Samsung Lands Google TPU and Neuralink Brain-Chip Orders, Turning TSMC’s Bottleneck Into Its Breakout

Google’s motivation is straightforward: TSMC’s leading‑edge capacity is effectively sold out to Nvidia, Apple, and other hyperscalers, leaving too little room for Google’s own AI chip ramp . Adding Samsung as a second source is a supply‑chain diversification play, not a technology defection . Notably, The Information first reported the talks in June 2026, citing two people familiar with the matter, and multiple Korean outlets have since characterized a deal as “likely” but not yet signed .

Key details at a glance:

• Chip: 10th‑generation TPU “Icefish” (co‑designed with MediaTek) .
• Samsung’s role: 2nm I/O die fabrication; TSMC retains the main compute die .
• Target production: Mass production as early as 2028 .
• Status: Active negotiations; no final contract publicly confirmed .

Neuralink “O1” Brain‑Computer Interface Chip: First Order in Hand

While the Google deal is still taking shape, Samsung Foundry has already locked in its first order from Neuralink. According to reports from The Korea Economic Daily and other Korean media, Samsung started process development and pilot production for Neuralink’s fourth‑generation brain‑implant chip, internally codenamed “O1,” using its mature 4nm SF4 process .

The engagement began in late 2025, with the first test chips taped out in May 2026 and delivery planned for the first half of 2027 . If testing goes smoothly, mass production could follow in the second half of 2027 . The project marks the first time Neuralink has outsourced a chip to Samsung, expanding a wafer‑fabrication supply chain that previously relied on TSMC for the third‑generation implant .

This win is part of a deepening Samsung‑Musk relationship. In 2025, Samsung signed a mammoth $16.5 billion contract with Tesla to manufacture its next‑generation AI6 inference chips at Samsung’s Taylor, Texas fab . The Neuralink deal extends that partnership into neurotechnology, giving Samsung a new and highly regulated niche to demonstrate its 4nm maturity .

Neuralink O1 deal summary:

• Chip: Fourth‑generation brain‑computer interface chip “O1” .
• Process: Samsung 4nm (SF4) .
• Timeline: R&D began late 2025; pilot production started May 2026; mass production possible in H2 2027 .
• Status: First‑ever Samsung–Neuralink wafer partnership; development underway .

The Bigger Picture: Samsung Foundry’s Turnaround Strategy

These wins are not random—they thread directly into a deliberate turnaround plan that Samsung Foundry has been executing since its utilization rate cratered to about 50% in 2025 . By Q1 2026, utilization had already recovered to roughly 80%, driven by the Tesla deal, an 8nm order from Intel, and ongoing 2nm discussions with AMD . The Google and Neuralink engagements would further fill capacity and improve the unit’s path toward profitability, which some analysts now expect as early as Q3 2026 .

Samsung is executing a dual‑track process strategy:

• 2nm GAA (SF2) is being pitched for high‑volume AI accelerator and HPC applications. The Google Icefish I/O die would be a marquee customer win that de‑risks the process node for other logic customers .
• Mature 4nm (SF4) is positioned for lower‑volume, specialized chips where proven yield and stability matter more than ultimate transistor density, which matches Neuralink’s needs exactly .

Samsung Foundry’s messaging is unmistakable: it wants to be seen as a credible second‑source fab for advanced logic, not just a memory giant. The more Big Tech firms that hedge their TSMC bets, the more Samsung stands to gain—provided it can execute on yields and on‑time delivery. The Google deal is still an open negotiation, and the Neuralink project is in early pilot production. Both are promising but not yet high‑volume realities. If Samsung turns these beachheads into volume ramps, the foundry landscape will look meaningfully different by 2028.