OpenAI’s first custom-designed silicon chips allegedly might be manufactured by Taiwan Semiconductor Manufacturing Firm (TSMC), the identical outfit churning out processors for Nvidia, Apple, AMD, Intel, and others.
United Each day Information Group, certainly one of Taiwan’s largest media orgs, this week mentioned trade sources declare OpenAI has booked capability for TSMC’s A16 course of node, which is concentrating on mass manufacturing within the second half of 2026.
A16 is a 16 Angstrom or 1.6-nanometer manufacturing course of. “In contrast with N2P,” TSMC’s 2nm-class course of node, “A16 presents 8-10 p.c velocity enchancment on the similar Vdd [working voltage], 15-20 p.c energy discount on the similar velocity, and 1.07-1.10X chip density,” based on the manufacturing unit large.
The chip manufacturing goliath didn’t reply to a request for remark.
Apple is claimed to have been the primary main buyer to have reserved A16 manufacturing capability.
The iPhone maker in June introduced a partnership with OpenAI to combine ChatGPT into iOS 18, iPadOS 18, and macOS Sequoia through a service referred to as Apple Intelligence. Whereas Apple says that most of the machine studying fashions utilized by Apple Intelligence run on-device, the iBiz additionally plans to deploy a service referred to as Non-public Cloud Compute, operating on Apple Silicon, to make use of server-based AI fashions to deal with advanced requests.
There isn’t any indication presently that OpenAI anticipates Apple, which has its personal Neural Engine {hardware} for accelerating AI workloads, will find yourself utilizing OpenAI silicon when it turns into out there.
OpenAI didn’t reply to a request for remark.
OpenAI has reportedly explored investing in its personal chip fabs to the tune of $7 trillion, however now these ambitions seem to have been scaled again. As an alternative of negotiating with TSMC to construct a devoted wafer manufacturing unit, which is ridiculous when you consider it, the AI mannequin maker is believed to be pursuing the fabrication of its personal machine-learning accelerating ASIC with the assistance of Broadcom and Marvell on a TSMC node.
That appears extra real looking and regular: A software program firm partnering with a chip design firm to have a {custom}, app-specific processor fabbed by a contract producer. Broadcom helped Google design the online large’s TPUs in any case.
UDN expects TSMC to fab Broadcom- and Marvell-designed ASICs on its 3nm node and its subsequent A16 course of. Ergo it is not likely a shock that TSMC would fab a 1.6nm chip for OpenAI with Broadcom or related aiding within the design advert testing.
OpenAI is mentioned to be engaged on a funding deal that will see the corporate valued at $100 billion. The AI super-lab, which as of June had a reported annualized income of $3.4 billion, now says it has greater than 200 million weekly energetic customers of ChatGPT, double the quantity cited final November.
And it claims that 92 p.c of Fortune 500 corporations are utilizing OpenAI’s merchandise to a point. Additionally, API utilization is claimed to have doubled for the reason that launch of GPT-4o mini in July. However, OpenAI took greater than $10 billion in pledged assist from Microsoft to get thus far, amongst different investments, and might dive $5 billion into the pink this 12 months because of its non-trivial neural internet coaching and employees prices. ®