Turing Self-Driving Startup Adds AMD GPUs and AMD Ventures Backing

Japanese self-driving startup Turing Inc. has added AMD Ventures to its backers and started integrating AMD GPUs into its AI training pipeline, marking a notable shift in GPU supplier strategy for a company that previously relied entirely on Nvidia hardware. The move is part of Turing's push toward a commercial launch and aims to diversify supply and reduce costs.

What happened

Turing Inc., a five-year-old Japanese self-driving technology developer, has added AMD Ventures to its list of backers and begun adopting Advanced Micro Devices Inc.'s AI accelerators in its systems. According to company executives, Turing now handles roughly 10% of its AI training needs with AMD graphics processing units. The startup had been reliant on Nvidia hardware for both training and inference since its founding. AMD, headquartered near Nvidia in Santa Clara, California, presented an opportunity to diversify suppliers and achieve lower costs, the executives said.

Why AI builders should care

For teams building AI products that depend on GPU compute, Turing's move illustrates a growing trend: reducing single-vendor lock-in by adopting multi-vendor hardware strategies. Nvidia has long dominated AI training, but AMD's MI-series accelerators are becoming viable alternatives for certain workloads. Turing's decision to shift 10% of training to AMD GPUs suggests that the performance gap is narrowing enough for production use, at least for some tasks. This matters for AI builders who want to negotiate better pricing, hedge against supply constraints, or optimize for specific workload characteristics.

The AMD Ventures investment also signals strategic alignment. Having a GPU vendor as an investor can improve access to hardware, engineering support, and roadmap visibility. For startups building capital-intensive AI systems, this kind of relationship can be as valuable as the funding itself.

Practical implications

Turing's commercial launch timeline is not specified, but the hardware diversification is a concrete step toward production readiness. Builders evaluating GPU options should note that Turing is using AMD GPUs for training, not just inference. The 10% figure is modest, but it establishes a pattern: multi-vendor GPU stacks are feasible for self-driving AI workloads, which are among the most compute-intensive in the industry.

For AI builders, the key takeaway is to evaluate AMD GPUs as a complement to Nvidia, not just a fallback. If a self-driving startup can integrate AMD hardware into its training pipeline, many other AI products can too. The cost savings and supply chain flexibility could be significant, especially for teams running large-scale training jobs.

Caveats

Several details remain unclear. Turing has not disclosed which specific AMD GPU models it is using, nor has it shared performance comparisons between AMD and Nvidia for its workloads. The 10% training share is a starting point, and it is unknown whether Turing plans to increase that percentage or keep AMD as a secondary supplier. The commercial launch date and the scope of Turing's self-driving system (Level 4 or 5) are also not specified in the available sources.

Additionally, AMD's software stack (ROCm) has historically lagged behind CUDA in maturity and ecosystem support. Turing's ability to integrate AMD GPUs may depend on custom engineering that is not easily replicable by other teams. Builders should test AMD hardware on their own workloads before committing to a multi-vendor strategy.