Nvidia will grow data center revenue over 40 percent while its multiple stays compressed through mid-2027

The prediction

Between now and July 31, 2027, Nvidia will report at least two quarters with data center revenue growth of more than 40 percent year over year, while its forward price-to-earnings multiple remains below its trailing three-year average for the majority of that period.

In plain terms: the earnings keep coming and the market keeps refusing to pay up for them. That gap is what I have called the durability discount — the market pricing AI compute earnings as a cycle rather than as a platform.

Why I think this happens

The setup as of July 2026 is unusually clean. Nvidia's most recent reported quarter showed revenue of $81.6 billion, up 85 percent year over year, with data center revenue of $75.2 billion, up 92 percent, a GAAP gross margin of 74.9 percent, and guidance for the next quarter of $91.0 billion — an acceleration. Committed 2026 hyperscaler capital expenditure runs roughly $660 to $690 billion, about double 2025, and is being revised upward rather than down.

Despite all of that, the stock was up only 7 to 9 percent year to date and fell 3.5 percent on July 17, losing the most-valuable-company title to an Apple that was flat on the day. The semiconductor index fell roughly 11 percent that week.

Three things arrived within 48 hours that I believe caused a durable re-classification rather than a temporary wobble. Moonshot's Kimi K3 took first on Arena's Frontend Code Arena with a 76 percent pairwise win rate as an open-weight model, demonstrating that frontier-class capability now diffuses. DeepSeek was reported to be developing its own chip, extending the pattern of major customers becoming competitors. And Gemini 3.5 Pro slipped months behind schedule on coding capability, hinting that the conversion of compute into capability may be getting harder.

Each of those attacks the scarcity premium that underwrites a 75 percent gross margin without touching near-term demand. That is exactly the condition that produces sustained multiple compression alongside continued revenue growth.

Confidence reasoning

I am at 60 percent rather than higher for three reasons.

Multiple compression is notoriously mean-reverting. Two or three more quarters of 90-percent data center growth could simply overwhelm the skepticism, and the market has been wrong-footed by exactly this company more than once.

The CUDA ecosystem is a deeper moat than most historical analogies allow for. In-house accelerator programs have been announced repeatedly for years and have consistently underdelivered against their stated ambitions.

And the buyer signal genuinely cuts against me. Hyperscaler capex is being raised, not cut, and that is the most concrete evidence available that the runway is longer than the market currently believes.

I am above 50 percent because the diffusion mechanism is structural rather than sentimental. Open weights do not get less available over time, and once a capability is downloadable the pricing power associated with its scarcity does not return.

Validation criteria

This prediction resolves correct if both of the following hold by July 31, 2027:

1. Nvidia reports at least two quarters in the window with data center revenue growth above 40 percent year over year, per its official quarterly results.
2. Nvidia's forward price-to-earnings multiple sits below its trailing three-year average for more than half of the trading days in the window.

It resolves incorrect if either condition fails — including the case where data center growth falls below 40 percent in three or more quarters, which would mean the market was pricing a real deceleration rather than applying a durability discount to healthy growth.

Key indicators to watch

• The interval between frontier model releases across the major labs. Stretching cadence supports the prediction; tightening cadence undermines it.
• Benchmark deltas between consecutive frontier generations at constant or rising training cost. Use third-party arenas, not vendor self-reports.
• How close open-weight releases track the closed frontier. Routine parity within a few months strongly supports the prediction.
• The first genuine downward revision to hyperscaler capex guidance. This would shift the situation from a re-rating to a forecast, and would likely mean data center growth falls below the 40 percent threshold — resolving this prediction incorrect for the opposite of the reason skeptics expect.

Full reasoning in The Durability Discount.