TAIPEI (Taiwan News) — OpenAI CEO Sam Altman visited Taiwan on Tuesday for a low-profile trip, Digitimes Asia reported, meeting with executives from TSMC and Foxconn to discuss artificial intelligence chip production and server infrastructure.

OpenAI, best known for models like ChatGPT, has been rapidly expanding its global computing capabilities. The company recently committed NT$9 trillion (US$300 billion) to Oracle over the next five years.

Altman’s discussions likely focused on aligning hardware capabilities with OpenAI’s growth plans and reviewing updates on ongoing projects. The talks gave the company a closer look at TSMC’s next-generation chip technology and strengthened collaboration on scaling AI systems.

TSMC’s 3nm process and chip-on-wafer-on-substrate packaging, technologies that efficiently combine multiple chips, are critical for AI infrastructure. They support the deployment of AI services across multiple data centers, which house specialized hardware needed to train, deploy, and deliver AI applications at high speed, according to IBM.

OpenAI has been linked to a US-backed AI initiative dubbed “Stargate.” Alongside SoftBank and Oracle, it plans five major AI data centers across the US.

Foxconn is said to be the largest AI server supplier for the initiative, with a visit by SoftBank CEO Masayoshi Son highlighting its role.

OpenAI is also developing its own AI chips to reduce reliance on Nvidia GPUs. The company has reportedly partnered with Broadcom to design custom AI chips for production at TSMC, targeting mass production next year, according to UDN Money.

The company also recruited a team of engineers who previously designed AI chips at Google’s parent company, Alphabet, to develop its own AI chips, which will also be produced at TSMC.

Supply chain analysts note that producing in-house AI chips is expensive and time-consuming. Designing chips and sending them to a foundry can cost tens of millions of US dollars.

Fabrication typically takes about six months. Any production errors require re-fabrication and debugging, which adds more time and expense.

OpenAI has begun deploying initial in-house chips for inference tasks, while full-scale production next year is expected to significantly boost computing performance.

TSMC’s leadership in advanced chip manufacturing and its neutral foundry model have made it a core beneficiary of the AI revolution. Nodes below 10nm are considered advanced, while those above 12nm are classified as mature.

The company pioneered 7nm, 5nm, and 3nm processes. Its 2nm chips, expected to enter mass production in the fourth quarter of 2025, are projected to be 10–15% faster than 3nm chips or reduce power consumption by 25–30% at the same speed, a key advantage for AI and high-performance computing applications.

Altman’s visit underscores Taiwan’s vital role in the global AI ecosystem. TSMC and Foxconn are expected to be major suppliers for the hundreds of thousands of chips needed, reinforcing the country’s position at the center of next-generation AI technology.