Chip makers are chasing the advancements made by big players like Apple and Google. The AI chip startup scene is particularly buzzing with activity in China, following a US government ban on selling high-end chips to the country.
In 2012, when researchers at the Google Brain project announced that a cluster of computers had trained themselves to recognize images of cats from YouTube videos, it was heralded as a breakthrough in artificial intelligence.
But Andrew Ng, who was part of the group that led the research, would later tell Wired that some researchers walked away from the project —which ran on 1,000 computers and cost roughly $1M — skeptical of its feasibility.
A year later, Ng and other scientists published a paper on achieving the same results using a cluster of Nvidia’s graphic processing units (GPUs) on just 3 machines. This marked the beginning of a new era in artificial intelligence. Along with the availability of big data and advances in machine learning algorithms, Nvidia’s chips provided researchers with the horsepower required for training AI models.
“Today, some of the world’s largest internet companies, as well as the foremost research institutions, are using [Nvidia’s] GPUs for machine learning.” – Larry Brown, Nvidia
When Nvidia launched the graphic processing units (GPUs) in 1999, they were primarily aimed at the gaming industry. But the chips soon showed promise for training processor-intensive AI algorithms. Today, the chips dominate the industry, with everyone from AI startups to corporations like Baidu and Google relying on GPUs from Nvidia.
Other big players have now entered the hardware arena to build chips specifically optimized for deep learning.
- As the dominant processor in data center stacks, Intel has had to play catch-up with a massive acquisition of programmable chipmaker Altera for $16.7B, whose chips help accelerate AI tasks, as well as snapping up AI-powered startups like Saffron, Nervana, Movidius, and Mobileye.
- Google released the Tensor Processing Unit (TPU) earlier this year that would integrate with its TensorFlow open-source library.
- Apple recently released its first machine-learning optimized A11 Bionic chip, with Apple-designed GPU cores, to enable AR and facial recognition in the upcoming iPhone.
An interesting trend is the emerging deals to startups in the space, some of which consist of employees who worked for competing projects like Google’s TPU and Baidu’s deep learning institute. The business model of these early- and mid-stage companies that are locking horns with big corporations is unclear, but it won’t be surprising if they fall into the aggressive M&A pattern that the AI industry is witnessing.
The Chinese chip market is particularly buzzing with activity following a US government ban on selling high-end chips to China, and the Chinese government’s push towards locally developed chips.
Using the CB Insights database, we surfaced 9 startups creating hardware optimized for AI applications.
