TSMC Raises Prices and Extends Bookings to 2028 While First Signs of Structural Disruption Emerge
As TSMC commands unprecedented pricing power with bookings through 2028, a startup's $800 million fundraise built on engineers poached from TSMC and Nvidia signals that the competitive dynamics around the world's dominant chipmaker are beginning to shift.
When Etched, an AI chip startup, disclosed this week that it had raised $800 million in a Series B at a $5 billion valuation — with over $1 billion in pre-orders already secured — the headline numbers were striking enough. More telling, though, was the sourcing of its more than 400 engineering hires: drawn substantially from Nvidia and Taiwan Semiconductor Manufacturing Company. The move signals that the AI infrastructure buildout has grown competitive enough to pull elite talent away from the two firms that arguably sit at its center, and it frames the broader question about TSMC's strategic position with unusual sharpness.
For now, that position remains formidable. TSMC reported 30% revenue growth in May 2026, a pace it has sustained since at least September 2025, when monthly revenues rose 31% year on year. Its order books are reportedly fully subscribed through 2028, and the company has converted its capacity constraints into pricing leverage: multiple reports from late June 2026 indicate that TSMC is raising wafer prices 5 to 10 percent across all advanced nodes — 3nm, 5nm, and 7nm — as well as on mature capacity, affecting customers including Nvidia, AMD, Apple, and Qualcomm. Advanced nodes account for roughly 75 percent of the company's wafer revenue, giving those price increases an outsized impact on earnings. TSMC has simultaneously raised its global semiconductor market forecast to $1.5 trillion for 2026, citing AI demand as the primary driver.
The company's expansion plans are calibrated to extend rather than merely defend that lead. According to industry reports from late June, TSMC is targeting a compound annual growth rate of 70 percent for N2 and A16 advanced-node capacity between 2026 and 2028, with first-year N2 output expected to run 45 percent higher than N3 achieved in its inaugural year. Advanced packaging — CoWoS and SoIC — is projected to grow even faster, at a CAGR exceeding 80 percent. TSMC is also advancing its next-generation CoPoS (Chiplets-on-Package System) architecture, which uses glass-core substrates that reportedly cut costs by 30 percent and push wafer utilization above 90 percent. On the mature-node side, the company has cut 28nm output by more than 25 percent since early 2026, reallocating wafer starts to advanced and specialty nodes where margins are materially higher.
The risks to this picture are real, even if none yet constitutes a near-term operational threat. The talent drain that Etched represents is modest in absolute terms against TSMC's tens-of-thousands-strong workforce, but it is qualitatively significant: engineers who leave foundry giants to build alternative silicon stacks carry with them process knowledge and customer relationships that take years to develop. The customer-concentration dynamic is also evolving: Apple is reportedly exploring Intel and Samsung as alternative AI chip suppliers to reduce its dependence on a single foundry, and TSMC's own capacity crunch has already pushed some customers to evaluate Samsung's expanding offerings. Compounding this is a regulatory overhang — reports from May 2026 indicate that potential Taiwan chip export restrictions remain an unresolved source of uncertainty, and any tightening of controls could affect shipments of leading-edge wafers to key customers. TSMC's recently announced packaging partnership with Amkor to extend its U.S. supply-chain footprint reflects both commercial opportunity and the political pressure accompanying its Arizona fab buildout.
Stepping back, TSMC's transformation from manufacturing contractor to the indispensable infrastructure layer of global AI compute has been a decade in the making, punctuated by inflection points that now read as a coherent ascent: surpassing Intel in quarterly revenue in mid-2022; absorbing successive Nvidia GPU generations — from Grace Hopper on the 4N process to the Rubin GPU, which taped out at TSMC in 2025 — and building the CoWoS packaging capacity that became the fulcrum of the 2023-2024 AI accelerator supply chain. AMD's chief executive stated publicly in 2023 that its MI300 GPU line could not be manufactured without TSMC. That dependency, replicated across custom silicon programs at Meta, Google, Amazon, and others, is what underwrites TSMC's current pricing power and near-100% fill rates.
Three signals are worth tracking in the months ahead. First, whether Etched and comparable startups can convert pre-order momentum into volume silicon production — doing so would require either TSMC's own fab services or a credible alternative foundry partner at the leading edge, neither of which is easily obtained. Second, how far Apple's foundry diversification actually progresses: exploratory supplier conversations are routine in the industry; actual transfer of leading-edge Apple silicon to Intel or Samsung would represent a structurally different threat to TSMC's customer concentration. Third, the trajectory of Taiwan's chip export controls: any escalation would introduce supply uncertainty that even a commanding two-year order backlog cannot fully insulate customers from. For now, the combination of pricing power, technology leadership, and locked-in demand through 2028 reflects a business operating near peak leverage — which is precisely when structural risks deserve the most careful attention.