Tesla is developing a chip for autonomous driving (TSLA)
Wikimedia Commons
- Tesla CEO Elon Musk recently confirmed that his company is developing a custom processor dedicated to AI.
- It's still unclear whether Tesla would be able to handle the production load that would accompanying an AI chip rollout.
Just in case it wasn't abundantly clear yet, Tesla CEO Elon Musk has confirmed that the company is indeed hard at work developing a custom processor dedicated to artificial intelligence (AI). The electric-car maker hired legendary chip architect Jim Keller in early 2016, and reports emerged in September that Tesla was making a custom chip that could potentially supplant the NVIDIA graphics processors that currently power Tesla's semi-autonomous driving system, Autopilot, in the future.
Tesla had initially declined to comment on those September reports, but as is often the case, Musk decides what Tesla does and doesn't share with the media. Musk often front-runs his communications teams with premature announcements, sometimes on social media. "Jim is developing specialized AI hardware that we think will be the best in the world," Musk reportedly said.
Why companies develop custom chips
Generally speaking, the reason why some companies pursue custom or semicustom chips is because they believe they can improve performance, power efficiency, and costs (or some combination of all of the above) by designing chips instead of buying off-the-shelf parts.
The most successful custom chip strategy is easily Apple's, which designs a growing portfolio of silicon specially catered to specific purposes, allowing the Mac maker to optimally balance performance and power efficiency while saving on costs. In doing so, companies can also exert greater control over their technological roadmaps, instead of having to rely on chip suppliers to pave the way for spec bumps and incremental refreshes.
Speaking at the event, Keller reportedly said Tesla can improve efficiency through specialization, while Musk said the company can potentially improve power while reducing costs.
https://twitter.com/mims/statuses/938991222034546689?ref_src=twsrc%5Etfw
@jekbradbury asks what's missing in current AI tech and whether Tesla use open source existing stacks or custom. Jim notes the efficiencies gained by specializing and removing overheads between components. Elon/@karpathy note this can give 10x the power at a tenth the cost.
Tesla is a huge fan of vertical integration, so designing a custom chip fits entirely within its broader approach to product development. Besides, it was immediately clear what Tesla had in mind when it hired Keller, given his extensive background in designing chip architectures. Alphabet's autonomous driving subsidiary Waymo is also exploring custom chips, but is mostly working with supplier Intel to do so. "You're going to see us building custom silicon and custom technology for Waymo," Intel CEO Brian Krzanich told CNBC in September.
Will it be worth it for Tesla?
The main consideration for investors is whether or not developing custom silicon makes financial sense. Developing custom chips in-house (as opposed to collaborations and partnerships like Waymo/Intel) is massively expensive and takes years. To justify those costs, scale is incredibly important, since ideally companies want to amortize those costs over a large volume of units, reducing the average unit cost.
For a company like Apple, which has shipped 260 million iPhones and iPads over the past year (each carrying a custom A-series chip), it's a no-brainer. The Mac maker also uses A-series chips in other products like Apple TV, iPod touch, and the forthcoming HomePod. For a company like Tesla, which has shipped 95,000 vehicles over the past year, it's less clear. Of course, Tesla hopes to exponentially ramp production and delivery in the years ahead, which would help justify the custom chip development if it can execute.
As if there wasn't already enough pressure to ramp production.
NOW WATCH: Here's why Boeing 747s have a giant hump in the front