AMD confirmed in a Wednesday investor briefing that its sixth-generation EPYC server processor, codenamed Venice, has entered volume production on TSMC's 2nm process technology at the foundry's Hsinchu fabs. Venice is the first high-performance computing product anywhere in the industry to enter production on the 2nm node, which TSMC began commercial qualification on earlier this year, and the first non-mobile silicon to ship on a sub-three-nanometre process.
AMD chief executive Lisa Su told the briefing that Venice samples had been delivered to "key hyperscaler customers" since the start of the second quarter and that volume cloud deployments were on track for the second half of the year. Su confirmed that AMD will also ramp Venice at TSMC's Arizona fabrication facility starting in calendar 2027, in keeping with the company's previously announced US production targets.
The Venice generation is positioned as AMD's primary competitive response to Nvidia's vertically integrated Grace-Blackwell stack and to Intel's delayed Granite Rapids successor. AMD said Venice would deliver roughly 1.6x the performance-per-watt of the current Turin generation on standard cloud-server workloads, and that the company's long-standing eighty per-cent share of the high-end x86 server market would "be defended aggressively" through the cycle.
TSMC's 2nm node, formally called N2, uses gate-all-around nanosheet transistors and represents the foundry's first move away from the FinFET architecture that has carried it through five generations. The 2nm process is currently yielding roughly seventy per cent at the high-volume tile size used in Venice, according to industry tracker SemiAnalysis, which is broadly in line with where the 3nm node was at similar volume ramp.
AMD also previewed its follow-on data centre CPU line, named Verano, which will also use the 2nm node but in a denser core configuration and with the company's next-generation CCD interconnect. Su said Verano was "on track" for production sampling in the first half of 2027 and that AMD's product cadence would continue to track its established eighteen-month rhythm.
The market reaction to the announcement was muted on Wednesday, with AMD shares closing 1.1 per cent higher in New York, but the stock gained a further 3.4 per cent on Thursday after Bank of America raised its price target to $245 from $220 on what the broker called "execution on the most ambitious foundry transition in industry history." AMD trades at roughly 38 times forward earnings, a premium to its five-year average but well below Nvidia's 51 multiple.
TSMC, which is the only foundry currently producing at the 2nm node, said in a separate statement that overall N2 wafer demand had been "very strong" through the first commercial quarter and that capacity would be expanded "as rapidly as equipment delivery allows" through 2027. The foundry has been allocating N2 wafers preferentially to Apple, AMD and Nvidia, with Qualcomm and MediaTek slated for the second wave starting in late 2026.