Intel (NASDAQ: INTC) announced on December 16, 2019, that it was ready to spend $2 billion for Habana Labs, a start-up that makes artificial intelligence chips. Habana Labs was founded in 2016 and has offices in Tel Aviv, San Jose, Beijing, and Gdansk, Poland. It has raised $120 million, including $75 million in a funding round led by Intel Capital last year.
Habana is not Intel’s first foray into the AI chip market. Intel has been shopping for several growth opportunities as it deals with challenges in its core markets.
In June of 2015, Intel announced that it intended to buy Altera, a supplier of chips known as field-programmable gate arrays, or FPGAs, for $16.7 billion. The rationale behind the acquisition was Intel’s belief that by 2020 about a third of the data center market could be using the type of chips in which Altera specializes.
In 2016, Intel bought two companies; Movidius, a specialized low-power processor chips maker focused on computer vision, and Nervana, a startup considered among the leaders in developing machine learning technology. Movidius acquisition gave Intel access to new devices outside of the traditional computer segments. Devices like Drones, Robots, Virtual Reality Headsets, and alike need computer vision and deep learning solutions that Movidius focused on. Nervana brought machine learning technology but also brought specialized chip designs to Intel, which the company unveiled in November 2019 with new products for data centers.
Later in 2017, Intel acquired automotive camera maker Mobileye NV for $15.3 billion betting on self-driving cars. Mobileye has developed a nerve center for how driverless vehicles operate via their sensors and camera technologies. Completing this acquisition put Intel squarely in one of the “driver’s seats” for driverless cars, estimated to become a $70B industry by 2030.
The four acquisitions signal to the market that Intel is following a broader strategy of chasing new growth markets vs. simply maintaining dominance in the traditional CPU market. It is no secret that Intel has missed the smartphone market. The company has no intention to miss the AI opportunity and has made artificial intelligence the focal point of the company’s growth strategy.
The company has a lot of upside. While Intel has a significant market share in the CPU market, but the company has a small percentage in the quickly growing AI chip market. That should not be surprising, as Intel has increasingly been depending on sales to data centers as PC sales stagnate.
Everyone expects the AI chip market to grow exponentially over the next few years. Intel believes that the demand for AI chips should grow to $25 billion by 2024. The company expects half of that revenue to come from selling chips to data centers.
Before announcing the Habana purchase, Intel expected to generate over $3.5 billion in AI-related sales in 2019, up around 20% from last year.
The Habana Labs acquisition comes with a lot of expectations. It gives Intel’s AI chip portfolio a substantial boost. Especially considering that AMD had recently launched new chips for data centers that outcompete Intel’s by many performance metrics, the acquisition should provide leverage to Intel.
We are unclear what the immediate revenue contribution from Habana Labs will be, but we expect it should be negligible in the short term. However, we believe that the Habana deal provides Intel a stable strategic position primarily because of the products and technology the startup brings.
Habana Labs has two main products, Gaudi processor for training and the Goya processor for inference referring to two different phases in machine learning projects.
In the training phase, the algorithm is tested on a massive data set. The system learns to extract patterns and create machine learning models that can capture these patterns.
In the inference phase, these models are fed with new data to generate desired outputs. Given the amount of data and real-time expectations, both training and inference require significant computational power. By way example, computer vision in an autonomous vehicle requires image analysis and object detection in real-time.
The AI system needs to detect the location of objects and identify their classes (e.g., cars, pedestrians, street signs, traffic lights, crossings) and recommend an action. The deep learning algorithms employed need to be trained using massive data while driving thousands of miles. And then they need to be employed and work in real-time driving situations. It turns out that CPUs are not specially designed for the type of calculations that machine learning needs.
Over the last few years, we saw companies like Nvidia win in the AI chip market with Graphical Processing Units (GPUs). Today Nvidia is the number one player in the AI training chip segment. Other players include Google’s Cloud TPUs (Tensor Processing Units), and startups such as Habana and GraphCore also compete in these segments. Habana’s website claims that their Gaudi AI training processor can “easily beat GPU-based systems by a factor of four.”
Intel’s acquisition, therefore, is an attempt to tackle both AI computational segments in a robust way.
Considering that the growth of the AI chip market is a function of the deployment of Artificial Intelligence in enterprises globally, Intel’s move should position them well for success.
Intel is becoming a serious player in the AI computational segments and has an opportunity to combine vision with the brain, as described by Karl Freund.
Considering that the next ten years we’ll see the explosion of autonomous vehicles and advanced driver assisted systems, we may see Intel squarely in the midst of delivering AI chips, and machine learning systems for new vehicles such as drones, autonomous cars, robots and alike.