OpenAI has formed a partnership with Taiwan Semiconductor Manufacturing Company (TSMC) to develop its first custom AI chips. The collaboration will employ TSMC's A16 process node, set for mass production by the second half of 2026, as per industry insiders.
Technological Advancements in Chip Manufacturing
The A16 process node, also known as a 16 Angstrom or 1.6-nanometer technology, signifies a notable progression from TSMC's existing 2nm-class N2P node. TSMC asserts that the A16 technology will yield an 8-10 percent enhancement in speed at identical voltage, a 15-20 percent decrease in power consumption at the same speed, and an increase in chip density by 7-10 percent.
Apple has also signaled interest in the A16 process and allocated production capacity for its output. In June, Apple disclosed a collaboration with OpenAI to incorporate ChatGPT into future versions of iOS, iPadOS, and macOS via a new service named Apple Intelligence. While Apple Intelligence will manage many machine learning operations on-device, more complex requests will be processed through a server-based solution known as Private Cloud Compute.
Financial Considerations and Market Position
Current reports indicate that OpenAI is in talks for a funding arrangement potentially raising the company's valuation to $100 billion. Despite achieving an annualized revenue of $3.4 billion as of June and maintaining over 200 million weekly active ChatGPT users, the company might face a $5 billion loss this year due to the high costs entailed in neural network training and staffing. Notably, OpenAI's technology is utilized by 92 percent of Fortune 500 companies, underscoring its prominent market position.
Initially, OpenAI explored the idea of constructing a dedicated wafer fabrication plant with TSMC. However, they shifted plans and instead opted to collaborate with American companies Broadcom and Marvell for ASIC chip development. These chips will leverage TSMC's 3nm technology and the to-be-released A16 process. Broadcom, known for assisting Google with its Tensor Processing Units (TPUs), will be a significant partner in this venture.
