From Bloomberg via Yahoo Finance, October 29:
OpenAI is working with Broadcom Inc. to develop a new artificial intelligence chip specifically focused on running AI models after they’ve been trained, according to two people familiar with the matter.
The AI startup and chipmaker are also consulting with Taiwan Semiconductor Manufacturing Co., the world’s largest chip contract manufacturer, said the people, who asked not to be identified because the discussions are private. OpenAI has been planning a custom chip and working on such uses for the technology for around a year, the people said, but the discussions are still at an early stage.
OpenAI declined to comment. A representative for Broadcom didn’t respond to a request for comment, and a TSMC spokesperson said the company doesn’t comment on rumors and speculation. Reuters reported on OpenAI’s ongoing talks with Broadcom and TSMC on Tuesday. The Information reported in June that Broadcom had discussed making an AI chip for OpenAI.
The process for taking a chip from design to production is long and expensive. OpenAI is less focused on graphics processing units, chips that are used to train and build generative AI models — a market that has been cornered by Nvidia Corp. Instead, it’s looking for a specialized chip that will run the software and respond to user requests, a process called inference. Investors and analysts expect the need for chips to support inference will only grow as more tech companies use AI models to field more complex tasks....
....MUCH MORE
If OpenAI is consulting with TSMC I would guess that Sam Altman has not found anyone to step up with the $5 to $7 trillion he's been panhandling for:
And on inference:
Chips: The Pivot From Training To Inference (where Nvidia's dominance isn't as strong)Elon Musk earlier today tweeted that xAI's supercomputer was built in under 17 weeks, just incredible. However...there's always an "however"....this means that the chips have been purchased, the die has been cast, etc.
See PC Mag, September 3: "Musk's xAI Supercomputer Goes Online With 100,000 Nvidia GPUs".
Additionally, the big buyers (AMZN, META, MSFT) will be transitioning the usage of their computers from training to inference and it is here we join the electronic geniuses at IEEE Spectrum, August 28....
"Making GenAI more efficient with a new kind of chip" (plus some of our history with Nvidia)
....A lot of people see the opportunity in the inference, rather than the training, end of things but inference at the edge could lead to the kind of serendipitous manufacturing—research—discovery feedback loop that Nvidia experienced when they were pushing the limits of using GPUs as accelerators for supercomputers in 2015 -2016.....
Nvidia Q2 2025 Earnings Call Transcript, August 28, 2024 (NVDA)
"NVIDIA Corporation (NVDA) Q4 2024 Earnings Call Transcript"
A Deep Dive Into Zuckerberg's New AI Chip (META; TSM)"We estimate in the past year approximately 40% of data center revenue was for AI inference," Colette Kress, Nvidia's Chief Financial Officer (CFO) reported.
Nvidia Chief Executive Officer (CEO) Jensen Huang later said that the 40% estimate is "probably understated" due to the use of deep learning and generative AI (GenAI) interfaces for online recommender systems.....
With MTIA v2 Chip, Meta Can Do AI Inference, But Not Training...
And many, many more, it's all anyone is talking about.
