Nvidia just posted one of the most impressive quarterly revenue figures in corporate history, pulling in $96.2 billion in its second quarter of fiscal 2027. And yet, buried in the filing, two details stand out: a $400 million charge tied to excess H200 GPU inventory, and the revelation that a single unnamed customer accounted for 16% of total quarterly revenue.
The H200 inventory problem
The H200 is one of Nvidia’s flagship GPUs built for large-scale AI training and inference workloads. It packs 141 GB of HBM3e memory and delivers the kind of bandwidth that makes data center operators salivate.
Yet Nvidia took a $400 million provision against excess H200 inventory during the quarter ending July 26, 2026. The precise cause remains somewhat murky: it could stem from demand shifts, production timing mismatches, or the geopolitical friction that has made selling advanced chips into certain markets a bureaucratic obstacle course.
For context, this is considerably smaller than the $4.5 billion charge Nvidia absorbed in Q1 of fiscal 2026. That earlier hit was tied to its older H20 inventory and directly linked to US export restrictions on chip sales to China.
Limited shipments of the H200 did reach selected Chinese customers, including ByteDance and Tencent, just before August 19, 2026. But those deliveries reportedly totaled roughly 10,000 units each, a fraction of the backlog Nvidia had previously accumulated, which exceeded two million units. That gap between supply and deliverable demand is where the $400 million write-down likely lives.
One customer, 16% of revenue
One unnamed account generated 16% of Nvidia’s Q2 revenue. On a $96.2 billion base, that works out to roughly $15.4 billion from a single buyer.
Nvidia doesn’t name the customer, which is standard practice in SEC filings when a company crosses the 10% disclosure threshold. The usual suspects in the AI infrastructure buildout include hyperscalers like Microsoft, Meta, Amazon, and Google, all of which have been spending aggressively on GPU clusters.
The concentration shows up most acutely in Nvidia’s Compute & Networking segment, which is where the data center GPU revenue sits.
Export controls continue to shape strategy
The $4.5 billion H20 charge from the prior quarter was the most dramatic example of regulatory exposure. The H200 situation suggests the friction isn’t limited to older product lines, with limited H200 deliveries to ByteDance and Tencent at volumes that are a rounding error compared to what was originally planned.
What this means going forward
The $400 million charge is manageable relative to Nvidia’s overall financial position. On $96.2 billion in quarterly revenue, it represents less than half a percent.
For investors tracking Nvidia, two things are worth watching closely. First, whether the H200 inventory situation resolves through eventual sales or requires further write-downs. Second, whether the 16% customer concentration figure trends upward or begins to diversify as more enterprises and sovereign AI initiatives enter the GPU market.