It’s hard to sell yourself as the last Mohican to Gaudi 3

The hardware itself wouldn’t be a problem, but the future of the product line would be.

Back in April, Intel unveiled the Gaudi 3 accelerator designed for machine learning, which the company talked about in more detail in September, and which was expected to generate $500 million in revenue this year. The latter will certainly not happen anymore, as Pat Gelsinger recently admitted that they only expect less income in terms of AI accelerators, but what is more interesting is that the problem is not really with the hardware itself.

We asked potential companies whether they considered the capabilities of Gaudi 3 to be good enough, and most of them spoke positively about the hardware itself, so in theory they would find it suitable for, for example, the inference stage of machine learning. So it cannot be said that Gaudi 3 was a bad success, it would have a market at a favorable price, but what keeps companies away from it is the future of the product range.

As you know, Intel announced in advance that Gaudi 3 will be the last of the Gaudi series. Practically, this marks the end of the history of the entire product line, and its place is taken by another design, which is produced under the name Falcon Shores. The latter already brings significant changes architecturally, so the accompanying software environment also changes. For this reason, companies do not consider it a very logical step to buy a hardware that will not have a direct successor that is fully compatible with the written software. No matter how good the Gaudi 3 may be in this respect, it will be very difficult to sell itself as the last representative of the product line.