// Global Analysis Archive
Apple and Broadcom are reportedly developing an AI server chip codenamed “Baltra,” expected to be manufactured by TSMC on the N3E 3nm process and potentially use Samsung Electro-Mechanics’ glass substrate. The chip is anticipated to debut in Apple’s security-focused cloud infrastructure to reduce reliance on costly NVIDIA GPUs and lower data center operating costs.
According to the source, TSMC will upgrade its second Kumamoto facility in Japan to 3nm, with 15,000 12-inch wafers per month and mass production expected in 2028. The move underscores a shift toward security-driven distribution of advanced semiconductor capacity among trusted partners, supported by Japanese subsidies and industrial policy.
TechNode, citing DigiTimes, reports that TSMC’s 3nm capacity has entered an unusually severe overload state, creating a major bottleneck for GPU/CPU designers and hyperscale cloud providers. The resulting supply-demand imbalance is reportedly disrupting product roadmaps and shifting industry focus from technology advancement to capacity allocation and procurement.
Nvidia CEO Jensen Huang used a high-profile Taiwan visit to press key suppliers, especially TSMC, to expand capacity amid strong 2026 AI demand. He also warned that memory chip shortages could become a critical bottleneck for AI server deliveries and broader supply-chain throughput.
Apple and Broadcom are reportedly developing an AI server chip codenamed “Baltra,” expected to be manufactured by TSMC on the N3E 3nm process and potentially use Samsung Electro-Mechanics’ glass substrate. The chip is anticipated to debut in Apple’s security-focused cloud infrastructure to reduce reliance on costly NVIDIA GPUs and lower data center operating costs.
According to the source, TSMC will upgrade its second Kumamoto facility in Japan to 3nm, with 15,000 12-inch wafers per month and mass production expected in 2028. The move underscores a shift toward security-driven distribution of advanced semiconductor capacity among trusted partners, supported by Japanese subsidies and industrial policy.
TechNode, citing DigiTimes, reports that TSMC’s 3nm capacity has entered an unusually severe overload state, creating a major bottleneck for GPU/CPU designers and hyperscale cloud providers. The resulting supply-demand imbalance is reportedly disrupting product roadmaps and shifting industry focus from technology advancement to capacity allocation and procurement.
Nvidia CEO Jensen Huang used a high-profile Taiwan visit to press key suppliers, especially TSMC, to expand capacity amid strong 2026 AI demand. He also warned that memory chip shortages could become a critical bottleneck for AI server deliveries and broader supply-chain throughput.
| ID | Title | Category | Date | Views | |
|---|---|---|---|---|---|
| RPT-3638 | Apple’s ‘Baltra’ AI Server Chip: TSMC N3E Manufacturing and Glass Substrate Signals a Push to Reduce GPU Dependence | Apple | 2026-04-09 | 0 | ACCESS » |
| RPT-3434 | TSMC’s Kumamoto 3nm Upgrade Signals a Security-Led Rewiring of Indo-Pacific Chip Supply Chains | Semiconductors | 2026-04-04 | 0 | ACCESS » |
| RPT-3287 | TSMC 3nm ‘Overload’ Intensifies Global Battle for Leading-Edge Chip Capacity | Semiconductors | 2026-03-30 | 0 | ACCESS » |
| RPT-472 | Huang’s Taiwan Push Signals 2026 AI Boom—and a Memory Supply Squeeze | Nvidia | 2026-02-01 | 0 | ACCESS » |