The artificial intelligence (AI) industry is booming. Thanks to widespread public acceptance of tools like OpenAI’s ChatGPT and Google’s Bard, the technology is in demand like never before. Without question, the mass majority of chips powering those AI applications come from Nvidia.
The GPU maker controls upwards of 95% of the AI chip market and has seen its stock price rise by more than 100% since the start of 2023. While it’s clear that Nvidia is the uncontested frontrunner in AI chip production, the industry is heating up.
After several setbacks and failures of its Ponte Vecchio chip, Intel is preparing to rejoin the fray. The company recently unveiled the first details of its new “Falcon Shores” chip at a supercomputing conference in Germany.
Slated for release in 2025, the all-new Falcon Shores design will be Intel’s flagship foray into the AI chip sector. The company’s recent announcement offered a small, but noteworthy glimpse of what the new chips will offer.
Falcon Shores will feature 288 gigabytes of HBM3 memory, making it attractive to businesses looking for chips to run powerful AI models. Meanwhile, the new chip will also support 8-bit floating point computation. This is another in-demand feature for AI applications.
Intel’s Falcon Shores chip is designed to compete with offerings from both Nvidia and AMD. The latter currently plans to challenge Nvidia’s stranglehold on the AI market with a powerful chip called the MI300. The 146-billion transistor chip is the largest AMD has produced to date and includes 128 gigabytes of memory. AMD claims the MI300 challenges Nvidia’s most recent offering, the H100 which is packed with 80 billion transistors and 188 gigabytes of memory.
Given that Falcon Shores won’t be released until 2025, it’s likely Nvidia will already have another flagship model to market. Time will tell whether Intel’s upcoming offering will be able to outpace the market leader’s next effort—not just its current.
At this time, we don’t have much information about Falcon Shores, nor Intel’s greater plan for competing in the AI world. The chipmaker currently has essentially no market share in this segment thanks to the failures of its previous design strategy with Ponte Vecchio.
Years of delays have let Nvidia and AMD gobble up market share while Intel fumbled and changed course. Though, it seems like good news is finally on the horizon for the beleaguered chip.
The company noted that it has nearly completed shipments for the Ponte Vecchio-based Aurora supercomputer at the Argonne National Lab. It claims its current AI chip outperforms Nvidia’s H100, which is a bold statement. Whether or not this performance holds up outside of internal benchmarking remains to be seen.
Meanwhile, Intel is taking a careful approach to its AI aspirations. The company has rolled back its previous strategy—combining GPUs with CPUs—admitting the approach was incorrect.
Jeff McVeigh, corporate vice president of the company’s super computer group, said in a statement to Reuters, “While we aspire to have the best CPU and the best GPU in the market, it was hard to say that one vendor at a time was going to have the best combination of those.”
“If you have discrete offerings, that allows you at the platform level to choose both between the ratio as well as the vendors,” he added.
Given the current and future relevance of AI technology, it’s better late than never for Intel to join the party. But time will tell whether the chip giant is simply too far behind to catch up to Nvidia and AMD.
One thing is certain. In the coming years, there will be no shortage of demand for AI chips. So, if Intel can indeed perfect its Falcon Shores chip, there will be buyers available. In addition, more competition in the AI chip space is always good for innovation.
The next move belongs to the current market leader, Nvidia. While we still don’t know much about Falcon Shores, there is no doubt Nvidia is already looking for ways to outdo Intel’s forthcoming offering.