Record Profits and the Concentrated AI Network Behind Them

The S&P 500’s 29% earnings surprise falls to 11% without Google and Amazon. Much of their contribution came from unrealized investment markups, giving record profits an important asterisk.

Record profits are real, but not equally durable

With 88% of S&P 500 companies reported, second quarter revenue rose 15% and earnings rose 50% year over year. Projected margins reached 16.9%, full year 2026 earnings expectations climbed to $362 per share, and more companies are raising guidance.

These are strong underlying results, not merely an accounting illusion. But composition matters.

Google reported $98 billion in other income and Amazon reported $53 billion, reflecting unrealized markups on investments including SpaceX and Anthropic. Those gains are real under current accounting, but they are not equivalent to recurring operating earnings.

Anthropic’s rumored October IPO valuation is $2 trillion, compared with its last private funding valuation below $1 trillion. That suggests part of today’s reported profitability depends on private market valuations continuing to rise.

Several pillars rest on the same AI network

Reported figures indicate that OpenAI accounts for roughly $300 billion of Oracle’s approximately $640 billion backlog. It also represents about $280 billion, or 45%, of Microsoft’s backlog. Anthropic, meanwhile, is spending heavily with Amazon and Google.

Google’s and Amazon’s investment gains, Oracle’s and Microsoft’s backlogs, and demand from OpenAI and Anthropic may appear to be separate pillars of market strength. They are connected through the same small AI ecosystem.

Steve Eisman argues that large language model moats are shallow. If model competition produces a price war, the pressure may spread beyond private AI companies into hyperscaler revenue, equity valuations, and the estimated $400 billion of tech related corporate bond issuance this year.

That is a transmission risk, not evidence that a collapse is inevitable. AI demand may remain durable enough to support today’s backlogs, valuations, and investment gains. Neither a model price war nor a revenue collapse has been established.

High financing costs leave little room for disappointment

The 30 year Treasury yield remains above 5% even after 175 basis points of Fed cuts since September 2024. When financing costs remain elevated, the durability of cash flows and competitive advantage matters more, not less.

Transformative technology does not automatically produce attractive investment returns. The relevant question is whether a business can generate owners’ earnings at a price supported by those economics.

AI can transform the world while disappointing investors if today’s profits, backlogs, and valuations depend on financing and competitive advantages that have not yet been proven durable.