AI War: OpenAI vs. Anthropic and the Compute Struggle
An analysis of the strategic competition between OpenAI and Anthropic, focusing on compute resources, revenue reporting, and the race toward IPOs. The discussion also covers the financial health of SpaceX's AI ambitions and the shifting landscape of AI security.
The Great AI Compute War
The battle for supremacy in Large Language Models (LLMs) has shifted from pure architectural innovation to a brutal war of attrition centered on compute resources. While OpenAI and Anthropic are the primary contenders, the struggle is now defined by how much raw processing power they can secure and how they report their financial viability to potential investors.
Strategic Discrepancies in Scaling
OpenAI has pursued a strategy of massive over-investment in compute, aiming for a structural advantage. In contrast, Anthropic has been more cautious, leading to higher operational profitability but facing significant scaling bottlenecks. This has resulted in frequent outages and potential delays in releasing new models due to the lack of-capacity to run them. OpenAI's leadership has internalised this as a competitive edge, though some argue that Anthropic's better unit economics might make them more attractive in the long run.
Financial Engineering and IPO Readiness
As both companies eye the public markets, the debate over revenue reporting has intensified. OpenAI has accused Anthropic of inflating its revenue run-rate by including revenue shares from partners like Amazon and Google, whereas OpenAI reports net revenue. This accounting difference highlights a critical lesson for investors: focusing on gross profit (Revenue $\times$ Gross Margin) rather than top-line revenue is the only way to accurately compare the efficiency of these AI giants.
The SpaceX Factor
Elon Musk's xAI is integrating deeper into the SpaceX ecosystem. While SpaceX remains highly profitable through Starlink and Space Launch, the AI segment is currently a significant cash-burn center, with estimated losses of billions per year. The impending SpaceX IPO is expected to absorb a massive amount of market liquidity, potentially affecting the timing and valuation of other AI-driven IPOs.
Conclusion
For leadership and investors, the AI sector is no longer just about the 'magic' of the model; it is about the physical infrastructure of the data center and the balance sheet. The winner will be determined by who can balance aggressive scaling with sustainable unit economics.
Key insights
-
The competition between OpenAI and Anthropic is now primarily a battle over compute capacity. OpenAI's aggressive investment in data centers provides a structural advantage in availability and reliability, while Anthropic's leaner approach increases operational profitability but limits growth.
Impact: Companies must prioritize secure access to compute over pure software optimization to avoid scaling bottlenecks.
-
Comparing AI company revenues is misleading due to different accounting treatments of cloud partner revenue shares. Focusing on gross profit and raw margins is the only reliable way to assess the actual financial health and efficiency of these LLM providers.
Impact: Investors should demand gross margin data over top-line revenue figures to avoid being misled by inflated run-rates.
-
The AI sector's financial landscape is heavily influenced by the 'compute-heavy' nature of the business, where high CapEx is often masked by Adjusted EBITDA. True financial viability is determined by the actual cash flow after depreciation and massive infrastructure investments.
Impact: High valuation premiums may be unsustainable if cash flow remains deeply negative despite high adjusted profits.
Action items
-
Shift financial analysis of AI-based companies from Top-Line Revenue to Gross Profit (Revenue $\times$ Gross Margin). This removes the noise created by revenue-share agreements with cloud hyperscalers.
Impact: Provides a more accurate baseline for comparing the efficiency and sustainability of AI service providers.
-
Evaluate AI implementation strategies based on 'compute-readiness' rather than just model performance. Ensure that the chosen provider has the structural capacity to maintain uptime and reliability during scaling.
Impact: Reduces operational risk for enterprises integrating LLMs into critical business workflows.
Quotes
“The problem is that OpenAI has reports Microsoft Revenue Share Net, which is more in line with standards we would be held to as a public company.”
“I believe that the one who has good unit economics has, perhaps, the ability to secure additional capacity.”
“The winner will be determined by who can balance aggressive scaling with sustainable unit economics.”