THE Excellent CATCH-UP
But, there are arguments to refute the claim that this is not one of those situations, and it is possible that the same may occur.
First, our knowledge of DeepSeek may become inadequate. The single pre-training expense at the reported US$ 6 million was for pre-training, not the entire R&, D cost. Although US$ 6 million is still less than the US counterparts ‘ pre-training budget, the difference is in the thousands rather than the trillions as claimed.
Experts who monitor export controls have made the same speculacies about DeepSeek having access to tens of thousands of border Nvidia cards, some of whom are not all of the highest specs. According to Semianalysis ‘ experts, DeepSeek’s technology spend is” properly over US$ 500 million over the firm story.” This indicates that both the general R&, D costs and the hardware and software needed have no significantly decreased.
Second, perhaps with DeepSeek’s value efficiency gains, it might not weaken infrastructure investments. Dario Amodei, CEO of one of the border LLM businesses, Anthropic, argues that the notion of” second everything is expensive, then it gets cheaper” does not use to AI.
He stated that in the past, when prices go lower due to advancement ( or when the “scaling curve transitions”, in specialized terms ), businesses end up spending more, rather than less on training concepts. The logic is that the price of higher-performance AI significantly outweighs any expense savings, so firms reinvest those savings into more sophisticated AI growth, rather than reduce infrastructure investments.
Third, greater productivity was increase demand. Some businesses may lower the barrier to entry because of DeepSeek’s efficiency, which allows them to create or employ applications using the model. This is very dependent on how effective the concept, and its smaller varieties, are embedded in other daily tools, and how much this is used by customers.
What DeepSeek does is to make an opportunity and possibility. It overtook Alibaba for 2024 as the best AI concept, and it has now surpassed OpenAI in the field of frontier AI models. This so-called “great catch-up” is only the start. Either approach, Malaysia’s wager on AI-driven system will soon be put to the test.
Sang Kancil ( Penguin Random House ) is the author and social scientist James Chai writes.