DeepSeek, the Chinese artificial intelligence startup, is developing its own specialised inference chip to reduce its dependence on NVIDIA and Huawei. This move follows in the footsteps of giants like Google, Amazon, and Meta.
DeepSeek, the Chinese company that burst onto the artificial intelligence scene with efficient models, has decided to take a strategic step towards hardware. The company is developing its own AI chip, as Reuters has learned from sources close to the company. The goal is to compete directly with NVIDIA in the inference chip segment, the phase following the training of models.
What is inference and why is DeepSeek betting on it?
Inference is the process by which a pre-trained AI model applies its knowledge to make predictions or classifications. It is the operational part, used daily when a chatbot responds or a vision system identifies an object.
A chip specifically designed for inference allows these tasks to be accelerated and energy consumption to be reduced, which translates into lower operational costs. Until now, DeepSeek relied on NVIDIA and Huawei chips for both training and executing its models. With its own chip, the company seeks independence and control over its infrastructure.
According to sources, the project is still in the early design phases and a functional prototype is not expected in the short term. However, the commitment is clear: DeepSeek wants to stop being just a software company and become a full player in the AI ecosystem.
The industry trend: proprietary chips to avoid dependence on NVIDIA
DeepSeek is neither the first nor will it be the last. Google has been using its TPUs for years, Amazon has the Trainium and Inferentia chips, Microsoft has developed Maia, and Meta is working on its own designs. All are seeking the same goal: to reduce dependence on NVIDIA, which dominates more than 80% of the AI chip market.
DeepSeek's move is particularly relevant because the Chinese company gained global notoriety with models that offered performance comparable to industry leaders but with much lower computational costs. Now it applies that same efficiency philosophy to hardware.
For Spanish companies using cloud AI services, this diversification of suppliers could translate into more options and more competitive prices in the medium term. The current dependence on NVIDIA is a risk that many corporate clients are beginning to take into account in their procurement strategies.
The challenges for DeepSeek: investment, time, and U.S. restrictions
Designing an AI chip is no easy task. It requires an investment of several hundred million dollars, specialised engineering teams, and years of development. Additionally, there is the manufacturing aspect: DeepSeek will need partners like TSMC or Samsung, and here geopolitical complications arise.
U.S. export controls limit Chinese companies' access to advanced semiconductor technology. DeepSeek will have to navigate these restrictions to obtain the most advanced manufacturing processes or settle for less efficient nodes that could hinder the performance of its chip.
Despite this, the mere announcement that DeepSeek is working on its own hardware has already generated excitement in the market. The Chinese startup shows that it is not content to be a secondary player and is willing to fight on all fronts of artificial intelligence.
For investors and analysts, the news represents a new element to watch: if DeepSeek achieves its goal, it could alter the balance of power in the AI chip market, which is currently dominated by NVIDIA with an iron fist. But the road is long and filled with obstacles.
You may also like
