Sphere Semi CEO Steven Glinert argues that Huawei's AI hardware and rack-level engineeringAn AI inference chip is a processor optimized to run trained AI models efficiently when they generate predictions or other outputs. are becoming competitive enough to attract national infrastructure projectsAI compute infrastructure is the hardware, facilities, networks, power, cooling, storage, and software used to train and run AI models. outside the closest US allies. He says lower costAI model affordability is the practical ability to access and use a model within available financial and technical resources., configurability and compatibility with Chinese open-source models can matter more to some countries than access to the newest Nvidia systems.
Glinert says restrictions on advanced Nvidia sales to ChinaA chip export control is a government rule that limits where certain advanced computing chips or related technology may be sold. create more incentive for Chinese buyers to fund and adopt Huawei alternatives. From his slower-takeoff view of AI progress, the rest of the world remains economically and strategically important because productivity gains and infrastructure standards will spread across many markets over years.
Glinert examines the former US diffusion-rule tier system and proposes a middle ground combining clear limits with competitive incentives, more flexible domestic hardware options, cloud accessA cloud computer is a remotely hosted computing environment that provides applications, files, networking, and execution without running directly on the user's device. and coordinated model governance.
Watch on YouTube
