Intel Foundry Losses, China Export Controls & RTX 5090 Vanishes

Jamie Cole (0:00)

A Hardware & Chips Daily News.

I'm Jamie Cole. Thanks for joining me.

Today, can Intel's Foundry Math ever work?

Intel just posted its strongest revenue growth in 15 years. The data center segment is up 59% year over year, hitting $6.3 billion.

By almost any normal measure, that's a turnaround story. Here's the problem. The Foundry unit is still bleeding cash, and nobody has a clear answer for when that changes. That tension is the signal worth tracking right now. Intel's manufacturing bet is enormous. The company has committed more than $20 billion to building domestic fab capacity backed by Chips Act subsidies. And to be fair, there's a real positive data point this week.

The new Xeon 6 Plus server chips, built on Intel's 18A process, launched ahead of internal schedule. That matters. Process execution has been Intel's most persistent weakness for the better part of a decade. Getting ahead of targets, even by a small margin, is meaningful. The important distinction though, is between process progress and foundry economics. Intel isn't just trying to manufacture chips for itself. It's trying to compete with TSMC and Samsung for third-party customers. And those competitors have a 5-7 year CapEx advantage, plus scale that Intel can't replicate quickly. Being ahead of schedule on one node doesn't close that gap. It just means Intel is running a slightly faster race from further behind. Markets are making their own judgment. Intel and AMD both dropped 3% this week as investors rotated out of broader AI chip exposure. Nvidia held steady. That divergence is telling. The earlier thesis that a rising AI tide lifts all chip stocks is losing credibility. What's replacing it is a more selective view. Nvidia holds the monopoly position in GPU compute, and everything else has to prove its economics separately. For Intel, that means the foundry losses aren't just an accounting problem. They're a confidence problem. Revenue growth is real, but investors are watching whether the manufacturing bet can ever generate returns that justify the capital commitment. And right now, that proof point doesn't exist. Here's the thing meanwhile. The geopolitical picture got more complicated. China's Ministry of Commerce proposed export restrictions on 40 technologies this week, including 23 AI-related items and chip design tools.

A temporary registration requirement is already in effect. Final rules follow a consultation period closing after November 15

The key implication here is the direction of travel. The US has restricted advanced AI chip exports to China since 2022 China is now building a reciprocal framework. Whether this is a mirror or an escalation depends on enforcement, and that won't be clear for months.

But the structural trend is obvious. Two parallel technology stacks are forming, and every company with cross-border semiconductor exposure has to plan around that. On the GPU, the day, the publications of giving property convergers based address. The real question is whether this reflects supply chaos or deliberate allocation, with data center and professional GPU demand taking priority over gaming. Nvidia hasn't confirmed that, but the pattern is consistent with a company managing extreme demand against limited TSMC capacity. Consider this. One more development worth noting, Anthropic signed a 20-year lease for a data center campus in Nova Scotia, Canada. Phase 1 targets 95 megawatts by late 2027, scaling to 420 megawatts by 2028, with 2 billion in wind power integration. The signal here is strategic geography. Power availability and cost are becoming decisive factors in where AI infrastructure gets built. The two things worth watching closely from here, whether Intel can show any path to foundry profitability before the CapEx burden becomes unsustainable, and how China's export control frame will take shape after November 15th. Both will have real consequences for how the semiconductor industry organizes itself over the next several years.

Thanks for listening. This podcast was built using AI technology, a YesWe production.