Through a non-binding agreement, Intel is about to acquire SambaNova Systems. The latter is an AI chip company where current Intel CEO Lip-Bu Tan has been chairman since May 2024. However, the estimated value of the deal is lower than SambaNova’s latest valuation, falling below $5 billion.
That valuation dates back to 2021, when SambaNova last conducted an investment round. Intel Capital is already an investor in the company, reports Wired, which was the first to highlight the current progress of the negotiations.
Progress
Two sources with direct knowledge of the matter confirm to Wired that Intel and SambaNova have signed a letter of intent. This means that the deal is not yet finalised. It could take weeks or even months before all legal procedures, liability issues and financial checks have been completed.
Bloomberg reported in late October that Intel was in talks with SambaNova about a possible acquisition. That news organization was early to the story, as the two parties had only just begun discussions at that point. The report suggested at the time that SambaNova could be acquired for less than its April 2021 valuation of $5 billion.
Intertwined with Intel
It is noteworthy that Intel CEO Lip-Bu Tan currently serves as executive chairman at SambaNova Systems. The intertwined nature of the two companies is also evident in the fact that SoftBank, an Intel investor, also has money invested in SambaNova.
SambaNova Systems was founded in 2017 in Palo Alto, California, by Kunle Olukotun, Rodrigo Liang, and Christopher Ré. Olukotun and Ré are professors at Stanford; Liang previously worked as a senior manager at Oracle. The company makes an AI chip platform for inference computing, the process by which large language models make predictions based on enormous amounts of data.
By early 2025, the startup had raised $1.14 billion in funding, according to PitchBook data cited by Wired. In 2020, it raised $250 million from asset manager BlackRock, Intel Capital, and venture capital fund GV, among others, bringing its valuation to $2.5 billion. The following year, SambaNova was valued at $5 billion after a massive $676 million funding round led by SoftBank’s Vision Fund 2.
The implied valuation has since declined. BlackRock is said to have reduced the value of its SambaNova shares by 17 percent over the past year, according to The Information. That likely made the company an attractive acquisition target for Intel, which is also lagging behind the rest of the chip industry when it comes to AI chips.
AI chip strategy under pressure
After taking over as CEO earlier this year, Tan said he wants to restructure Intel’s debt, divest non-core businesses, and switch to an AI-first strategy. The beleaguered chip company also received an $8.9 billion capital injection from the US government in August, which it plans to use to expand domestic semiconductor manufacturing.
For Intel, the acquisition may be a way to regain ground in the AI chip market more quickly. According to market analysts, Intel had a 22 percent market share in data center AI chips in early 2024, far behind market leader NVIDIA with 65 percent. AMD followed with 11 percent.
Not a new HabanaLabs?
The question with any such acquisition, if it happens, is whether the potential of the acquired party will be realized by the new owner. Intel has already ruined an AI chip manufacturer in the past, according to many. Israeli company Habana Labs was acquired for $2 billion in 2019 and could have been a springboard for Intel to be well-prepared for the AI advance three years later. However, Intel squandered that opportunity and, according to a particularly damning report by CTech, several key mistakes were made following the acquisition. For example, the cultures of the two parties did not seem to be a good match, and Intel showed little patience, bizarrely opting for a completely separate GPU development in another division simultaneously that was only cut off at a late stage.
Above all, Intel did not seem to really believe in an impending AI boom, something that, with hindsight, looks far more foolish than it would have been at the time. Now, that mistake cannot be made again. In fact, the question is whether it is already too late to catch up as an AI chip company. Only Google, with its Tensor Processing Units (TPUs), seems to be a serious alternative to Nvidia, despite frantic attempts by other parties to achieve greater market penetration. AMD is considered a reliable third option and mostly supports open standards such as Ethernet instead of Nvidia’s InfiniBand. Groq, SambaNova, and a few other parties promise amazing AI performance, but do not yet seem to be deployed at scale by hyperscalers for AI training or inferencing.
Read also: Why Nvidia’s rivals think they have a chance to topple it