Nvidia Powers Through China Setback as Inference Demand Ignites Next Wave of AI Growth

Jay's InsightThursday, May 29, 2025 9:00 am ET
3min read

Nvidia’s latest earnings report once again showcased its status as the bellwether of the AI revolution, with data center revenue surging and investor sentiment swinging bullish despite a noisy quarter marred by export control fallout. Shares climbed approximately 5% in post-earnings trading, buoyed by management’s resounding message that global demand for AI compute—particularly inference workloads—is not only intact but accelerating. While the company absorbed a $4.5 billion write-down due to the U.S. government's ban on H20 chips to China, investors and analysts appear more focused on the growth engine that is the Blackwell architecture and its rapidly expanding presence across cloud service providers, enterprises, and sovereign AI projects.

The headline numbers told a mixed story. Fiscal Q1 revenue rose 69% year-over-year to $44 billion, with adjusted EPS of $0.96, modestly beating expectations. Data Center revenue of $39.1 billion grew 73%, underscoring Nvidia’s dominant AI position. However, the results were skewed by the aforementioned H20 issues: $4.6 billion in revenue recognized before the April 9 export ban, a $4.5 billion inventory write-down, and an estimated $2.5 billion in lost Q1 revenue. Q2 guidance of $45 billion (plus or minus 2%) was just shy of consensus, but management emphasized that excluding the impact of lost China sales (~$8 billion), the guide would have easily surpassed expectations.

Despite these disruptions, analysts were largely positive. Truist raised its price target to $210, arguing that demand “from CSPs, enterprises, and sovereigns is still increasing.” Raymond James and Rosenblatt both highlighted the unexpectedly strong ramp in Blackwell production, with the latter noting Nvidia is “not stopping with Blackwell,” as Blackwell Ultra enters qualification and GB300 systems begin sampling. Citi’s takeaway was blunt: Nvidia “cleared the final hurdle of the China H20 ban transition quarter” and will likely break out of its recent range to make new highs. Deutsche Bank raised estimates on the belief that Nvidia’s China headwinds are now de-risked and margins will benefit from scale and architecture transition.

From the earnings call, the commentary on demand trends was relentless. CEO Jensen Huang described an explosion in inference workloads—specifically reasoning AI models—as a structural shift, saying these models require “hundreds to thousands of times more tokens per task than previous one-shot inference.” Microsoft alone processed over 100 trillion tokens in Q1, a fivefold increase. Nvidia’s GB200 NVL72 racks, now in full-scale production, are powering hyperscalers like Microsoft and Google at a pace of 1,000 racks per week—each housing 72 Blackwell GPUs. Huang emphasized that inference is now “a significant part of the compute workload,” transforming how data centers are built and scaled.

Networking strength was another pillar. Sequential growth of 64% to $5 billion marked a sharp rebound, driven by NVLink and Spectrum-X. Spectrum-X in particular is now annualizing over $8 billion, with new customers including Meta and Google Cloud. Huang positioned NVLink as a core “scale-up” platform while Spectrum-X brings InfiniBand-like capabilities to Ethernet, boosting utilization in AI clusters by up to 40%. These gains are far from trivial in multi-billion-dollar infrastructure investments.

While gaming, visualization, and automotive all posted healthy results, the company’s focus remains firmly on what Huang called “the next industrial revolution”: sovereign AI, enterprise AI, and industrial AI. There are now nearly 100 NVIDIA-powered AI factories in development globally, doubling from a year ago. Countries like Saudi Arabia, Taiwan, Sweden, and France are building national infrastructure with Nvidia at the center. “Every nation now sees AI as core to their digital economy,” said Huang. “We’re at the very beginning.”

Investors are also watching Nvidia’s evolving China strategy closely. The $50 billion China TAM remains off-limits under current rules, and Huang reiterated that the company has “limited options” to comply. He confirmed that Hopper cannot be further downscaled and admitted that no replacement product exists yet. But the market viewed this as an old problem already priced in, with analysts like Munster pointing out that Nvidia’s core growth story no longer depends on China.

Looking ahead, the top story to watch is whether Nvidia can sustain this Blackwell-fueled momentum amid growing competition and political scrutiny. The Senate recently raised national security concerns over Nvidia’s planned Shanghai office. Yet the company is doubling down on U.S. manufacturing with new factories in Arizona, Fort Worth, and Houston. Another key issue is the duration and magnitude of the inference wave: if reasoning models continue scaling token generation, Nvidia could capture a prolonged and lucrative upgrade cycle.

Finally, upcoming events like Citi’s Silicon Valley Bus Tour and GTC Paris in June may act as further catalysts, especially if Nvidia announces new sovereign AI partnerships or Blackwell Ultra milestones. As Truist noted, Nvidia’s CY26 EPS is now expected to reach $5.99, and the stock trades at a roughly 10x discount to other high-growth semis—leaving room for rerating. If demand holds and margins expand as expected, it’s hard to argue against the idea that Nvidia is not just weathering regulatory storms but leading the next wave of digital infrastructure.