NVIDIA Just Hit Its 13th Straight Record. The Real Question Is What Comes Next.

The Number No One Is Discussing

NVIDIA just posted its 13th consecutive quarter of record revenue. That is not a footnote. It is a signal that the AI infrastructure buildout has not slowed — it has simply shifted from speculative commitment to operational reality.

Second quarter FY2027 revenue came in at $96.2 billion, beating consensus estimates of $92.2 billion by roughly $4 billion. Net income reached $59.7 billion, up 126% year over year. Data center revenue alone hit $89 billion — 93% of total revenue — growing 117% from a year ago and 18% from the previous quarter. Gross margin held at 75%. Jensen Huang, NVIDIA’s CEO, confirmed that the next-generation Vera Rubin platform has entered production.

All of this is spectacular. But the story hiding inside the headline is in the Q3 guidance, and it is more revealing than anyone is letting on.

The $108 Billion Guidance That Excludes China

NVIDIA guided for third-quarter revenue of $108 billion. On its face, that is a 12% increase from the current quarter and suggests demand shows no signs of fading. But the company attached a qualifier that matters more than the number itself: this outlook explicitly excludes data center revenue from China.

That is not an accounting footnote. It is a strategic boundary line. NVIDIA is telling the market it has consciously decoupled a significant portion of its growth story from Chinese data center spending, likely reflecting both export controls and the company’s own risk assessment of that market.

The implication is twofold. First, the $108 billion figure should be read as a floor for non-China demand, not a ceiling. Second, the Chinese gap represents real revenue NVIDIA could have captured — and the company’s silence on that number is a deliberate act of transparency. Most tech companies bury that kind of exclusion. NVIDIA put it in bold.

Who Actually Wins When Data Center Revenue Dominates

For years, NVIDIA’s valuation was anchored to gaming and data center as separate growth engines. Now data center accounts for 93% of revenue and is growing faster than any other segment in the company’s history. This is not a diversification story anymore. It is a monopoly story — one built on the combination of CUDA software lock-in, custom silicon design cycles that take years to replicate, and the sheer capital intensity of building advanced packaging capacity.

The winner here is straightforward: NVIDIA captures pricing power that would be impossible in any competitive semiconductor market. At 75% gross margins on what is effectively a commodity-adjacent product, the company is earning like a software platform selling to infrastructure buyers who have no fallback.

Samsung and SK Hynix also win — but conditionally. NVIDIA’s guidance and Huang’s comments point directly to increased HBM demand tied to Vera Rubin’s production ramp. Both Korean memory giants are pushing HBM4 (the sixth generation of high-bandwidth memory) into volume. If NVIDIA’s accelerator shipments continue climbing, Samsung and SK Hynix become secondary beneficiaries of a demand curve they did not create but are structurally positioned to serve.

The risk for them is timing. HBM is a margin-intensive business, and the second-source qualification cycle is long. Whoever ramps fastest on HBM4 while NVIDIA moves to its next architecture after Vera Rubin will capture disproportionate value. Right now, that race is tight.

The Real Question: Is the Capex Cycle Inflecting?

Here is where most analysts are wrong. They look at 13 straight record quarters and assume the cycle is durable because the numbers keep compounding. But compounding is not the same as accelerating.

The QoQ data center growth rate of 18% in this quarter is meaningful, but it is a deceleration from the triple-digit year-over-year growth that dominated earlier periods. The market is not pricing in a cliff — it is pricing in a gradual flattening. That is the more accurate read.

Big tech companies have committed tens of billions to AI infrastructure, but commitments are not purchases. The distinction matters because NVIDIA’s next few quarters will reveal whether capital expenditure plans are being executed or renegotiated. If hyperscalers hit budget constraints or pivot toward more efficient models that require fewer chips per training run, NVIDIA’s growth trajectory changes fast.

What makes this cycle different from previous infrastructure booms is the software layer. CUDA is not just a moat — it is a dependency. Every model trained on NVIDIA hardware creates switching costs that compound over time. That is why the demand curve is flatter but potentially longer than previous hardware cycles.

What Comes Next

NVIDIA returned approximately $26 billion to shareholders this quarter through buybacks and dividends, with roughly $99 billion in remaining authorization for share repurchases. That is a company confident enough in its forward position to return capital at scale while simultaneously investing in production ramp for its next architecture.

The Vera Rubin ramp is the near-term catalyst. If shipment volumes meet expectations and HBM demand from Samsung and SK Hynix scales accordingly, the next two quarters could reinforce the current trajectory. But the $108 billion guidance excluding China also sets a high bar — and any miss against that number, particularly if driven by softness in non-Chinese data center demand, would be the first crack in the 13-quarter streak.

The market is not yet pricing in the inflection point because inflection points are invisible until they arrive. What this quarter shows is a company operating at peak execution in what is still, by every measurable metric, a supersonic growth cycle. The question is no longer whether AI infrastructure spending is real. It is how long the current pace sustains before the law of large numbers and budget constraints begin to bite.

NVIDIA’s numbers say not yet. The guidance says the same — just without China.