TSMC has sold out its manufacturing capacity through 2028, according to Piper Sandler, with the analyst firm pointing to sustained demand for AI-related silicon as the primary driver of the multi-year backlog.

The call, reported by Investing.com India, lands at a moment when the entire advanced-node supply chain is straining under orders for AI accelerators, high-bandwidth memory, networking chips and the foundry slots needed to build them. A sold-out order book stretching three years ahead signals that TSMC's leading-edge capacity — the N5 and N3 families and the N2 ramp now underway — is fully committed before the first high-volume 2-nanometre wafers have shipped.

What does a sold-out book through 2028 actually mean?

In practical terms, it means customers cannot secure additional leading-edge wafer allocations on TSMC's current timeline. Any new demand must either wait in line, displace someone else's allocation, or move to a second source.

For buyers of AI silicon, the constraint is structural rather than cyclical. Hyperscalers and AI chip designers have committed to multi-year roadmaps of accelerator products, and each new generation consumes more advanced-node wafers per unit than the one before. Larger reticle sizes, bigger die pairs and chiplet-based packaging multiply the wafer count behind every training cluster a customer plans to deploy.

For TSMC, the report reinforces the durability of the pricing power that the company has exercised across recent years. A foundry that cannot meet demand until 2028 does not discount. It negotiates allocation, terms and prepayments.

Who feels the squeeze first?

The pressure propagates down a clear hierarchy of customers:

• AI accelerator vendors, whose products sit at the top of TSMC's allocation priorities and whose volumes keep climbing;
• Fabless designers of networking, switching and serdes-heavy silicon that ride the same AI infrastructure build-out;
• Smartphone and PC chipmakers, who compete for the same advanced nodes during cyclical upturns;
• Smaller fabless firms, which face the longest queues and the weakest leverage when capacity tightens.

The report's framing — capacity spoken for through 2028 — implies the constraint extends beyond the current generation of AI chips into at least the next two product cycles. That horizon covers the N2 ramp and the early years of the A16 generation, the node family TSMC has slated for the second half of the decade with backside power delivery.

Does this change the competitive picture?

A three-year sold-out position strengthens the argument that leading-edge foundry demand has entered a structurally higher regime. AI training and inference demand has not followed the sharp correction that mobile and PC markets experienced in recent downturns. Instead, capital expenditure commitments from hyperscalers keep compounding, and each commitment converts directly into wafer allocations.

The dynamic also raises the strategic value of every competing option: Intel Foundry's external ambitions, Samsung Foundry's advanced-node roadmap, and the capacity expansions TSMC itself is funding in Arizona, Japan and Germany. Capacity announcements made now will not relieve a 2028-bound backlog overnight — new fabs take years to qualify and ramp — but they define who can absorb the demand that TSMC cannot.

For investors, the Piper Sandler assessment frames TSMC less as a cyclical semiconductor stock and more as an infrastructure bottleneck with a locked-in revenue profile. For procurement teams across the electronics industry, it converts a market rumor into a planning assumption: leading-edge capacity is a scarce, allocated resource through at least 2028, and product roadmaps that assume otherwise need revision.

The report does not specify pricing, node-level breakdowns or named customer allocations. What it does establish is the headline condition: the world's dominant advanced-node foundry has nothing left to sell for three years, and AI demand is the reason.