Turing’s $600 million self-driving bet shifts part of AI training to AMD GPUs, indicating intensifying GPU competition

Turing, a self-driving startup described in a new report as backed by $600 million, is moving part of its artificial intelligence training to AMD GPUs, a shift that highlights how rapidly buyers are diversifying chip suppliers in the AI buildout. The company also disclosed that AMD Ventures is joining the funding roster as a new backer.

According to the report, Turing will allocate 10% of its AI training to AMD GPUs. AI training refers to the compute-intensive process of teaching machine-learning models to recognize patterns, such as how to interpret sensor inputs for driving and navigation. In practical terms, the split suggests the startup is treating hardware mix as a lever, balancing performance, cost, and supply considerations while developing its autonomy stack.

The same report states that Turing has brought in AMD Ventures as part of its broader capital base. The addition matters because venture support can come with more than money, including engineering collaboration and access to hardware and software enablement that can reduce deployment friction for a compute-heavy project.

The report also frames Turing’s technical and commercial timeline around a 2028 launch, implying that the training pipeline choices being made now could determine how quickly the company iterates model performance and scales compute as it moves from experimentation toward production-grade systems.

For the broader industry, the development underscores a central shift in AI infrastructure purchasing: hyperscale and high-performance computing buyers are no longer limited to a single GPU ecosystem. Even when one vendor has deep mindshare, workloads can be partitioned across suppliers to manage risk, negotiate pricing, and address specific performance characteristics of different architectures.

Competition at the GPU level has become a major theme in AI markets. When a promising robotics company chooses to allocate even a portion of training to an alternative chip supplier, it indicates that the addressable market for competing GPU vendors is not confined to research labs, but can extend to startups building real-world autonomous systems.

What is still unclear from the report is how Turing will implement the 10% split in practice, including which AMD GPU models it plans to use, whether the workload is for specific model families, or how the company’s training results compare against any retained baseline using its other hardware. The report also does not detail any software stack changes, performance metrics, or vendor commitments beyond AMD Ventures’ role.

The next thing to watch is whether Turing’s GPU mix becomes more pronounced over time, and whether other autonomy and robotics companies follow a similar pattern as they plan long training horizons ahead of production timelines. Additional disclosures, such as hardware specifications or benchmarks, would be key to assessing the durability of this shift.