> The US trade war with china which will place deepseek compute availability at disadvantages
I doubt it'll make much difference. Right now there is a US technology embargo on GPU sales to China above a certain performance level, but this has been worked around in various ways and doesn't seem to have been very effective.
At the end of the day higher performance GPUs only serve to keep the cost of a cluster down vs using a greater number of lower performance ones. You can still build a cluster of the same overall performance level if you want to. Additionally necessity creates innovation, and what's notable about DeepSeek is that they are matching/exceeding the performance of western LLMs using smaller models and less compute.
Not only that, but having a constraint often feeds innovation. Having to work with less compute might mean new ways of doing things that leads to faster iteration, etc.
I doubt it'll make much difference. Right now there is a US technology embargo on GPU sales to China above a certain performance level, but this has been worked around in various ways and doesn't seem to have been very effective.
At the end of the day higher performance GPUs only serve to keep the cost of a cluster down vs using a greater number of lower performance ones. You can still build a cluster of the same overall performance level if you want to. Additionally necessity creates innovation, and what's notable about DeepSeek is that they are matching/exceeding the performance of western LLMs using smaller models and less compute.