Artificial Intelligence (AI) chip-based startups are a magnet for venture capitalist (VC) funding and fast-paced innovation around the globe. But not everyone is in agreement about their worth.
Some argue that these companies are doomed from the start―due to their high startup costs and pricey software subscriptions―but AI chip startups have found strong footholds in Silicon Valley, Beijing, and other tech startup hubs. The year 2020 is only the beginning of AI chip startups, so let's discuss where we expect to find them and why.
Server Computing vs. Edge Computing
As of early 2020, AI hardware exists in two realms: server and edge computing.
- Server computing is most synonymous with AI training models and the development of profoundly complex algorithms.
- Edge computing is the field-implementation aspect of AI, whereby these models and algorithms are deployed in a finished product to serve a specific purpose.
Where AI Chip Startups Will Not Be
We expect to see edge computing-focused AI startups flourish in the coming year. At the moment, server computing is dominated by cloud computing giants like Google, Amazon, Facebook, and Apple. Silicon Valley heavyweights such as AMD, Intel, and NVIDIA have general server-grade CPUs and GPUs that are popular with cloud computing giants. Still, the majority of the server computing hardware market is moving towards custom hardware that's designed to run custom server applications. This trend means the Googles, Amazons, and Facebooks of the world are hard at work designing and building their own highly optimized servers for their niche applications.
While there certainly are startups that focus on optimizing inference processes and other server-level AI developments, these startups tend to fail or get acquired quickly. Ultimately, the same companies that are developing the proprietary AI networks are the ones making most of the progress in this market.
Where to Find AI Chip Startups
Currently, you can find AI chip startups solving power and processing capability issues in edge AI applications. Companies are also creating CPUs, GPUs, and TPUs―such as the Intel Compute Stick or NVIDIA Jetson Nano―that can run edge AI loads, but these may fall short in some AI applications. Edge AI Applications
Edge AI applications that become widely adopted in a specific sector tend to have remarkably fine-tuned networks and algorithms, enabling them to solve specific problems. In some instances―such as automated delivery drones or robots―these networks and algorithms are highly complex and require substantial levels of energy to run correctly. Unfortunately, many edge AI applications are powered by a battery or solar panels, which directly limit the amount of energy available for heavy compute loads. However, like server computing, increasing chip-level efficiencies can effectively decrease energy consumption across the device. A startup company that develops a more efficient chip will be poised for great success in the edge AI world.
AI Chip Startups 2020 Focus
With most of 2020 still in front of us, we predict that this year's new AI chip startups will focus on research and architecture with the goal of reducing latency, increasing efficiency, or even designing AI-geared 5 nm chips. Regardless of their specific goals, AI chip startups in 2020 will likely target edge computing applications as opposed to server-level computing. Either way, we expect 2020 to be a year that pushes the boundaries of AI technology as we know it.