For US players, it’s basically going back to the drawing board.
The out-of-nowhere emergence of OpenAI rival DeepSeek uncovered a new reality where the biggest budgets, fastest tools, and best resources wouldn’t equate to holding the position of ‘global leader’ in AI. China’s ability to develop its own powerful AI despite significant limitations has ignited a global race towards AI dominance, challenging traditional assumptions about what it takes to lead in AI development.
The extreme cost of training models in terms of energy usage, processor requirements, and massive data centers is well-documented. But the newest models might be able to do more with less.
We spoke with Manil Vasantha, Independent Consultant and CEO/COO of Custx.ai and tenured VP at a large legacy cloud software provider, to understand the global perspective behind DeepSeek and how it’s shapeshifting the global leaders of AI:
On China: Vasantha set the scene for the constraints China faced while developing DeepSeek:
“Putting China into perspective, they had their arms forced by giving them H800, which was completely stubbed compared to the H100s. They tore apart the three critical pieces: compute power, the RAM available, and the floating point notation,” Vasantha explained.
Despite these limitations, China approached the challenge strategically. “They considered the architectural limitations of what was given to them and they worked around it by getting down to the root of the problems and solving it with technology that was always available,” he noted.
China’s willingness to be resourceful in the face of great odds allowed them to create DeepSeek on an unparalleled small budget and computing power.
Putting China into perspective, they had their arms forced by giving them H800, which was completely stubbed compared to the H100s. They tore apart the three critical pieces: compute power, the RAM available, and the floating point notation.
U.S. response: “DeepSeek didn’t shock me. It’s just that why didn’t America think about it? That’s what shocked me,” Vasantha remarked. In his view, the abundance of resources in the U.S. has led to inefficiencies and complacency in AI development. “For US players, it’s basically going back to the drawing board,” he said. Vasantha pointed out that the vast budgets of tech giants like Meta and Microsoft, often in the tens of billions, may hinder innovation rather than help it.
“It would have been better for companies to constrain resources, test performance on the constrained resources, then see how well you perform because it will expose the challenges of your software,” he suggested. With this approach, he believes the US could drive more efficient advancements and compete with other global players.
Opportunity in India: While the US and China have dominated perspectives around DeepSeek, Vasantha sees India as an important part of the equation. He highlights India’s substantial financial resources and its ability to deploy large teams to develop new AI technologies at a lower cost compared to the U.S. Confident in its future leadership in AI, he added, “India is already working on their DeepSeek model hosted internally with the bells and whistles to protect their IP. There will be an explosion of computing.”
