The latest financial results from Taiwan Semiconductor Manufacturing Company (TSMC) show demand for AI and high-performance computing (HPC) is compelling and expected to stay that way – without detriment to existing capacity for networking giants.

That's according to Futurum research released this week, after July saw TSMC hit revenue of $30.07 billion in Q2 2025 to become the first Asian company in almost twenty years to surpass one trillion US dollars in market value.

Futurum research director Ray Wang highlighted AI and HPC’s 60% share of quarterly revenue as a historic milestone for TSMC, which underscores its role as the world’s go-to AI foundry.

In an interview with SDxCentral, Wang is “quite confident” that the same market demand will endure in the coming quarter.

“This is given the strong AI demand we are seeing driven by explosive inference deployments, global data center buildouts, and a global push for sovereign‑AI infrastructure.

“We have yet to see a sign suggesting a slowdown of this, and believe TSMC will continue to be the lead beneficiary of such a trend.”

Wang also said there are no signs of "overbuilding capacity" either, noting the flat demand for smartphones, which will make it easier for companies to project market demand and manage their capacity.

This should be good news for firms such as Broadcom and Marvell, who rely on external foundries such as TSMC for networking silicon.

While headlines have been taken up by AI-ready chips manufactured by TSMC for Nvidia and Apple, the majority of networking silicon today is not at this level yet.

While it may be true that TSMC’s revenue remains smartphone and HPC-heavy, TSMC still allocates “notable capacity” for networking firms, Wang explained, adding a caveat that CoWoS (Chip-on-Wafer-on-Substrate) are essential to AI chips.

“Broadcom, Marvell, and Intel are some of the biggest customers for TSMC today, so their ability to secure adequate capacity is not in doubt,” the analyst said.

“I think what could be a little more tricky would be TSMC's advanced packaging capacity - CoWoS, which so far has even more supply constraints than chip fabrication end.”

High-end chips with or without native AI capability are used in switches and data center routers. Any CoWoS bottlenecks would see blowback via chips like these from Broadcom, down to the likes of Arista, Cisco, and Juniper.

In Wang’s view, though, the CoWoS danger is mainly theoretical, saying TSMC "should still be able to serve its customers at its best capability."

That factor would likely sustain what the report highlights as TSMC’s pricing power in the market.

But with Nvidia, Apple, AMD, and Qualcomm accounting for a huge share of TSMC’s advanced-node revenue, how does the chipmaker manage pricing power without triggering customer insourcing?

In Wang’s view, TSMC achieves this through disciplined capacity management and sustained technology leadership.

“As [essentially] the only foundry able to deliver high‑volume, reliable, and customized nodes of production below  5nm, TSMC commands a strong pricing negotiation capability with leading‑edge customers.

"Persistent tightness in advanced‑node capacity bolsters TSMC’s bargaining power, allowing it to secure more favorable terms from customers vying for priority or higher volume allocations.

“Nonetheless,” he added, “TSMC has long been a customer‑first company. When it does raise prices, it typically does so very cautiously and only after extensive, collaborative discussions with clients.”