Intel’s acquisition of Habana Labs for $2 billion in 2019 was supposed to be a game-changer in the AI race. Amazon’s announcement in December 2020 that it would use Habana’s Gaudi chips to train large language models in the cloud seemed to validate Intel’s bet. However, just a few years later, the story has taken a dramatic turn.
Last week, Intel revealed that its next-generation Habana processor, Falcon Shores, had received negative customer feedback and would not be commercially released. This followed an earlier announcement that the Gaudi line had failed to meet its projected $500 million revenue target for 2024, leading Intel to halt further development beyond Gaudi 3. The failure of Habana Labs under Intel’s ownership is particularly surprising given the track record of its founder, Avigdor Willenz.
Willenz has a history of successful exits, including the $2.7 billion sale of Galileo to Marvell and the $350 million sale of Annapurna Labs to Amazon. Habana’s decline began even before the public announcements, with nearly all of its original founders, managers, and engineers leaving Intel by 2024, most immediately after their four-year retention period ended. Intel’s struggles with acquisitions are well-documented.
Intel’s AI struggle with Habana
Initially, the company tried to keep Habana as a separate unit, similar to its approach with Mobileye, but eventually dissolved Habana and absorbed its operations into existing Intel teams. Intel’s AI efforts were fragmented and ultimately unsuccessful, with the company acquiring multiple startups like Nervana for $400 million, only to realize it had backed the wrong technology.
The acquisition of Habana was partly a response to losing out on Mellanox, which Nvidia acquired for $7 billion. However, the integration faltered, and Habana became another name on Intel’s list of acquisition failures. Former Habana employees cited Intel’s bureaucracy as a major obstacle, with decisions that could have been made quickly at Habana taking days and multiple meetings at Intel.
Intel’s AI strategy became even more convoluted with the departure of its AI leader, Raja Koduri, and the decision to merge Habana with its GPU division to develop Falcon Shores, a hybrid GPU-CPU chip. However, this approach failed, and Falcon Shores will now only be used for internal testing. Intel is pinning its hopes on a new processor, Jaguar Shores, in an attempt to catch up to Nvidia.
Despite the financial windfall for Willenz and Habana’s employees, the startup’s implosion represents a significant missed opportunity for Intel and the Israeli tech landscape. As the former Habana team moves on to new ventures, Intel serves as a cautionary tale of how not to handle cutting-edge technology acquisitions in the rapidly evolving world of artificial intelligence.
