Li Auto Unveils Self-Developed 5nm Chip Mach M100 with Over 3x the Computing Power of NVIDIA's Thor U
Taylor Wilson
Li Auto disclosed the Mach M100, a 5nm self-driving chip delivering 1,280 TOPS — roughly triple Nvidia's Thor U — debuting in the L9 Livis. If mass-produced successfully, the chip would give Li Auto full-stack control across silicon, software, and AI models, ending its reliance on Nvidia's supply chain.
What makes this chip stand out?
Mach M100 is built on a 5nm process and delivers 1,280 TOPS (trillions of operations per second — the key measure of how fast a chip can crunch self-driving data). That is roughly three times Nvidia's Thor U on a single chip.
End-to-end latency — the total time from camera "seeing" to car "reacting" — drops by 40%. This means → the car decides faster in the same traffic scenario, widening the safety margin.
Li Auto stresses this is not a tweak on an existing Nvidia architecture but a ground-up redesign on a proprietary architecture. In plain terms = they didn't tune someone else's engine — they built a new one from scratch.
Why is Li Auto building its own chip?
If Mach M100 reaches mass production, Li Auto will control the full stack — chip, software, and AI model — with end-to-end hardware-software integration.
This means → Li Auto no longer waits on Nvidia's allocation schedule or supply cadence. It sets its own iteration pace.
This reflects a deeper industry signal: China's leading EV makers are shifting from "buy parts, assemble cars" to "vertically integrate starting at the chip level."
Is Li Auto alone — how broad is China's EV chip wave?
BYD earlier this year unveiled its in-house 4nm chip "Xuanji A3," paired with its "Eye of God" driver-assistance system, pushing the feature down to budget models.
NIO launched its 5nm chip NX9031, already deployed across multiple models. NIO's founder noted that semiconductors and batteries together account for over half of a new car's cost — in-house chips are the key lever for vertical cost compression.
In plain terms = chips and batteries make up more than half the cost of a car. Whoever can build their own holds an extra card in the price war.
The bigger picture: losses are forcing an efficiency pivot
Per Digitimes, in Q1 2026 Leapmotor, XPeng, Li Auto, and NIO all posted net losses. The industry is pivoting from "trade market share for growth" to efficiency and profitability first.
In-house chips can lower component costs long-term, but they require heavy upfront R&D spending, and switching away from third-party chips carries high migration costs.
This means → building your own chip is a "spend big now, save big later" bet — it deepens losses in the short term and only pays off if production scales successfully.
The key question: how far from launch to mass production?
Mach M100 is at the spec-disclosure stage. It is slated for the L9 Livis first, but no firm timeline for volume production has been announced.
Whether Li Auto can deliver on both cost and performance targets will be the definitive test of its full-stack strategy's commercial viability.
In plain terms = specs on paper are one thing. Getting the chip into cars at scale — and keeping costs under control — is an entirely different challenge.
Content is for reference only, not financial advice.