As TSMCs 2nm mass production countdown begins, speculation mounts over its major clients. KLA, a key chipmaking equipment supplier, reports that around 15 customers are designing at the N2 node, including roughly 10 high-performance computing (HPC) players, according to Investing.com, etnews and MyDrivers.
www.trendforce.com, Sept. 22, 2025 –
As per the transcript from Investing.com, which cites KLA CFO Brent Higgins at Goldman Sachs Communicipia on September 10th, from a design perspective, the company expects N2 to be the largest node over the next three years, with TSMC leading the charge.
TSMC announced on its July earnings call that its 2nm (N2) node is on track for mass production in the second half of 2025, with a ramp-up comparable to the 3nm node, the Economic Daily News reported. With a higher price of roughly $30,000 per wafer, N2 is expected to give a significant lift to revenue and outperform 3nm in profitability, the report added.
Cloud Giants Could Eye TSMC’s 2nm Node
While KLA did not name specific customers, Higgins noted that HPC customers are expected to play a key role TSMC’s 2nm lineup. MyDrivers suggests that cloud giants—including Google, Broadcom, Amazon, and OpenAI—are pursuing custom AI chips that could tap into TSMC’s cutting-edge node.
Notably, AMD confirmed in April that its Venice processor, set for a 2026 launch, will be built on TSMC’s 2nm process. As MyDrivers notes, NVIDIA and AMD are planning products around TSMC’s 2nm process, including NVIDIA’s Rubin Ultra and AMD’s Instinct MI450 AI series.
This is in line with TechNews’ report, which notes that beyond the mobile device market’s aggressive adoption, 2nm is also beginning to penetrate other key high-performance computing sectors.
As TechNews reports, during the early ramp-up phase, six tech giants are expected to participate in 2nm production: Apple, AMD, Qualcomm, MediaTek, Broadcom, and Intel. Among them, Apple, TSMC’s long-standing top customer, is reportedly set to take nearly half of the initial 2nm output for its mobile processors, further cementing its position as TSMC’s largest client.
For now, most of TSMC’s cloud-based HPC customers are still sticking with its 3nm node. The Economic Daily News noted that Google’s sixth-generation TPU, Trillium—built on TSMC’s 3–4nm process—is its fastest processor yet. Meanwhile, Reuters reported in February that OpenAI is preparing to mass produce its first custom chip using TSMC’s 3nm technology in 2026.
However, as HPC demand accelerates in the AI era, 2nm is emerging as a key battleground. Marvell, according to its press release in March, has already introduced its first 2nm silicon IP for next-gen AI and cloud infrastructure. Meanwhile, Anue reports that Meta is pushing forward with two ASIC projects on the 2nm node, both slated for mass production in the second half of 2027.