A company that used to buy chips now designs them
Waymo said on 20 August 2026 that its newest robotaxi generation, built on the Zeekr-manufactured Ojai platform, runs a chip the company designed itself rather than one it bought from a supplier. In a post titled "A look under our trunk: what's in our compute," Waymo described co-designing hardware, sensors and algorithms together to fit the constraints of real-time edge compute inside a moving vehicle, rather than treating the chip as an off-the-shelf component bolted onto a self-driving stack built by someone else.
The custom application-specific integrated circuit, or ASIC, is manufactured on TSMC's 5-nanometer process and is rated at more than 1,000 trillion operations per second, according to Bloomberg's reporting on the announcement. That figure matters because it is not a downgrade dressed up as independence: Waymo's own reporting places the new chip's raw throughput in the same range as Nvidia's latest automotive-grade autonomous-driving platform, the supplier Waymo has relied on alongside AMD since the earliest generations of its fleet.
The economics behind the switch
Every robotaxi mile requires a vehicle to process camera, lidar and radar streams fast enough to react to traffic in real time, and that processing has historically run on chips bought from Nvidia at Nvidia's margin, or AMD at AMD's. Designing that silicon in-house removes a per-vehicle line item that scales with every car Waymo adds to its fleet in San Francisco, Phoenix, Los Angeles, Dallas and now Houston, where the company confirmed full public availability the same day as the chip announcement.
Alphabet is not new to this move. Google has designed its own Tensor Processing Units for AI workloads in its data centers for years, a decision that let Google Cloud avoid buying every accelerator from Nvidia at list price. Waymo's chip is the same logic applied to the edge: instead of a data center rack, the constraint is the trunk of a minivan, and instead of training a model, the job is reacting to a pedestrian in under a second. Bloomberg noted the chip's TOPS rating matches Nvidia's current automotive silicon, meaning Waymo did not trade performance for cost control.
What this changes for everyone who is not Alphabet
The obvious reading is that this is good news for Nvidia's rivals in autonomous driving, since one of its largest customers just became a competitor in edge AI silicon. The more useful reading is the opposite: Waymo's move raises the capital and design bar for every robotaxi operator that does not have Alphabet's balance sheet, its TSMC allocation, or a decade of Google TPU design experience to draw on. Tesla and Amazon's Zoox, the two operators most directly chasing Waymo's US lead, either continue paying supplier margin on every vehicle or attempt the same vertical chip program without Alphabet's manufacturing relationships.
This table sets Waymo's new chip against the Nvidia platform it displaces and the two rival approaches still buying merchant silicon.
| Platform | Operator | Compute | Process node | Chip sourcing |
|---|---|---|---|---|
| Waymo custom ASIC | Waymo (Alphabet) | 1,000+ TOPS | TSMC 5nm | Self-designed |
| Nvidia autonomous-driving platform | Third-party OEMs | ~1,000 TOPS class | TSMC (Nvidia design) | Purchased from Nvidia |
| Prior Waymo generations | Waymo (pre-Ojai) | Not disclosed | N/A | Nvidia and AMD |
For a European operator watching this from the outside, the lesson is not about robotaxis specifically. It is that the deepest-pocketed AI platform companies are quietly moving compute cost out of their supplier's margin and into their own balance sheet, one product category at a time, and every company that competes with them on thinner capital will feel that cost gap widen before it narrows.
Read next: Meta Starts Making Its Own AI Chip in September | TSMC's $29 Billion Buys a Tenth, Not a Fix
