One of the things that has differentiated Intel in the AI space is that while their CPU technology is strong for inferencing, it wasn’t good for training. And, without training, getting optimal inferencing performance can be problematic.
However, break out firms like Deep Instinct, the firm HP is partnered with for next-generation AV, have shown that you can train on GPUs and have powerful inference engine on even mobile X86 CPUs. Now they have a powerful GPU under development – so why not jump ahead and instead of using a technology that evolved around a different use case – graphics – build something that was laser-focused on Deep Learning Artificial Intelligence (DL AI)?
That appears to be the strategy with Intel’s buying of Habana. Yet rather than developing the technology, hey are buying it with the company that is developing it and agreeing to leave the company alone.
Deep Learning Artificial Intelligence
Currently, AI can be delineated in several ways. By process, where you get ML (Machine Learning), which requires a lot of Data Scientist time for training, has smaller data sets, but often has issues concerning accuracy and bias.
Deep Learning eliminates much of the Data Scientist requirement. Still, it moves to massive training data sets, making it both more expensive and far more effective and accurate (critical for things like Autonomous Driving where mistakes can be fatal). And there's also Symbolic AI, which is a hybrid of the two but, until recently, was out of favor.
Then you have Focused AI, which is what we currently have in the market where they AI can do one thing well. The next phase, expected mid-next decade, is Broad AI, where an AI can do multiple jobs but is still far less flexible than people are. And, finally, we have General AI which is the AI that can do what people can do and more, and it is expected in the 2035-2045 time frame.
Habana is a very secretive company out of Isreal. It is a startup and well-run startups that survive often do so because they hold their technology very close during the early years to prevent theft.
We do know they are in many trials with large companies and Facebook, a firm with a critical need for DL AI to mitigate the kinds of problems they’ve had in politics, which is the customer they referenced in a recent press release.
Intel had already dropped a whopping $75M into Habana, and they don’t do that unless they see a clear long term opportunity with the company. This early investment also suggests they could fast track due diligence because they would have already done that when they made this earlier investment.
The move was designed specifically to provide stronger competition for the market leader, NVIDIA, in the space as it ramps to its $25+ estimated billion-dollar potential by 2024. Habana's Chairman Avigdor Willenz has an impressive success record, having sold Galileo Technologies to Marvell Technology Group for $2.7B in 2001 and Annapurna Labs to Amazon for $370M in 2015. Intel paid $2B for Habana.
This purchase will bring Intel’s presence in Israel to five facilities, making them well established in this unique tech and security-focused country.
Intel’s history with acquisitions isn’t great, but given the founder will be staying on to protect his company, the firm is relatively small compared to Intel. And Intel’s CEO, Bob Swan, is their ex-CFO and has performed impressively well to date. There is good chance this merger will set a better standard.
Besides, while there will be some conflict for inference, Intel lacks a good DL AI training engine so the tendency will be to look to Habana for that critical skill. And Intel’s sales and support force are world-leading, which should give Habana, if it successfully integrates into Intel’s sales process, substantial upside.
This acquisition should also help Habana move outside of Data Centers, where they are initially focused, into other interesting AI areas like Autonomous Driving and into areas where Intel now has a focus like Autonomous Driving and Security.
Intel’s acquisition of Habana adds additional competition to the AI technology space and should enhance their already strong Inference AI capability. On paper, and eventually, this could result in a stronger solution than NVIDIA has. However, Intel is significantly behind in terms of developer support and market penetration giving NVIDIA, which remains the market leader, significant time to assess and respond to this threat.
So, while this will make Intel more competitive, without a significant developer, marketing, and additional sales effort, it won’t alone make them a true market leader.
Intel does have the resources to make this a fight, but only if they execute and, even so, it will take several years to build the other resources to make them a true challenger for the top. Much like Azure was the test of the current Microsoft management, which they passed with flying colors and now challenge Amazon Web Services as a peer, this will be a test of the new Intel management. Let’s see if they do as well. The strategy looks solid but execution will be critical for success and Intel does have a mixed record here.