After nearly a year of regulatory limbo, Nvidia (NVDA) is finally poised to resume shipments to China, its largest semiconductor market. CEO Jensen Huang confirmed at the company’s GTC conference this week that export licenses have been secured and purchase orders from Chinese customers are already in hand. “We’ve been licensed for many customers in China,” Huang said. “We’ve received purchase orders, and we’re restarting manufacturing. Our supply chain is getting fired up.”
The announcement marks a dramatic reversal for the chip giant, whose China business was effectively frozen in 2024 when the Trump administration imposed strict export controls on high-performance AI chips, later adding a requirement that 25% of proceeds be remitted to the U.S. government. As recently as late February, CFO Colette Kress told analysts that “we have yet to generate any revenue” from China-bound products and expressed uncertainty about when—or if—imports would be allowed.
Nvidia isn’t simply dusting off old inventory. According to sources familiar with the matter, the company is developing a variant of its recently acquired Groq AI chips specifically for the Chinese market. Last year, Nvidia paid $17 billion for Groq, an AI chip startup, and the technology is now being adapted to meet both performance needs and regulatory constraints.
The new chips will focus on inference—the stage where trained AI models answer questions, write code, or execute tasks—rather than the training workloads where Nvidia dominates globally. At GTC, Nvidia outlined a dual-track product strategy: for global markets, it plans to pair its forthcoming Vera Rubin chips (which cannot be exported to China) with Groq chips; for China, the Groq‑based variant will shoulder inference workloads, potentially filling a gap left by export restrictions.
One source told Reuters that the China‑bound chips are not simply “downgraded” versions, but rather adaptable variants designed to work with other systems. The new chips are expected to be ready by May.
The market opportunity is substantial. Huang has previously estimated China’s AI chip market at $50 billion annually. Before the export clampdown, Nvidia’s China revenue in 2024 stood between $12 billion and $15 billion. A successful relaunch could push those numbers significantly higher.
“Despite continued U.S. export controls, Nvidia is finding a practical path to serve China by offering products that comply with regulations,” said Bernstein analyst Stacy Rasgon. “The key question is whether the performance and interconnectivity of these Groq‑based chips will meet the demands of China’s leading AI customers. If inference performance holds up, Nvidia could regain a foothold in China’s inference market.”
That market is increasingly contested. While Nvidia remains the undisputed leader in AI training, domestic players like Baidu have developed their own inference chips. Nvidia’s edge lies in its mature software ecosystem and seamless integration, which could sway customers despite local alternatives.
Wall Street is already pricing in robust growth. Consensus estimates project Nvidia’s revenue over the next 12 months at $368 billion, a 70% increase. Any contribution from China would add further upside.
At a forward price‑to‑sales multiple of about 22x, Nvidia’s valuation remains attractive, according to KeyBanc Capital Markets analyst John Vinh. “Given the company’s moat in AI acceleration and the potential incremental revenue from inference in China, we see room for upside surprises in coming quarters,” Vinh said.
With the May launch window approaching, all eyes are on whether Nvidia can replicate its global AI dominance in China. For the semiconductor industry, this comeback story is only just beginning.
