TSMC Earnings Preview: CoWoS Crunch, H200 China Restart & Intel's CPU Comeback

**Jamie Cole** (0:00)
AI Hardware & Chips Daily News.
I'm Jamie Cole. Thanks for joining me.
Today, TSMC as AI Sector Oracle. TSC reports second quarter earnings on Thursday, and this particular set of numbers matters well beyond the company itself. Three things to watch, CoWAS packaging capacity, CAPEX guidance, and weather management raises the full year outlook. Those three signals, taken together, will tell you whether the AI infrastructure buildout is holding or quietly decelerating. The expected profit is around $19.65 billion, up roughly 59% year on year. That would be a record. The signal year isn't just the number, it's weather guidance moves with it. One piece of evidence already in hand, TSMC imposed 5-10% price increases across its 3-manometer, 5-nanometer, and 7-manometer nodes. Apple, Nvidia, AMD, and Qualcomm are all absorbing those hikes. Advanced nodes now represent 74% of TSMC's revenue. The key implication is that clients are paying more without walking away. That's not speculative demand, that's a capex cycle that's still running. The extension of those increases to 5- and 7-manometer nodes is the part worth noting. Earlier signals suggested cutting edge nodes only. The broader footprint tells you the inference acceleration, edge AI and hyperscaler workloads outside pure training are pulling real volume through mid-generation processes.
Investors who focus only on FAD capacity are watching the wrong constraint. COOS, the advanced packaging process that stacks memory directly next to compute, is the actual choke point in AI chip supply. TSMC controls this entirely. There's no merchant market, capacity has grown roughly 80% annually, and the target is 125,000 to 130,000 wafers per month by end of 2026
That roadmap, and whether Thursday's call revises it upward, matters more than transistor density debates. Thursday's call is also expected to offer updated detail on the A14 node, targeting 1.4 nanometers and a 30% power reduction. That's not shipping soon, but the roadmap signal shapes how customers plan long cycle CAPEX. TSMC CEO CCY has been framing the shift from generative to agentic AI as a driver of persistent, rather than episodic hardware demand. Agentic workloads require continuous compute, not just inference bursts. Here's the thing. Away from Thursday's report, there's a confirmed development in the export control space. Limited H200 shipments to China have begun under approved licenses. Three entities, including a ZTE subsidiary, have received deliveries.
Volumes are small, Chinese import approvals remain a timing uncertainty. The important distinction is that this moves the situation from aspirational to operational, however narrowly. It's politically significant, precisely because earlier guidance suggested expecting nothing.
Intel's position in this environment is worth a closer look. The company's data center and AI segment beat expectations in the first quarter of 2025 on the back of hyperscaler demand for host CPUs and AI server configurations. The inference workload shift is real and it favors CPU-centric architectures. Xeon is becoming the workload consolation ware as AI infrastructure matures beyond pure training economics.
That's not a dramatic reversal of GPU dominance, but it's a structural shift showing up in earnings, not just in analyst notes.
The European Commission approved 659 million euros in German semiconductor subsidies this week. The funding targets four facilities covering silicon carbide epiwafers, power MOSFETs, retrology equipment, and specialized detectors. This brings cumulative EU semiconductor investment above 14 billion euros. The strategic logic is supply chain resilience outside Taiwan's concentrated advanced node capacity. Whether European subsidies produce globally competitive fabs is a longer-term question, but the capital commitment is no longer tentative. After Thursday's call, the near-term watchpoints narrow to two things. First, does TSMC raise full-year guidance, and does CoOS capacity expansion accelerate with it? Second, does the H200 China volume remain contained, or does it widen as more import approvals clear? Those two data points will frame the next leg of the AI hardware cycle more clearly than most of what gets said between now and then. Thanks for listening. This podcast was built using AI technology. A YesWe production.