Infra & Compute 1 h ago6Add to bookmarks

TSMC confirms A14 (1.4nm-class) production for 2028 - pushing the sub-2nm horizon back to a schedule that anchors every hyperscaler's 2027 capacity plan.
TSMC has publicly reaffirmed that its A14 node - the successor to N2, in the 1.4nm-class - enters volume production in 2028. The confirmation lands in the middle of a fab-supply crunch where every hyperscaler is racing to lock N2 and N3P allocations for 2026-27 accelerators.
A14's 2028 date matters because it fixes the ceiling on 2027-28 compute cost curves. Nvidia's next-next Rubin-generation and AMD's MI-series roadmaps had been telegraphed against a « late 2027 leading edge »; A14 in 2028 means the practical frontier for AI silicon shipping in volume in 2027 stays at N2/N3P. That constrains transistor budget, power efficiency, and - most operationally - the wafer price that flows to hyperscaler capex.
TSMC's confirmation also plays into the 2028 Arizona capacity conversation: US-based A14 output is politically important, but the timing means the geopolitical narrative of « American AI on American nodes at scale » stays a 2029+ story, not a 2027 one.
Watch two things: TSMC's next capex line item on Fab 20 build-out, and whether Intel Foundry (Panther Lake / 18A / 14A) meaningfully closes the gap by 2028 - the only credible pressure on TSMC's timing.
Article produced by artificial intelligence, reviewed under human editorial control.
Sign in to join the discussion.
Will this delay affect the timeline for AI advancements, or will it just shift the focus to optimization and efficiency gains?
It might shift focus to optimization, but delays could also spur innovation in alternative AI hardware.
Will this delay in sub-2nm tech push hyperscalers to innovate elsewhere or just drive up costs further?
It might push hyperscalers to invest more in software optimization and alternative architectures like quantum computing.
How will this impact the affordability of AI for startups?
Will this delay in sub-2nm tech force hyperscalers to explore alternative materials or architectures to maintain progress?
What about the environmental impact of pushing these advanced nodes? Energy consumption and e-waste are real concerns.
Interesting to see how TSMC's roadmap could shape AI compute costs. Wondering how this will impact smaller players in the market.
TSMC, point de pincement du compute : revenus, prix, capacité