My call: the AI party can cool off and data centers still win

Nvidia just printed an $81.6 billion quarter, with $75.2 billion from data centers and a rounding error from the thing it used to be famous for, gaming. The stock market has decided this is not a company, it is a religion with 75 percent gross margins.

Here is the bet: in Nvidia’s most recently reported quarter before August 31, 2027, data‑center revenue will still be at least twice gaming revenue. If the AI bonfire fizzles, I expect lower flames, not a sudden LAN party comeback.

Put differently, the AI compute “asset class” story can prove wildly overhyped and still leave gaming looking like the quaint origin myth on the corporate museum tour.

The numbers: a ten‑step head start

Start with the scoreboard. Q1 fiscal 2027: $75.2 billion from data centers, roughly 92 percent of Nvidia’s revenue. Gaming is now the side quest. Depending on the quarter, data center is an order of magnitude larger than gaming, not a friendly peer.

Wall Street, never shy about stacking fairy tales on top of facts, now models Nvidia’s revenue jumping from about $216 billion in fiscal 2026 to nearly $700 billion by 2029. Most of that is supposed to be data center. The underlying assumption is simple and deranged: every hyperscaler spends more on AI capex every quarter, forever, and new buyers materialize to soak up another few hundred billion in GPUs on top.

That ramp is fantasy. The ratio is not.

To break my call, one of two things has to happen by mid‑2027: either data‑center revenue collapses so far that it falls within touching distance of gaming, or gaming has an historic, console‑plus‑crypto‑plus‑miracle‑software supercycle that closes a gap that started at well over ten to one.

Both occurring on the same clock would take divine intervention and an SEC filing.

Why data centers stay the main story, even in a hangover

The bullish script you already know by heart. Blackwell and Vera Rubin keep rolling, Rubin allegedly drops inference costs by up to 90 percent, every “frontier model company” signs up at launch, and Nvidia graduates from chip vendor to the central bank of AI compute, complete with a $500 billion financing platform and CME futures on GPU rental.

If that script holds, the question is not 2x. The question is whether data center stays 8 to 15 times gaming. My call survives that scenario with room to stretch.

The more interesting case, and the one consensus hates to look at, is the hangover. Right now, hyperscalers are burning through more than a trillion dollars of AI data‑center capex in 2026, with pressure to spend even more in 2027. AI revenues are not paying that tab. Google and Amazon’s cash flows are creaking, OpenAI and Anthropic are loss machines, and banks are starting to whisper that they are “choking” on data‑center debt.

At some point, boards notice that the magic AI money printer is still in beta and the only guaranteed winners are the folks selling 75 percent gross‑margin shovels.

When that realization hits, you get capex discipline. Which is a polite term for: someone finally says no to another $20 billion GPU order.

But even then, this is not a software subscription you can cancel next Tuesday. Nvidia’s data‑center revenue is wired into multiyear build plans, supply commitments that run into the hundreds of billions, and a global roster of governments and enterprises that have already promised their boards an “AI strategy.” That creates a thick floor. Volumes can flatten. Prices can fall. Margins can retreat from their current 65 percent operating peak. The installed base and long delivery pipelines keep data center a massive, recurring line item anyway.

For data center to sink to parity with gaming in that world, gaming would need to skyrocket in absolute dollars while AI build‑out decays like a failed mobile game. Nvidia’s actual behavior, however, says the opposite. The company has reorganized its soul around the data‑center stack: GPUs, networking, CUDA, systems. Gaming is treated less like a growth engine and more like a nostalgia brand that still sells hoodies.

The real risk: Wall Street is right on direction, wrong on dosage

Where the market is likely wrong is not in segment mix, it is in magnitude. To hit the rosy 2029 numbers, hyperscalers have to keep compounding an already absurd base of AI capex. Not just maintaining today’s $1 trillion‑plus pace, but escalating it, even as investors ask awkward questions about returns and regulators ask even more awkward questions about energy, competition, and China.

On the demand side, the math is uglier. Only a tiny handful of buyers, mostly the big clouds plus their captive model labs, currently buy compute at the necessary scale. Those labs are losing tens of billions of dollars a year. The clouds are starting to rely on debt to keep the party going. At the same time, every one of those customers has a standing incentive to build or buy an alternative: custom ASICs, AMD accelerators, anything that loosens Nvidia’s hand around their gross margin.

That is why I expect the “compute is a new asset class” narrative to age like DeFi: there will be serious infrastructure, serious money, and some honestly useful products, wrapped in a financialization story that turned out to be several sizes too big. Yes, Nvidia and CME can wrap GPUs in futures and structured financing. No, that does not repeal cycles, overbuild, or the basic physics of a balance sheet.

When the cycle turns, it will show up first in Nvidia’s numbers as slower data‑center growth, order slippage from a few hyperscaler whales, and creeping margin compression. But even a sharp deceleration leaves data center towering over gaming. The denominator in this bet is just too small.

How this call can be wrong

To break the 2x ratio by August 31, 2027, you need a very specific train wreck.

On one track: a genuine AI bust. A credit shock, or aggressive regulation, or a visible field of underutilized GPU barns that forces hyperscalers to slash AI capex hard and fast. Several of Nvidia’s top five customers, who currently account for roughly half or more of data‑center revenue, would have to pull back in unison, or flip aggressively to in‑house accelerators.

On the other track: a freak gaming boom. Think new consoles, a monster PC GPU cycle, and some consumer AI craze that really does run best on local graphics cards, all stacked on top of one another. Gaming would have to add tens of billions per quarter while data center is shrinking.

Could that happen? In theory, yes. In the same way it is possible that by 2027 everyone is doing their taxes inside a metaverse built on quantum blockchains. I assign low odds to both.

The scorecard for August 2027

This call resolves cleanly. On or just after August 31, 2027, we look at Nvidia’s most recent reported quarter. We take the reported data‑center revenue, we take the gaming revenue, we divide. If data center is at least 2x gaming, this column was right. If it is under 2x, it was wrong.

Watch a few signals on the way there: hyperscaler AI capex disclosures, especially any talk of “plateauing” spend; Nvidia’s data‑center margins, which will tell you when scarcity turns into negotiation; the mix of Nvidia versus custom silicon inside the big clouds; and whether Rubin arrives roughly on time and close to spec or shows up like a delayed AAA game with patch notes.

The consensus right now is that Nvidia can grow like an early‑stage SaaS startup while minting Apple‑tier profits and financing the entire AI industrial revolution on the side. I am more boring. I think the industrial revolution will go over budget, some CFOs will rediscover subtraction, and data‑center growth will slow from ludicrous to merely obscene.

And when that happens, the verdict will be simple. AI compute may not be the new oil, but by 2027 it will still be worth a whole lot more than watching your roommate play Fortnite at 240 frames per second.