Intel Purchases Habana Labs in a $2B Push into AI
Intel has been working hard to differentiate its silicon offerings and expand into new markets. The company has just purchased Israeli-based Habana Labs for $2B in an acquisition intended to “turbo-charge” its AI products.
“Habana turbo-charges our AI offerings for the data center with a high-performance training processor family and a standards-based programming environment to address evolving AI workloads,” said Navin Shenoy, executive vice president and general manager of the Data Platforms Group at Intel.
Intel has been signaling that it’s quite serious about expanding its reach beyond personal computing. Intel CEO Bob Swan said as much during a recent Credit Suisse conference in early December.
“I’m trying to destroy the thinking about having 90 percent share inside our company because, I think it limits our thinking, I think we miss technology transitions,” Swan said. “We miss opportunities because we’re, in some ways preoccupied with protecting 90 percent, instead of seeing a much bigger market with much more innovation going on. So we come to work in the morning with a 30 percent share, with every expectation over the next several years that we will play a larger and larger role in our customers’ success, and that doesn’t just mean CPUs.”
Intel’s decision to acquire Habana Labs, a company working on specialized accelerators for the AI market, has to be interpreted in the context of what Swan has said. Habana Labs has demonstrated its Gaudi AI processor, which it claims can outperform a GPU by a factor of 4x. Let’s assume, for the moment, that this is true. It makes sense for Intel to buy a company with a credible product that can challenge Nvidia or Google’s TPU in this space, but what does it say about the performance of the compute products Intel already owns, like Nervana, or the GPU IP that it’s currently working to commercialize?
This is where there may be trouble. Patrick Kennedy at ServeTheHome points out that the Intel NNP-T was supposed to compete against Gaudi AI and Habana. Intel’s decision to buy Gaudi could reflect problems with that product or that Intel has other ideas for where it wants to take its AI training products. It’s also possible that Intel is trying to hedge its bets and move quickly in the market rather than waiting to get its own solution into play.
One interesting point Kennedy raises is that Habana Labs’ Gaudi AI is designed to scale out using 100GbE Ethernet fabrics — and Intel now owns a company devoted to building that kind of solution. Nvidia’s purchase of Mellanox fits this pattern as well. Both companies have made purchases that increase their total addressable markets (TAM) within the datacenter.
The way I read these changes collectively is that Intel is genuinely trying to expand the markets it addresses with its CPU, GPU, FPGA, and adjacent products. Even if the purchase of Habana represents Intel making a different bet on a product other than the Nervana NNP-T, that might be the right move for the company to make and this would likely be the right time to make it. Not to overdraw the smartphone comparison, but one of the problems both Intel and Microsoft had (in different ways and for different reasons) was the issue of being late to market. Buying new companies with better approaches might be a better option than continually doubling down on what you’ve already got in-house. A lot of proverbial mud is being slung at the AI wall, and it’s not yet clear whose solutions and approaches are going to stick.
Continue reading
Intel Launches AMD Radeon-Powered CPUs
Intel's new Radeon+Kaby Lake hybrid CPUs are headed for store shelves. Here's how the SKUs break down and what you need to know.
SpaceX Launches ‘Better Than Nothing’ Starlink Beta
Those lucky few who have gotten invitations to try the service will have to pay a hefty up-front cost, and the speeds aren't amazing. Still, it's a new generation of satellite internet.
Samsung, Stanford Built a 10,000 PPI Display That Could Revolutionize VR, AR
Ask anyone who has spent more than a few minutes inside a VR headset, and they'll mention the screen door effect. This could eliminate it for good.
New Intel Rocket Lake Details: Backwards Compatible, Xe Graphics, Cypress Cove
Intel has released a bit more information about Rocket Lake and its 10nm CPU that's been back-ported to 14nm.