OpenAI, Broadcom and TSMC looking at making an AI inference IC

OpenAI, Broadcom and TSMC are working on an AI inference chip, reports Bloomberg.

The move is seen as a departure from OpenAI’s reported strategy of trying to raise the money to build a network of fabs to make AI ICs.

Intel has also said that its concentrating on inference ICs. This could be because Nvidia’s grip on the market for ICs for training AI programmes  seems solid.

OpenAI is also said to be looking to make investments in datacentres in the belief that the strategic  future in the AI business is to have control over the whole AI infrastructure from software to chip to server.

Although OpenAI recently raised $6.6 billion, it’s burning cash at the rate of $5 billion a year and its revenues this year are expected to be $3.7 billion  while hardware infrastructure investment is  costly.