Waymo says it has built an ASIC chip that will improve its robotaxis' reflexes and navigational skills and help it diversify away from third parties like Nvidia
https://lnkd.in/...Satish Jeyachandran:Today, we're providing a literal peek under the trunk of the Waymo Driver at our compute, including our new custom 5nm ASIC. …
Context & Ripple Effects
Waymo has long pursued control of its autonomous-driving stack, from its earlier move to produce self-driving technology in house to its work with DeepMind on driving AI. Its more recent use of synthetic worlds for edge-case training adds a software-and-data counterpart to the new compute effort.
The new ASIC makes that vertical-integration strategy more consequential: Waymo is targeting the hardware running the Driver while reducing reliance on Nvidia for a core part of the robotaxi system.
First-order effects
- Waymo gains a custom 5nm compute component intended to improve the Waymo Driver's reflexes and navigation, while shifting part of its robotaxi compute stack away from third-party hardware.
- Nvidia faces a major autonomous-driving customer explicitly seeking less dependence on its chips, even as Waymo continues to build out its own Driver platform.
Second-order effects
- Waymo's robotaxi rivals must increasingly compete on how tightly they co-design hardware, driving software, and training workflows—not only on vehicle deployments.
- Nvidia's automotive position faces added pressure from customers that can turn specialized driving workloads into internal ASIC programs rather than relying solely on general-purpose AI hardware.
Third-order effects
- If more autonomous-vehicle developers follow Waymo's path, the competitive unit shifts toward an integrated stack spanning simulation, models, and purpose-built inference hardware, rather than a standalone chip supplier.
- That structure may widen the advantage of developers able to sustain both AI training programs and custom-silicon design, while making external compute vendors more dependent on serving the remaining layers of the stack.
The trend: Autonomous-driving developers are moving from buying AI compute toward vertically integrated systems that pair proprietary training pipelines with workload-specific silicon.