DeepSeek AI was in the news worldwide after its launch. It was being compared to global AI models such as ChatGPT and Google Gemini. The very fast emerging Chinese contender was then seen as a threat by America’s Silicon Valley tech companies. The Chinese government declared that DeepSeek AI was based on purely Chinese technologies.
DeepSeek AI was being trained on HUAWEI’s Ascend chips. They stated that such chips would bring the development of AI to the next scale. The physically printed statement said that the company’s technologies are free of any American dependencies. Nothing could be further from the truth.
The engineers working for DeepSeek have now admitted that the HUAWEI chips do not offer reliability. They are unreliable in performance and applicability for scaling training for AI models.
To put this in, low-speed network connections and the absence of tools in support of the software; these considerations have not worked alongside.
According to the engineers, the chips could only help in developing user interfaces but not model training.
To make further advances on DeepSeek’s R2 model, they needed NVDIA chips with the powerful A100 and H100 series. These chips are crucial for sufficiently training models and efficient high-performance computing. DeepSeek AI will not have a legitimate chance to challenge leading global models without these chips.
The restrictions imposed by the US-China trade war now block advanced chips from getting to Chinese businesses; restrictions that were enforced during Donald Trump’s presidency and now threaten to keep tech businesses from emerging. DeepSeek has little prospect of obtaining enough high-performance hardware to move their development forward.
According to reports, DeepSeek has been acquiring NVIDIA chips through shell companies in Singapore and throughout the Middle East. That does not give DeepSeek a reliable supply of chips long-term, and DeepSeek still requires a huge number of NVIDIA chips to maintain and improve their performance.
DeepSeek is also still blocked with the CUDA software ecosystem, which is a prerequisite for rapid AI development. In the absence of CUDA, the performance of DeepSeek’s AI systems declines significantly. This has continued to slow down DeepSeek’s efforts to catch-up with higher order models.
Share
How to Save Money on Petrol and Diesel with Credit Card?
FASTag Fraud Alert: How To Protect Your Account From Online Scams
