From Anthropic to OpenAI: How AI Accounting Is Shaping Big Tech Earnings

Big Tech’s AI Bets Blur the Line Between Startup and Incumbent WinsThis earnings season has offered fresh clarity on a question investors have asked since ChatGPT reshaped markets in 2022: in the race for artificial intelligence, will startups or tech giants come out ahead?

Three years later, the answer appears to be both—at least among the industry’s biggest names.

Alphabet and Amazon have seen major paper gains from their stakes in AI startup Anthropic, whose valuation has surged from $61 billion in March to about $183 billion after its September funding round. The jump boosted both companies’ quarterly results: Amazon recorded a $9.5 billion pretax gain, while Alphabet reported $10.7 billion in net equity securities gains that included its Anthropic holdings. Analysts estimate each holds a double-digit ownership share in the firm.

Microsoft, by contrast, faced the downside of a different accounting method. Its quarterly profit was weighed down by a $4.1 billion net loss related to its roughly 30% stake in OpenAI. Because Microsoft uses equity accounting, it must record its share of OpenAI’s operating losses, rather than marking its investment to market value—even as OpenAI’s valuation has soared and Azure continues to benefit from exclusive access to its AI models.

Analysts describe this as a form of AI “round-tripping”: hyperscalers funding leading AI developers, who then spend much of that capital on their investors’ cloud infrastructure and chips. What once seemed circular is now generating billions in paper gains for some, and volatility for others.

Outside Big Tech, high-profile investors are also repositioning. Michael Burry, known for “The Big Short,” disclosed put options against Nvidia, accusing cloud giants of using aggressive depreciation schedules to inflate earnings. He claims major AI investors could be understating depreciation by $176 billion between 2026 and 2028, potentially overstating profits at firms such as Oracle and Meta by more than 20%.

Burry argues that as GPU lifecycles shorten—Nvidia now refreshes hardware annually—companies are stretching accounting assumptions to preserve margins. Yet the secondary market for older GPUs remains robust: filings show CoreWeave depreciates chips over six years, while rivals like Nebius use four.

SoftBank’s Masayoshi Son, meanwhile, has sold about $5.8 billion in Nvidia shares to help fund a $22.5 billion investment in OpenAI and related AI ventures. The firm has raised over $41 billion through recent financings, giving it flexibility to expand its hardware and AI portfolio while keeping leverage low.

Both moves reflect growing scrutiny of AI valuations and spending cycles. While Burry is betting against what he sees as accounting excess, Son is redeploying capital for liquidity and growth—different motives, same signal: investors are rethinking how long the AI boom can sustain its momentum.

IPO Market Reawakens as AI Capital Needs GrowThe market for tech and AI IPOs is also showing signs of life after a two-year lull. Accel reports that eight venture-backed tech firms have gone public in 2025, up from just four last year and one in 2023. Notable listings include Figma, CoreWeave, and Navan—still far below the 46 IPOs seen in 2021’s peak, but enough to suggest renewed appetite.

Magdalena Heinrich, head of U.S. technology ECM at Bank of America, said her team’s pipeline is “busy again,” with strong demand building for 2026. She attributes the revival to AI’s enormous capital requirements: “These companies need more money than ever to compete and invest in infrastructure, and public markets are becoming attractive again because they offer a lower cost of capital.”

That same logic is driving speculation that OpenAI could explore a public listing of its own. For AI leaders, going public isn’t just about liquidity—it’s about securing the resources to keep pace in an increasingly expensive technological arms race.