NVIDIA's 6th Consecutive Beat Masks the Real Fracture in AI Chip Dominance

The Number That Matters Most

NVIDIA reported its sixth consecutive quarter of beating consensus estimates, and on paper the results are extraordinary. Revenue came in at $96.22 billion for the second quarter of fiscal 2027, up 105.8 percent year-over-year. Data center sales alone surged 117 percent to $89.02 billion. Gross margins held at 75 percent for a second straight quarter, and operating margins climbed to roughly 66.5 percent — a figure that signals pricing power so fierce it would make most hardware companies weep.

But the single most revealing sentence in the entire earnings release is buried in the guidance section: NVIDIA explicitly stated that Q3 revenue guidance of $108 billion does not include any data center sales from China.

That is not a footnote. It is a fault line.

The Engine Room

The Blackwell Ultra platform is now in full production ramp, and it is replacing the Hopper generation (H100 and H100-class chips) that has powered the AI boom through 2025 and into 2026. NVIDIA claims training speed improvements of up to four times over the prior generation and inference acceleration of thirtyfold. That translates directly into fewer chips needed per petaflop and tighter control over who gets access to what compute capacity.

The hyperscalers are still buying aggressively. Amazon Web Services disclosed a purchase of two million NVIDIA GPUs this quarter alone and is adding NVIDIA’s next-generation Vera CPU to its lineup. The five largest hyperscalers are projected by NVIDIA’s CFO to increase capital expenditure from $800 billion in 2026 to $1.3 trillion in 2027 — a 62.5 percent jump in a single year.

At that pace, $100 billion in quarterly revenue is not a milestone. It is a floor.

The Supply Chain Is a Triangle

Here is what the source article calls the “three-legged stool” of the modern AI accelerator: GPU, high-bandwidth memory (HBM), and advanced packaging. Each leg belongs to a different ecosystem with very different geopolitical risk profiles.

The GPU leg is NVIDIA itself, designing in California and fabricating on TSMC’s most advanced nodes in Taiwan. The HBM leg is dominated by Samsung Electronics and SK Hynix in South Korea. The packaging leg — co-packaged optics and 2.5D interconnects — lives primarily in Taiwan through TSMC’s CoWoS process.

All three legs sit on shaky ground, but in different ways.

TSMC’s foundry dominance means the United States has placed enormous strategic faith in a company operating in a potential conflict zone. Taiwan remains the single most concentrated point of fragility in the global technology supply chain, and no amount of NVIDIA earnings will change that calculus.

South Korean memory makers face a more subtle squeeze. Samsung and SK Hynix benefit from a surge in HBM demand, but Samsung’s foundry division continues to struggle against TSMC on leading-edge process nodes. The Korean government and chaebol leadership are acutely aware that their semiconductor destiny is split between memory — where they lead — and logic manufacturing — where they are years behind. Every quarter of strong HBM revenue temporarily deflates that anxiety. Every quarter of foundry underperformance reinforces it.

The China Problem That Isn’t a Problem (Yet)

NVIDIA has been navigating US export controls since 2022, when the Biden administration first restricted sales of advanced chips to China. The Trump administration, as the source notes, has moved to close remaining loopholes — effectively prohibiting even lower-tier AI chip exports to Chinese buyers. By Q3, NVIDIA expects to have lost meaningful data center revenue from mainland China entirely.

This creates a paradox. NVIDIA’s margins are at historic highs precisely because it has successfully rationed supply and maintained price discipline in an environment of intense global demand. But the company is simultaneously dependent on a market it can no longer fully serve. China is not merely a demand sink — it is a strategic competitor building its own chip ecosystem, with Huawei’s Ascend line and state-directed investment campaigns accelerating domestic alternatives.

The exclusion of China revenue from guidance suggests NVIDIA’s management views the restriction as structurally binding rather than cyclical. That changes how you value the business. A company that has lost one major market permanently does not deserve the same multiple as one that merely paused in it.

Who Wins, Who Loses

NVIDIA wins — for now. The company’s ability to sustain 66 percent operating margins while ramping a new product generation and absorbing geopolitical disruption is unprecedented in semiconductor history.

The hyperscalers win secondarily. AWS, Google, Microsoft, and Meta are locked into multi-year supply agreements with NVIDIA at prices that favor the chipmaker, but they are also building redundant compute pathways — Amazon’s Trainium, Google’s TPU, Microsoft’s Maia — that may one day reduce NVIDIA’s dominance from within.

South Korean memory stocks win conditionally. HBM is a commodity with premium pricing power in this cycle, but Samsung’s foundry troubles mean the upside is narrower than the headlines suggest.

Taiwan’s tech ecosystem wins and loses simultaneously. TSMC’s CoWoS packaging capacity is the bottleneck that determines how many Blackwell GPUs reach the market each quarter. That gives TSMC leverage, but it also makes the island an even more consequential target in any geopolitical scenario.

China’s domestic chip industry faces the sharpest headwind. Export controls are forcing a faster, messier decoupling that benefits no one’s efficiency but accelerates a parallel semiconductor track outside US jurisdiction.

What Happens Next

The market’s muted reaction to the blowout results tells you everything. Investors are not questioning whether AI demand is real — the data center numbers demolished that argument. They are pricing in the structural uncertainty beneath the surface.

NVIDIA’s next earnings will test whether the company can replace China revenue with growth elsewhere, or whether the export control regime is tightening further. The Blackwell Ultra ramp will reveal whether TSMC can scale packaging capacity fast enough to meet hyperscaler demand without quality degradation. And Samsung’s foundry results will determine whether South Korea’s semiconductor strategy regains credibility or continues to drift.

The six-quarter winning streak is impressive. But streaks end. The question is whether NVIDIA’s next down quarter arrives from a demand shock or a policy decision made in Washington — or Beijing.

The $108 billion guidance is a record. The silence around China revenue is the louder signal.