S&P 500 earnings grew roughly 48% this past quarter, a number strong enough to make almost any investor pause and double-check it. According to a CNBC analysis published August 3, 2026, a meaningful chunk of that growth wasn't earned. It was a paper gain, and it traces directly back to a handful of Big Tech companies holding equity stakes in Anthropic and OpenAI, two private AI labs whose valuations have kept climbing even as their businesses remain unprofitable.

Strip out those specific gains, and the picture changes considerably. According to Tajinder Dhillon, LSEG's head of earnings and equity research, removing investment gains tied to Alphabet and Amazon's private company stakes alone brings aggregate earnings growth down to roughly 29%, a figure much closer to the 24% growth analysts had originally forecast for the quarter.

How a Software Company's Profits Depend on a Startup's Valuation

The mechanism behind this is a straightforward accounting rule with an increasingly outsized effect. When a public company holds an equity stake in a private business, it has to periodically mark that stake's value to reflect the private company's latest valuation, even though no shares actually changed hands and no cash came in the door. As Anthropic and OpenAI valuations have surged past levels each now sits just south of a trillion dollars, the paper value of stakes held by their public-company investors has surged right along with them, and that increase flows straight through to reported earnings.

Microsoft and Amazon are both private shareholders in OpenAI and Anthropic, while Alphabet holds its stake specifically in Anthropic. All three companies posted results this earnings season carrying real, quantifiable boosts from exactly this dynamic. Amazon's most recent quarter included a genuine investment gain tied to its Anthropic stake. Alphabet's reported earnings growth reached nearly 300% year over year, a figure driven substantially by mark-to-market gains on its own private holdings rather than organic growth in its core advertising, cloud, or search businesses. Microsoft's results included their own smaller but still material gain tied to its Anthropic position.

Real Numbers Attached to a Fictional Sale

What makes this distortion different from ordinary earnings noise is scale. These aren't rounding errors buried in a footnote. Individually, each of these gains was large enough to meaningfully move the specific company's headline earnings-per-share number, and collectively, across three of the largest companies in the index, they were large enough to shift the entire S&P 500's aggregate growth rate by nearly 20 percentage points in a single quarter. That's the kind of swing that changes how strong an earnings season looks in the financial press, even though the underlying operating businesses, cloud revenue, ad sales, software subscriptions, grew at a considerably more modest, if still healthy, pace.

Gil Luria, managing director and head of technology research at D.A. Davidson, put the dynamic plainly to CNBC, noting that these types of moves tend to even out over time. That's the core reason most professional analysts already treat these gains as noise rather than signal: valuation swings on private holdings are inherently temporary markers, subject to reversal the next time a funding round, IPO, or secondary transaction resets the underlying company's price, whether up or down.

Why Analysts Aren't Panicking, Exactly

To their credit, most Wall Street analysts have been factoring this in already. Non-GAAP earnings estimates and forward-looking forecasts typically exclude one-time items like unrealized investment gains specifically because they don't reflect a company's actual operating performance. The practical consequence this quarter wasn't that analysts were fooled, it's that the reported, headline GAAP numbers generated more earnings upside surprises across the market than the underlying businesses alone would have produced, a gap between the story investors read in a press release and the story a more careful reading of the balance sheet actually tells.

Importantly, LSEG's analysis doesn't suggest these companies' core businesses are performing poorly. Even with the investment gains excluded, aggregate earnings growth in the high 20s remains a genuinely strong quarter by historical standards. The concern isn't that Big Tech's fundamentals are weak. It's that the fundamentals are getting obscured by a separate, much larger number that has nothing to do with how well any of these companies actually ran their core operations this quarter.

The Risk Sitting Inside a Number That Can Reverse

The uncomfortable implication is symmetrical. If Anthropic or OpenAI's private valuations keep climbing, the same accounting mechanism keeps flattering Microsoft, Amazon, and Alphabet's reported earnings in future quarters. But if either lab's valuation stalls or drops, whether from a funding round that prices lower than the last one, a public listing that undershoots private-market expectations, or broader AI investor sentiment souring, the same mechanism works in reverse, producing paper losses that would drag reported earnings down without reflecting any actual deterioration in Microsoft, Amazon, or Alphabet's real businesses either.

That's a genuinely awkward position for investors trying to evaluate these companies cleanly. A metric that can swing a company's headline earnings growth by dozens of percentage points, in either direction, based entirely on a third party's private valuation, is a metric that requires real scrutiny before anyone treats it as a signal of business health.

What This Reveals About How Tangled the AI Industry Has Become

This story fits a broader pattern that's become impossible to ignore across the AI industry this year: the largest tech companies aren't just building AI products, they're deeply financially entangled with the AI labs they compete against, partner with, and increasingly, invest in directly. Google has a stake in Anthropic while also competing against it with its own Gemini models. Microsoft and Amazon both hold OpenAI and Anthropic positions while running their own competing AI efforts internally. Untangling how much of any given company's reported success comes from its actual product execution versus the rising private-market tide lifting every AI-adjacent balance sheet has become genuinely difficult, and this quarter's earnings season made that difficulty impossible to dismiss as a minor accounting footnote.