Analysis: Intel’s AI Chip Efforts Stall As AMD Gets A Boost Against Nvidia

Analysis: Intel’s AI Chip Efforts Stall As AMD Gets A Boost Against Nvidia


The fact that AMD in the past quarter managed to unleash new data center AI chips sales opportunities equal to the full-year revenue goal that Intel now expects to miss is yet another reflection of how far behind Intel is in creating a meaningful AI chip business.


Intel said it will fall short of the modest revenue goal it set to gain a foothold in the Nvidia-dominated data center AI chip market— a stark contrast from the upgraded sales forecast for such chips recently given by rival AMD.

Pat Gelsinger, Intel’s CEO, said during the chipmaker’s third-quarter earnings call last Thursday that it will “not achieve our target of $500 million in revenue” for its Gaudi accelerator chips this year, blaming slower-than-expected sales on the “product transition” from Gaudi 2 to the recently launched Gaudi 3 as well as “software ease of use.”

[Related: AMD Says Instinct MI325X Bests Nvidia H200, Vows Huge Uplift With MI350]

Despite the slowness, Gelsinger said he remains bullish on Intel’s long-term Gaudi strategy, which rests on the idea that enterprises will seek out Gaudi chips for AI systems that are more cost-effective than Nvidia GPU-based systems for running smaller AI models.

Among Gaudi 3’s early adopters are IBM Cloud, which plans to launch Gaudi-based offerings in early 2025, and a reshaped Inflection AI, which is using Gaudi 3 compute from the recently rebranded Intel Tiber AI Cloud for a solution that may count Intel as a future customer.

“There is clear need for solutions with superior [total cost of ownership] based on open standards, and we are continuing to enhance the Gaudi value proposition,” Gelsinger said.

This is not the kind of news Intel was hoping to give, especially given that its $500 million revenue goal for 2024 amounted to merely 2 percent of the data center compute revenue Nvidia made in the last quarter alone. Nvidia’s data center compute revenue, which was $22.6 billion for the May-July period, largely came from GPUs like the H100 and H200.

Meanwhile, AMD CEO Lisa Su said earlier last week that her company has upgraded its 2024 sales forecast for Instinct data center GPUs by $500 million to $5.5 billion.

In AMD’s third-quarter earnings call that day, Su said the upgraded forecast was based on the completion of “some important customer milestones” such as meeting reliability requirements in data centers and optimizing the chips on certain AI workloads.

Two major customers that helped fuel Instinct sales in the third quarter were Microsoft and Meta, which both expanded their use of AMD’s MI300X GPUs for internal workloads, including multiple Copilot services and other inferencing infrastructure.

The company also saw the availability of MI300X public cloud instances expand among Microsoft, Oracle Cloud and smaller cloud providers while “multiple startups and industry leaders” such as Databricks and Essential AI drove “strong” adoption of such instances.

However, AMD’s $5.5 billion Instinct GPU forecast for 2024 is still a small fraction of what Nvidia makes from AI chips every quarter.

But the fact that AMD in the past quarter managed to unleash new sales opportunities equal to the full-year revenue goal that Intel now expects to miss is yet another reflection of how far behind Intel is in creating a meaningful AI chip business.

Intel may have fallen short of its small sales forecast, but top sales executives at the company recently indicated in interviews with CRN that it is still in the early stages of scaling up sales efforts for Gaudi 3.

The company announced in September that the first Gaudi 3 systems will arrive in October from Dell Technologies and Supermicro. Hewlett Packard Enterprise is then expected to follow up with its own offering in December. System availability from other server vendors, including Lenovo, has not been disclosed.

Michael Green, the head of Intel’s new North America partner scale group and U.S. channel chief, admitted to CRN in an interview last month that the Gaudi 3 rollout is a “slow process” in part because the company is scaling up a product line that was previously supported by only one OEM, Supermicro, with the predecessor chip.

“It’s a brand-new product that we’re ramping. We have visions of making this a channel-available product in 2025,” he said, adding that Intel’s channel partners will play a “massive role” in selling Gaudi 3 systems.

An AMD executive, on the other hand, told CRN in June that the company was expected to make a greater emphasis on enabling channel partners to sell servers equipped with its Instinct chips in the second half of the year.

“You’ll see us shifting very hard to enable enterprise through multiple channels this year, including the channel,” said Forrest Norrod, AMD’s top data center executive, at the time.

A senior leader at a solution provider that is in the top 100 of CRN’s 2024 Solution Provider 500 list told CRN on Monday that there is a need for stronger competition against Nvidia because of long lead times for its GPUs combined with how the AI chip giant determines allocations.

“There’s an appetite out here, for sure, because we want some ability to get this in our customers’ hands without having to wait nine months and go through that internal vetting system of essentially, ‘Is your customer deemed worthy enough to get these GPUs from Nvidia?’” said the executive, who asked to not be named to speak candidly.

The solution provider leader said Nvidia’s customer vetting process, which determines when chips are allocated to customers, is “not enjoyable.” (While an Nvidia spokesperson declined to comment on this complaint, Nvidia CEO Jensen Huang said in February that the AI chip giant does its “best to allocate fairly and to avoid allocating unnecessarily.”)

This situation portrayed by the solution provider executive means his company and his customers would be in a better place if there were one or two more AI chip suppliers.

“There’s an inherent need to get it out there,” the solution provider executive said.

But while the solution provider leader is glad both Intel and AMD have plans to get their AI chips in the hands of channel partners, the executive said the latter has a head start on the technology side and a more active presence on the selling side.

“Their sellers are out with our sellers. They are trying to proactively win that affinity battle. And Intel’s just been a little too lax in that that respect,” said the executive, who added that Intel’s massive job cuts have been a continued source of concern.

In response to the solution provider executive’s comments, an Intel spokesperson echoed what Gelsinger said about Gaudi 3’s prospects in last week’s earnings call.

“Gaudi 3 benchmarks have been impressive, and we remain encouraged by the market opportunity available to us. We will continue to enhance the Gaudi value proposition given the clear need for solutions with superior TCO based on open standards,” the representative said.



Source link
lol

The fact that AMD in the past quarter managed to unleash new data center AI chips sales opportunities equal to the full-year revenue goal that Intel now expects to miss is yet another reflection of how far behind Intel is in creating a meaningful AI chip business. Intel said it will fall short of the…

Leave a Reply

Your email address will not be published. Required fields are marked *