The turning point of Intel's downfall was when it decided to let Israeli startup Habana Labs, which it had acquired in 2019, die.
By
Intel has long held an overwhelming share of the semiconductor market, but due to delays in AI chips, at the time of writing, its market capitalization has been significantly surpassed by NVIDIA. Israeli technology media CTech pointed out that Intel's loss in the AI race was largely due to its lack of control over Habana Labs, a new semiconductor company it acquired in 2019.
How Intel ruined an Israeli startup it bought for $2B—and lost the AI race | Ctech
https://www.calcalistech.com/ctechnews/article/s1tra0sfye
When Amazon made the much-talked-about announcement in 2020 that it would use Habana Labs' chip, Gaudi, to build large-scale language models (LLMs), Intel, which had acquired Habana Labs the previous year for $2 billion, welcomed the move, calling it 'the first challenge to Nvidia's dominance in GPUs.'
However, in its financial results released in January 2025, Intel announced that the commercial deployment of Habana Labs' next-generation AI chip, Falcon Shores, had been postponed due to negative feedback, along with its disappointing performance. In 2024, Intel also announced that it would not develop a successor chip to Gaudi 3 due to the poor performance of Habana Labs' Gaudi series.
Intel's fourth-quarter 2024 sales fall 7% year-on-year but exceed expectations, postponement of launch of next-generation AI chip 'Falcon Shores' - GIGAZINE
'This effectively means that the failure of Habana Labs marks another chapter in Intel's long history of failed acquisitions,' CTech said.
According to CTech, this failure is typical for Intel, but unusual for Avigdor Willens, founder of Habana Labs. Willens, known as a 'genius of the semiconductor world,' is known as a serial entrepreneur who has launched and successfully exited numerous semiconductor companies, including selling Galileo Technologies to Marvell Technology for $2.7 billion in 2000 and Annapurna Labs to Amazon for $350 million in 2015.
Meanwhile, Habana Labs, which was sold to Intel, had begun to collapse even before the impasses at Gaudi and Falcon Shores came to light, and by 2024 most of the key executives and engineers from Habana Labs had left Intel.
Unlike those that missed the mobile revolution, Intel has correctly recognized the future of AI and acquired emerging AI company Nervana for $400 million (about 60 billion yen) in 2016 as part of its AI promotion. However, Intel, which was trying to strengthen its AI processors by acquiring Nervana, soon realized that it had made a wrong bet. In 2017, with NVIDIA's rise in AI GPUs, Intel appointed former AMD executive Raja Koduri to take charge of its AI business, but Intel's products were underperforming and did not meet market expectations.
In an attempt to catch up, Intel attempted to acquire Israeli semiconductor company Mellanox Technologies, but failed when NVIDIA offered a higher bid than Intel.
by
After losing the Mellanox bidding war, Intel turned its attention to Habana Labs. Habana Labs, which Willens founded with former Apple veterans David Dahan and Ran Halutz, already had a functional AI training processor, Gaudi, at the time, and was said to have attracted Amazon's interest, but Intel, which wanted to recover from the failure of Mellanox, was willing to pay a higher acquisition price. By 2020, Intel had completed the acquisition of Habana Labs and closed Nervana.
Thus, Habana Labs and Mellanox Technologies, both of which were acquired by major semiconductor companies at roughly the same time, had very different fates. NVIDIA successfully integrated Mellanox Technologies, which earned the company $13 billion in annual revenue and brought thousands of jobs to Israel.
Meanwhile, Habana Labs struggled under Intel's leadership. According to CTech, some point out that Intel acquired Habana Labs to cover up Nervana's failure and to appeal to investors about its investment in AI, and that it never intended to make it a rival to NVIDIA. This is evident from the fact that Habana Labs was placed under the Data Platform Group (DPG) rather than the GPU division led by Koduri, which was highly regarded within Intel and won a large budget, CTech pointed out.
'When the Habana Labs acquisition was completed, people inside Intel didn't see why they wanted to run both Habana Labs and their GPU division, both of which were developing competing architectures,' a former Intel executive told CTech.
A former Habana Labs employee also said, 'At Habana Labs, we could make decisions in a five-minute hallway, whereas at Intel it took three meetings with dozens of people to make the same decision, and nothing ever got done.' He recalled that Intel's bureaucratic inefficiency was a major obstacle.
Despite these criticisms, Intel continued to sell Gaudi processors and develop a competing GPU, Ponte Vecchio, at the same time. However, as NVIDIA's dominance became impossible to ignore with the rise of generative AI models such as ChatGPT in 2022, Intel again faced customer dissatisfaction. Ponte Vecchio was then discontinued in 2024, and no new generation of Gaudi was developed.
After Koduri left, Intel tried to merge Habana Labs with its GPU division to develop a new AI processor, Falcon Shores, but as mentioned above, Falcon Shores was never released to the market and was only used for internal testing.
Intel is now focusing on its next-generation Jaguar Shores chips, trying to catch up with NVIDIA in one leap, but industry experts say it's impossible to close the gap with NVIDIA, which has been working on AI chips for 20 years.
'While Willens' sale of Habana Labs to Intel was an unexpected success, based on the acquisition price alone, the subsequent failure of Habana Labs was a rare blemish on Willens' resume and, more importantly, a missed opportunity for Israel. Had it gone well, Habana Labs might have become a cornerstone of Intel's AI strategy and not become a casualty of Intel's troubled acquisition history,' CTech said.
Meanwhile, Intel told CTech, 'Based on customer feedback and market trends, we plan to leverage Falcon Shores as our internal test chips and focus our resources on bringing Jaguar Shores to market instead. As the foundation for Jaguar Shores, Falcon Shores remains critical to our key efforts and its learnings will feed directly into Jaguar Shores.'
Related Posts:
in Hardware, Posted by log1l_ks