Has Intel beaten NVIDIA with a 50% faster AI chip?

Intel has announced the launch of a new AI chip, called Gaudi 3, which is expected to rival NVIDIA’s most popular AI chip, the H100.

When Intel tested Gaudi 3 on large language models (LLMs)--such as Meta’s open-sourced Llama and Abu Dhabi-backed Falcon–it was reportedly 50% faster than NVIDIA’s H100, 1.3 times faster than H200 (its bigger chip), and 40% more power-efficient.

NVIDIA currently owns 80% of the AI chip market, but with this faster, more cost-efficient option entering the market, its spot at the top could be in jeopardy, as Intel claims its open-source approach will allow customers to use the services and software that they want.

They specifically state that the hyper-scalers–like Microsoft, OpenAI, Google, Amazon, and Meta–are developing their own AI chips, using NVIDIA chips. The Gaudi 3 will give these hyper-scalers another option, so they don’t have to rely on one, single source to meet their high-powered AI needs.  

Although pricing for the Gaudi 3 chips is unknown, Intel is already shipping samples to big tech companies, and it will be available for consumer purchase in the third quarter of this year, with multiple companies, including Dell, Hewlett Packard Enterprise, and Supermicro already confirming that they will use this chip to power their AI systems.

Back in December, Intel also released an AI-powered PC (Core™ Ultra) that has built-in chips that can handle AI tasks. These advancements in AI hardware and chips are part of Intel’s strategy to reinvent itself as the leader in advanced AI hardware and chip technology. But their investment in AI is costing them, with announcements that they made a $7B loss last year, and with their stock value falling by 25% so far this year, this new chip comes at a crucial time for Intel.

‍