Intel +12%, AMD +107% & Google's TPU Goes Merchant | AI Hardware Briefing

Jamie Cole (0:00)

A Hardware & Chips Daily News.

I'm Jamie Cole. Thanks for joining me. Today, the CPU revenge, how Meta's Muse is rebalancing AI hardware demand.

Intel's stock jumped 12.1% this week, and the company doing the heavy lifting wasn't Nvidia. It was Meta. Meta's Muse AI agent platform is driving a certain server CPU demand that most of the market wasn't pricing in. The reason is structural. AI inference workloads don't just run on GPUs. They require CPU coordination for task scheduling, memory management, and orchestration across mixed node architectures.

Muse, which deploys AI agents at scale, is exposing exactly that. Intel's data center revenue grew 59% year-on-year to $6.3 billion. That's not a rounding error, that's a demand signal. The key implication is this. The GPU or nothing narrative that defined AI hardware over the past two years is cracking. CPUs are back in the conversation. Node is a replacement for accelerators, but is an essential partner to them. AMD is making the same case from a different angle. Data center revenue surged 107% year-on-year to $6.7 billion, driven by instinct accelerators and EPYC server wins, with hyperscalers including Meta. The annualized run rate is approaching $34.6 billion.

The stock is trading near all-time highs around $616.

Here's the catch. AMD's trailing price to earnings ratio is around 142 times. The growth is real. The execution risk is also real. 50% CPU growth targets and continued hyperscaler concentration are ambitious assumptions. If demand moderates even slightly, that valuation leaves very little room. The other development worth watching is Google. TPU version 8 is now being sold to external cloud customers, a meaningful strategic shift from Google's previous internal-only deployment model.

The efficiency claim is three times the training performance per watt, compared to version 7 South Korea has been designated as a major TPU business base. The signal here is that Google is moving from a captive accelerator model to something closer to a merchant model. That puts it in more direct competition with Nvidia for hyperscaler and enterprise workloads. Whether external customers adopt TPUs at scale, rather than just piloting them, is the unresolved proof point.

Here's the thing. TSMC gained 2.5 percent after analysts raised the fiscal year 2026 earnings per share forecast to 16.78 and fiscal year 2027 to 21.91. The underlying driver is sustained AI accelerator demand. TSMC is also opening a new supplier hub in Kaohsiung, targeting packaging bottlenecks, a practical capacity move that shortens supplier qualification cycles without requiring a new fab. Quarterly capital expenditure is running at $15.59 billion. The packaging constraint has been a quiet throttle on overall chip output. This facility is a direct response to that pressure. On policy, three bipartisan AI chip export control bills remain stored in the National Defense Authorization Act.

The proposals include location tracking requirements, allied supplier equipment controls, and an 18-month ban on the most powerful AI chips to adversaries.

Senate Democrats are pushing them forward. The technology industry through the Information Technology Industry Council is pushing back hard. The complicating factor is timing. These bills are moving while the Trump administration is in active talks with Beijing. And Nvidia's chief executive has a documented relationship with the Trump White House. The important distinction is between export control as national security architecture and export control as a negotiating chip. Right now, it's unclear which framing is winning. Away from the Washington dynamic, Chinese open-weight models are quietly gaining infrastructure layer status in emerging markets. Alibaba's Qwen and Huawei's hardware are now underpinning a $251 million supercomputer project in Brazil. Similar arrangements are advancing across Africa. Consider this. The pattern resembles the Linux moment in enterprise software. Once local ecosystems build services and data pipelines around the platform, switching becomes structurally expensive. If Qwen adoption becomes self-reinforcing in Brazil and across the Global South, the long-term competitive implications for US South AI firms are significant and genuinely unresolved. The near-term signals worth tracking are specific. Can Intel sustain data center revenue growth as MUSE scales, or does this call to reflect a one-time pull forward? Will Google secure external TPU commitments beyond pilots in South Korea? And will the NDAA export control bills survive into the next legislative window, or does the Trump XG dynamic sideline them entirely? The CPU revival is real. The policy environment is unstable. And the emerging market AI story is moving faster than most Western observers are acknowledging. Those are the three threads to watch.

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