Intel’s AI Strategy Will Favor a “Broadcom-Like” ASIC Model Over the Training Hype, Offering Customers Foundry & Packaging Services

Intel's AI plans have been under uncertainity for several quarters now, but it appears that the firm is expected to target two main segments: ASICs and edge AI.

Intel's ASIC Ambitions Intend to Scale Up The Firm Alongside Broadcom & Marvell By Offering Foundry Services As Well

There's no doubt that when it comes to computing solutions providers, Intel has dominated in the past, mainly in segments like general-purpose and server-grade computing. However, when it comes to artificial intelligence, Team Blue has lagged behind both NVIDIA and AMD by a huge margin. Intel's former CEO, Pat Gelsinger, did admit that Intel's AI strategy was not up to mark. Even with CEO Lip-Bu Tan, we hadn't seen a defined structure until probably now. Speaking at the Barclays Annual Global Technology Conference, Intel's VP, John Pitzer, provided insight into what to expect from the firm on the AI front.

We're really looking to be focused on a power-optimized GPU for inference as AI moves from the center out to the edge. We've got a lot of work to do there. So stay tuned. It's going to take some time. I think the other dynamic that Lip-Bu has introduced that's worth talking about is the ASIC portion of this business.

I think the other interesting dynamic that we have in this ASIC business is we can pursue it in a Broadcom Marvell type model. But given also Intel Foundry, there are a lot of hyperscalers that are looking to go directly to the foundry and to try to do that, to circumvent kind of the traditional model.

Now, one of the main focuses of Intel is to capitalize on edge AI, and one of the ways the firm is already doing this is through its 'AI PC' portfolio, which includes the laptop-focused Meteor Lake, Lunar Lake, and the upcoming Panther Lake lineup. We have seen Intel driving up edge AI performance by integrating newer NPU generations, which have led to a significant increase in computing capabilities on mobile SoCs. More importantly, Intel also plans to expand its edge offerings through products like Crescent Island, an inference-focused option that utilizes LPDDR5X memory onboard.

The more attractive strategy is the dedicated ASIC unit, which Srini Iyengar is currently leading under the Central Engineering Group (CEG). We know that companies are looking to tape out custom silicon to tailor it for their personal workloads, and so far, solutions like Google's TPUs and Amazon's Trainium have seen massive interest in the industry. ASIC is an emerging segment in the AI race, which is why Intel is expected to center its AI strategy around it, providing customers a 'one-stop' shop solution for manufacturing and advanced packaging.

Intel's VP John Pitzer mentions that the firm has a 'vibrant' business running in the custom networking ASICs segment, as they have seen customers for SmartNIC ASICs, which are tailored for network-bound workloads like network packet processing, telemetry, traffic management, and much more. Eventually, Intel expects to position itself alongside Broadcom and Marvell, but the firm plans to distinguish itself by offering internal foundry services to customers, ensuring improved time-to-market figures, as well as close collaboration with ASIC customers.

Interestingly, Lip-Bu Tan has a deep history in the push towards a custom silicon business model, as evidenced by his position at Cadence, where he has emphasized IP business, design tools, design ecosystem partnerships, and vertical markets for custom silicon. His experience and market connections accelerate Intel's ability to capitalize on the 'ASIC hype'.

Follow Wccftech on Google to get more of our news coverage in your feeds.