- Nvidia and AMD had big export and chip manufacturing news this week
- But the impacts are likely to ripple well beyond the semiconductor industry
- China, and more specifically Huawei, could benefit in the end
A move by the U.S. to ban Nvidia and AMD from exporting AI chips to China could end up backfiring spectacularly, analyst Jack Gold warns.
The issues in the headlines this week largely centered on semiconductors – where they’re being made and where they’re allowed to be shipped. But as J. Gold Associates Founder and Principal Jack Gold told us, people talking about this news seem to be missing the forest for the trees.
“That’s one of the issues that people don’t think through. Right now, we’re talking about chips, but those are a component of a larger system,” he explained. In this case, those systems are the cloud and AI.
The U.S. government’s shortsighted move to stymie China and bring chip manufacturing back stateside could end up increasing costs and slowing domestic investments in AI and the cloud, Gold said. And by creating uncertainty around whether or not U.S. chip vendors can deliver, the U.S. could inadvertently create a massive opening for Chinese companies like Huawei to swoop in and pick off customers.
Nvidia, AMD and new chip export licensing requirements
But let’s back up for a moment. What exactly is happening?
Nvidia and AMD both kicked off the week by revealing the U.S. government has placed restrictions on certain chip exports to China and a handful of other countries. These vendors must now obtain special licenses before they can ship anything to these locations. The restrictions currently apply to Nvidia’s H20 GPU and AMD’s MI308 accelerator.
Both companies expect to take significant financial hits due to the change. Nvidia said it will record a $5.5 billion charge in the current quarter, while AMD expects a charge of up to $800 million.
At the same time, Nvidia and AMD both announced moves to manufacture their chips in the U.S.
Nvidia is investing in a million square feet of manufacturing space to build its Blackwell chips in Arizona and is also standing up plants to manufacture AI supercomputers in Texas. Production of its Blackwell chip has begun at TSMC’s plants in Arizona and manufacturing in Texas is expected to ramp over the next 12-18 months.
“It is clear that the company sees a world where diversity of supply chain and manufacturing is critical for resilience in fulfilling the needs of its customers,” Moor Insights and Strategy analyst Matt Kimball noted on LinkedIn.
AMD, meanwhile, is also partnering with TSMC to bring up production of its EPYC chips in Arizona.
“I like to see that it is expanding geographic support in these somewhat uncertain times,” Kimball said of the move.
While the licensing requirements aren’t great, the manufacturing news is a good thing, right?
If only it were so straightforward.
Catching up with chip production
The impacts are multifaceted.
First, Gold noted that if Nvidia and AMD can’t sell to China their revenues will suffer. Additionally, “it will give Huawei and others a big boost in selling their own increasingly competitive chips.” And not just selling locally in China either but “aggressively competing in the rest of the world.”
It’s worth noting that this is a muscle Huawei has already built up, having been forced to push its products domestically in China and then expand elsewhere after the U.S. banned its telecom equipment in stateside networks.
But even between TSMC and Intel – which is STILL trying to get production at its fabrication plant ramped up – the U.S. still doesn’t have enough capacity to bring all chip manufacturing here. It won’t be able to immediately remedy that either. As Gold noted, it takes three to four years minimum to build a new semiconductor factory and vendors aren’t exactly being incentivized to spend the $20-$40 billion required to do so since their addressable market is being restricted.
What the chip issues mean for the cloud and AI
Thus, chip prices seem poised to rise – whether because they’re made in the U.S. or due to tariffs for those being shipped from abroad. And that means the cloud and AI players using those chips will have to charge more for their services.
“When you raise prices you’re limiting the number of customers you can deal with,” Gold explained. “It’s a price-performance thing…What is the economic value that AI brings if the price of implementing it goes up?”
Think of chips like gas in a car. It’s a must have. If the cost of gas goes up, so does the cost to commute to work. That might make a driver reevaluate the need for non-essential trips (cough - cloud migration and AI implementation - cough) that require gas. It becomes a question of what’s a nice to have versus need to have.
Put it all together and we’re potentially looking at a world of hurt not just for Nvidia and AMD, but also for hyperscalers, AI vendors and U.S. AI ambitions as a whole.