Nvidia Takes $4.5B Inventory Charge Due to Chip Trade Policy

Nvidia CEO Jensen Huang told analysts Wednesday (May 28) that recent U.S. export-control changes are reshaping the company’s growth plans in China even as global demand for artificial intelligence (AI) infrastructure accelerates.

The new rules require U.S. chipmakers to secure licenses before shipping certain advanced devices to China. Huang said the restrictions effectively remove a market he values at roughly $50 billion and have forced Nvidia to take a $4.5 billion inventory charge on its H20 processor, a version of its flagship AI chip designed to meet earlier compliance thresholds.

“We cannot further reduce the chip’s performance to stay within the regulations, so it is inventory we cannot sell or repurpose,” he said.

While acknowledging that Chinese firms will continue building AI capabilities, Huang framed the policy debate in competitive rather than political terms: “Export-control decisions should strengthen U.S. technology platforms,” he said, adding that the company remains engaged with policymakers and customers to find compliant paths forward.

But new U.S. export controls have effectively “closed” the $50 billion Chinese market to American companies, Huang said. Nvidia took a $4.5 billion charge for excess inventory on the H20 chip, which was made specifically for the Chinese market under Biden’s export rules.

Read more: Nvidia and Partners to Produce AI Supercomputers in US

However, the current administration now requires a license to ship those chips, which effectively ended this business for Nvidia. “We cannot reduce (the H20 chip) further to comply. As a result, we are taking a multibillion-dollar write-off on inventory that cannot be sold or repurposed,” he said.

Despite the China setback, Nvidia is pushing ahead with AI infrastructure projects worldwide to meet soaring demand for AI workloads. Its projects globally include:

• A 500MW AI infrastructure initiative in Saudi Arabia with Humain.
• Stargate UAE to develop an AI infrastructure cluster in Abu Dhabi in partnership with G42, OpenAI, Oracle, SoftBank and Cisco.
• AI supercomputer in Taiwan in collaboration with the Taiwanese government and Foxconn.
• Opened a research center in Japan to host the world’s largest quantum research supercomputer.
• TSMC is building six fabrication plans and two advanced packaging plants in Arizona to make Nvidia chips.
• Partnering with Foxconn to construct a 1 million-square-foot facility for AI supercomputers.

Huang said Nvidia’s goal is to have chips and systems “built in America within a year,” citing investments to bring back advanced manufacturing.

Read more: Nvidia and Amazon Land Middle Eastern AI Deals Amid Trump Visit

Nvidia is building infrastructure because demand for AI inference is soaring, according to Huang.

This refers to AI workloads in which new data is applied to trained models to generate outputs. With reasoning models, where AI agents think back and forth while accessing tools and doing other tasks, the number of tokens (or word clusters) they generate is exploding. More tokens means more compute power is needed, which is where Nvidia’s chips come in.

“The number of tokens reasoning goes through is 1,000 times more than a one-shot chatbot,” Huang said.

Other areas of growth include enterprise and industry AI. With enterprise AI, Huang said Nvidia will bring its AI infrastructure to companies, most of whom still keep their data on premises at their own data centers instead of the cloud. Nvidia has put together a compute, storage and networking solution and is getting ready to go to market with it.

For industrial AI, Huang said Nvidia offers advancements in robotics, omniverse and working to embed AI into manufacturing operations.

According to recent research note from Bank of America Global Research shared with PYMNTS, Nvidia is expanding its portfolio, with a broad set of AI products across cloud, enterprise, humanoid robots, workstation, gaming, software libraries, quantum and 6G. This is what underpins what the chipmaker has called the “multitrillion-dollar AI factory” industry, analysts said.

Read more: Nvidia Reportedly Aiming to Expand AI Business Beyond ‘Hyperscalers’

In its fiscal first quarter, Nvidia reported net income of $18.77 billion, or 76 cents per share, compared with $14.88 billion, or 60 cents per share, in the like quarter a year earlier. Revenue rose 69% to $44.06 billion, up from $26.04 billion year over year.

Data center revenue was up 73% to $39.1 billion in the quarter from a year ago. The company said AI inference workloads has taken off, soaring tenfold in one year.

The earnings results exceeded Wall Street’s expectations of net income of $18.2 billion in the first quarter on revenue of $43.3 billion. Shares of Nvidia rose by 6% to $143.03 in after-hours trading.

“Nations are investing in AI infrastructure like they once did for electricity and the internet,” Huang said. “The age of AI is here, from AI infrastructures, inference at scale, sovereign AI, enterprise AI and industrial AI. Nvidia is ready.”

Read more: Oracle to Buy $40 Billion Worth of Nvidia Chips for First Stargate Data Center

The post Nvidia Takes $4.5B Inventory Charge Due to Chip Trade Policy appeared first on PYMNTS.com.