DeepSeek had the world’s stock markets in a churn the other day, giving blue chip Nvidia and several US tech titans a shock as their worth plummeted.
But as Pranay Kotasthane argued on The Gist, it’s not clear if DeepSeek is really the game changer in AI, is it as cheap as its makers claim it to be, will it be overtaken by others in the days to come?
“The $5 million talked about is the training cost, training cost for any AI model,” Kotasthane said, “they have used a bunch of optimizations, these are not new they are known to the software world, and in the US as well.”
He says it does not mean Chinese models are beating us because this is just the start with OpenAI and others coming up with new models.
Kotasthane heads the high tech geopolitics programme at the Takshashila Institution, a think tank based in Bengaluru, and his job requires him to closely monitor the tech space and its geopolitical implications.
What’s clear is that US tech titans have lost some of their sheen but could lose even more given the sheer number of companies in this game. These include not only China, also France where Mistral is a known competitor.
The other question being asked is do US export controls work or not? It’s clear DeepSeek used some Nvidia GPUs, how many and of what kind? Maybe they got them before the new export controls came in, so they had a ready stock of GPUs.
The DeepSeek phenomenon does not mean China is ahead in this game, Kotasthane says. India can also get a foot in because it may not come under the kind of restrictions imposed on the export of GPUs to China.
How come India did not think of this? Kotasthane attributes this to what he calls “creative insecurity”, where the sum total of external threats exceed the internal threats, it pushes countries to invest in R&D budgets that spur innovation.
Tune in for more in this chat with Pranay Kotasthane of the Takshashila Institution.
