Skip to main content
2025-01-01

Question of the Day

Question of the day · 2026-07-07 ·

One question per day to look beyond the headlines.

How does building an inference chip—rather than a training chip—actually reduce DeepSeek’s dependence on Nvidia and Huawei?

Take-away Inference is the scalable “deployment bottleneck,” so owning inference silicon cuts supplier risk even if training stays outsourced—and it sidesteps export bans aimed at top-end training GPUs.

DeepSeek aims to reduce its dependence on Nvidia and Huawei by developing an inference chip, which is designed specifically for generating responses from pre-trained models rather than training new models [2], [4]. Currently, DeepSeek relies heavily on Nvidia and Huawei chips for both training and inference tasks. By focusing on in-house inference chip development, DeepSeek can lessen its reliance on these external suppliers for deployments of its AI models, shifting towards hardware control in its operations [1], [4]. Additionally, due to export controls and limited access to advanced Nvidia chips, DeepSeek's strategy focuses on controlling the part of the AI process (inference) that does not require the high-performance chips restricted by these export controls [5]. This strategic move aligns with reducing potential supply chain vulnerabilities and ensuring consistent hardware availability for deploying AI models [3].

Sources · 2026-07-08