Nvidia Doubles Revenue to $96.2 Billion — and Says 70% Growth Next Year Is All Its Supply Chain Can Deliver
Q2 FY2027 revenue hit $96.2B, up 106% year over year, with data center alone at $89B. Jensen Huang's message: 'compute is revenue,' demand is accelerating, and the constraint is supply, not customers.
Nvidia reported $96.2 billion in revenue for its second quarter of fiscal 2027 — up 106% from a year ago and 18% sequentially — and then delivered the real headline on the call: management guided fiscal 2028 revenue growth to approximately 70%, a number CEO Jensen Huang described as supply-constrained rather than demand-limited. “Even though our demand is much greater than 70%, our supply allows us to confidently deliver 70%,” he told analysts.
The composition of the quarter is familiar and extreme. Data center revenue reached $89.02 billion, up 117% year over year, now roughly 93% of the entire company. Adjusted earnings per share came in at $2.22 against the $2.10 analysts expected. Gaming, networking, automotive — everything else Nvidia does — fits inside the rounding error of its AI accelerator business.
Huang’s framing was the quotable part: “AI has reached its inflection point. It’s doing useful work. Its tokens are productive and profitable. Now, compute is revenue. And demand is accelerating.” That last sentence is the one that matters for anyone pricing the AI trade — this is the company with the deepest visibility into global AI capex saying the curve is still bending upward, not flattening.
The cracks worth watching
It wasn’t a flawless print. Gross margins slipped on rising memory costs — HBM supply is the industry’s tightest bottleneck, and Samsung and SK Hynix have been raising prices into a fully booked market. When your product is 93% data center silicon wrapped in expensive stacked memory, HBM inflation lands straight on your margin line. A supply-constrained 70% growth forecast also cuts both ways: it’s a promise of scarcity, which props up pricing, but it means Nvidia is leaving demand on the table for rivals — AMD, Broadcom’s custom-ASIC programs, Cerebras — to absorb.
Context makes the numbers stranger. A year ago, bears argued hyperscaler capex had to cool. Instead, Nvidia doubled. The company that reported $46.7 billion in Q2 a year ago just did $96.2 billion, and its CEO is telling you the limiting factor for next year is how many wafers, HBM stacks, and CoWoS slots TSMC and the memory makers can physically produce.
For developers, the practical translation: compute stays expensive and rationed through at least 2027. If inference costs are a line item in your product’s unit economics, Huang’s “compute is revenue” applies to you as the buyer. Efficiency work — smaller models, caching, batching, right-sizing — keeps compounding in value while the supply curve, not the demand curve, sets the price.
The AI buildout’s scoreboard company says the game is accelerating. Believe the guidance or not, every cloud bill you’ll pay next year was just priced by it.