#英伟达财报周 Nvidia's earnings report exploded, but…

Nvidia turned in a perfect score
Early this morning Beijing time, Nvidia released its fiscal second-quarter earnings report for the quarter ended in July: revenue was $96.2 billion, up 106% year on year; data center revenue was $89 billion, up 117%; its after-hours share price surged more than 4.5% at one point.
Even more impressive was the guidance—the third-quarter revenue outlook was $108 billion, again above market expectations; the CFO directly threw out the line that “fiscal 2028 revenue growth will be around 70%,” while the market had previously dared to expect only 45%. Huang Renxun's exact words: “Computing is revenue, and demand is accelerating.”
In one sentence: the “class monitor” of AI computing power has once again left the entire class in the dust.

But first: that 70% growth is anchored to the “supply ceiling,” not “limitless demand”
A lot of people got carried away when they saw 70%. I came across an analysis from a Korean brokerage and realized something was off—it put it bluntly: Nvidia's 70% growth expectation is based on a “supply ceiling” (capacity is maxed out, so it sells as much as it can produce), not on demand truly exploding without limit.
Put simply, this is “how much I can sell depends on how much I can make”—growth constrained by capacity, not growth with demand breaking through the ceiling.
These are two completely different stories, with entirely different valuation logic. One is “orders are chasing me,” while the other is “I can't supply enough even if I produce at full speed.”
The latter sounds more powerful, but the risk is that if capacity catches up one day while demand does not, the story will reverse.

But second: gross margin is quietly slipping
Third-quarter gross margin guidance is 74%, below the market expectation of 75%. Don't underestimate this one percentage point.
Vera Rubin has just entered full-scale production, and the costs of the new production lines have not yet been diluted; profitability will initially decline slightly.
Even a perfect scorecard lost points in this subject.

But third: on the same night, someone was expanding a “printing press” at the other end of the school building
Kioxia announced last night that it would invest more than ¥1 trillion to build a new NAND flash memory plant in Iwate Prefecture, Japan. Note that SK Hynix and Samsung rose 5% and 3% in Korean trading today because AI-specific HBM memory is in short supply; but Kioxia is expanding traditional NAND, which is a different track. Global storage stocks were broadly higher today—don't treat all “storage” as the same thing: some are being fed by AI, while others are expanding their own capacity. The two tracks will eventually part ways.

The “rain” in the market has already fallen today. At tomorrow's opening, we'll see who brought an umbrella. $NVDA ‌