Inflection AI shifts to Intel Gaudi 3, challenging Nvidia’s AI chip lead

Inflection AI is shifting its latest enterprise platform to Intel’s Gaudi 3 accelerators, stepping away from Nvidia GPUs.

In a statement, Intel said that Inflection AI is introducing “Inflection for Enterprise,” a first-of-its-kind, enterprise-grade AI system powered by Intel’s Gaudi accelerators and Tiber AI Cloud.

The system is designed to offer empathetic, conversational AI capabilities tailored for employees, while also providing the control, customization, and scalability needed for large-scale, complex deployments, Intel said in a statement.

Founded in 2022, Inflection AI initially focused on developing Pi, a conversational personal assistant. However, following the departure of co-founders Mustafa Suleyman and Karén Simonyan to Microsoft earlier this year, the company has pivoted to building custom fine-tuned AI models tailored to enterprise needs.

The latest platform, Inflection 3.0, is designed to enable enterprises to fine-tune AI models using proprietary datasets to create customized, enterprise-specific applications.

The announcement follows IBM’s recent move to partner with Intel in deploying Gaudi 3 AI accelerators as a service on IBM Cloud, signaling a growing interest in Intel’s AI hardware across the industry.

Intel’s attempts at AI chips

The notable aspect of Inflection AI’s move is its departure from Nvidia, which continues to dominate the AI chip market. Previously, Inflection AI’s Pi consumer application operated on Nvidia GPUs.

“Nvidia has captured much of the market with hyperscalers and key enterprises,” said Neil Shah, partner & co-founder at Counterpoint Research. “However, AI is expanding across various industries, creating ample opportunity. This shift allows Intel to be competitive on price and potentially regain customers by offering a stronger price-performance balance compared to Nvidia.”

Significantly, Inflection AI’s decision showcases Intel Gaudi 3 as a viable alternative to Nvidia, which is largely sold out for the next several years, according to Hyoun Park, CEO and chief analyst at Amalgam Insights.

“Because large companies have taken up the majority of current AI chip inventory, enterprises must start looking at other options,” Park said. “With both IBM and Inflection AI making a bet on Intel, enterprises now have a potential option to look at a cheaper and arguably higher performance chip to build custom models.” 

While Nvidia’s chips may offer superior speed, Intel is relying on Gaudi 3’s lower price and reduced total cost of ownership (TCO) to bolster its competitive edge.

In a blog post, Inflection AI said that running Inflection 3.0 on Intel’s Gaudi 3 yields “up to 2x improved price performance, along with 128GB of high-bandwidth memory at 3.7 TB/sec, delivering optimal generative AI performance compared to current competitive offerings.”

Benefits of choosing Intel

In the statement from Intel, Inflection AI COO Ted Shelton highlighted that CEOs and CTOs are increasingly frustrated with the lack of availability of truly enterprise-grade AI tools.

“Enterprise organizations need more than generic, off-the-shelf AI, but most lack the expertise to fine-tune models on their own,” Shelton said. “We’re proud to offer a solution that addresses these challenges, and with the performance improvements we’re seeing on Intel Gaudi, we’re confident it can scale to meet any enterprise’s demands.”

Shah noted that Intel’s Gaudi 3 remains a strong AI accelerator compared to Nvidia’s offerings and is well-positioned for strong sales.

“Furthermore, Intel’s accelerator roadmap is solid, and its partnership with TSMC provides instant access to the most advanced process nodes, giving Intel an advantage to compete directly with Nvidia and AMD while it simultaneously works on its own foundry readiness for future accelerators,” Shah added.

Implications for Intel

The news arrives as Intel grapples with financial losses, a decline in market share, and recent layoffs. Reports have even hinted at acquisition discussions between Intel and its rival Qualcomm.

While Inflection AI’s adoption of Gaudi 3 is a positive step for Intel, concerns remain over the accelerator’s future, as the Gaudi series is expected to be phased out in favor of the upcoming Falcon Shores GPU starting next year.

“Intel is obviously dealing with a lot of challenges right now, but this transition [to Falcon Shores] is one of the most important ones,” Park said. “Companies seeking to build AI models would need to know that there is a clear roadmap from Gaudi 3 to Falcon Shores and that current investments are not an evolutionary dead end.” 

However, Falcon Shores may not enter full production until 2026 and Intel will need a chip to market for the upcoming year.

“Going forward, Intel must navigate and build its brand imagery in the AI hardware market, against the likes of Nvidia as well as emerging rivals,” said Prabhu Ram, VP of the industry research group at Cybermedia Research. “As competitors enhance their offerings, Intel will need to continuously validate its performance claims against those of Nvidia, Google, and AMD.”