasons, a notable shift buried in the model-race roundup: updates were "designed to cut costs and reduce the frequency of refusals of user requests for safety issues" — suggesting competitive pressure may be pushing labs toward relaxing safety friction points.