Observed Signal · May 21, 2026 · Technical Release · Source: Gary Marcus · Impact: 4/5 · Sentiment: Neutral
Assessing OpenAI Math Breakthrough and Anthropic Profit Claim
The author reviews two major recent items: OpenAI’s announcement that a new, unreleased reasoning model helped identify a counterexample to the planar unit distance problem (an Erdős conjecture), and reporting that Anthropic is projecting its first slightly profitable quarter. Independent mathematicians (including Thomas Bloom) reviewed OpenAI’s model output and helped extract a conventional proof from a long chain-of-thought transcript; critics urge caution because the internal model, training, prompting volume, cost, and generality remain undisclosed. Separately, reporting suggests Anthropic’s projected profit may be aided by a one-time compute discount disclosed in SpaceX’s S-1 filing; the exact discount and whether future quarters will be profitable are unclear. The piece also notes social-media commentary raising skepticism about Nvidia’s cash-flow dynamics and warns IPO buyers to exercise caution.
OpenAI (a major AI platform) announced a notable technical result using an unreleased LLM variant and Anthropic’s prospective profitability (with SpaceX compute concession) affects expectations for AI company financials—both developments materially affect the AI ecosystem and investor perceptions.
Track Anthropic Signals & Market Shifts in Real-Time
Polaris7 autonomous intelligence agents track regulatory filings, primary sources, executive changes, and deal flow 24/7. Create your free Explorer workspace to monitor these entities.
Key Takeaways & Evidence Grounding
- OpenAI announced that a new, not-yet-released reasoning model produced a counterexample disproving a conjecture related to the planar unit distance problem (an Erdős problem).
- OpenAI released a companion PDF of model output; mathematician Thomas Bloom reviewed the full model output and helped identify factors behind the result.
- Cal Newport and other commentators cautioned that the model’s training, prompting volume, costs, and generality across tasks are unknown and that chain-of-thought transcripts were used to extract the result.
- The Wall Street Journal reported (via Berber Jin) that Anthropic is projecting its first slightly profitable quarter, potentially aided by a one-time compute discount from SpaceX disclosed in SpaceX’s S-1 filing.
- Social-media commentary flagged concerns about Nvidia’s cash-flow and 'circular financing' absorbing a large share of operating cash flow in a recent quarter.
Connected Companies & Entities
5 Entities mappedOntology Mapping & Concepts
Related Market Signals & Shifts
Recent verified developments and strategic activity across this market segment.
OpenAI and Anthropic Report Massive Losses
This analysis examines leaked and reported financials for leading AI labs and concludes current business models are deeply unprofitable. Leaked OpenAI 2025 accounts (obtained by Ed Zitron and verified by the Financial Times) show $13 billion in revenue but a $38.5 billion net loss; the author calculates a no-nonsense operating loss of $20.92 billion based on disclosed line items. Anthropic reported $4.5 billion in revenue last year and a self-reported ARR of $47 billion in May, but the company also lost nearly $6 billion in 2024 and its path to profitability appears conditional and dependent on discounted contracts and uncertain compute costs. The author estimates Anthropic’s operating loss at roughly $11 billion under reasonable assumptions. The piece argues Anthropic could theoretically reach profitability under optimistic scenarios, while OpenAI remains far from viable without major revenue growth or aggressive cost cuts.
Anthropic vs OpenAI: Pre‑IPO BigAI Showdown
The article analyzes the pre‑IPO competition between Anthropic and OpenAI, highlighting major funding rounds, growth trajectories and product cadence. It reports Anthropic closed a $30 billion round while OpenAI is nearing a reported $100 billion round; Nvidia is reportedly in talks to invest up to $30 billion in OpenAI at a $730–$850 billion pre‑money valuation. Revenue projections and customer metrics from third parties (Epoch AI, Ramp) suggest Anthropic may outpace OpenAI in ARR by late 2026, supported by faster historical growth and recent model releases (Anthropic Sonnet 4.6, Google Gemini 3.1 Pro). The piece cites Ramp data showing high customer overlap (79% of Anthropic customers were already OpenAI customers), low churn (~4%), and growing dual adoption (16% of businesses pay for both). It also discusses agentic AI, Model Context Protocol (MCP), and profitability timelines (Anthropic possibly profitable by 2028; OpenAI possibly by 2031).
WSJ Reveals OpenAI and Anthropic Divergent Financials
The Wall Street Journal published confidential financial documents from OpenAI and Anthropic that were shared with investors ahead of recent funding rounds and anticipated IPOs. The filings show strikingly different economics: OpenAI projects roughly $121 billion in compute spending in a single year, while Anthropic’s training costs are described as starting near zero relative to revenue and declining rapidly as a share of sales. The piece argues the key takeaway is not just the headline numbers but the structural divergence in how the two companies plan to survive, scale, and monetize frontier AI models — implying different bets on capacity, cost structure, and long-term competitiveness within the AI infrastructure market.
Track Real-Time Market Signals & Shifts
Set up custom watchlists to receive automated, evidence-grounded executive digests whenever material signals or shifts occur across your tracked landscape.