The launch of DeepSeek was one of the defining moments in the evolution of Artificial Intelligence. The AI startup from China unveiled a model costing over $6 million, which is a fraction of the billions being spent by the flag bearers of AI in the United States. 

Advertisment

For investors, it was a major reality check or rather, shock. AI is definitely the future of work and human interactions, and thus, it was a sure-shot bet to succeed. However, they failed to factor in the costs, which spiked to billions in the name of developing AI infrastructure. Nvidia became the most valuable AI company, with its high-processing chips like the H800, used to train AI models like OpenAI. Then comes DeepSeek, which revealed a training cost far less than anyone in the market. This led to a large market sell-off, with investors looking at AI beyond its scope. Nvidia suffered the biggest single-day loss for any company in history.

Cost efficiency is the final frontier in the evolution of any innovation

DeepSeek's latest claims affirm its position as a disruptor. The company revealed a theoretical cost-profit ratio of 545 per cent per day. The company attributes its success to AI model distillation, a technique that compresses large AI models into smaller, cost-effective versions. DeepSeek leveraged open-source models from Meta and Alibaba to build its technology, challenging Silicon Valley's AI leadership. While some argue that smaller models lose some complexity and intelligence, the technique is increasingly being seen as a game-changer.

Advertisment

Meanwhile, US tech firms remain undeterred, highlighting the need to develop strong AI infrastructure for a long-term advantage. The companies have outlined AI spending of over $320 billion in 2025, up from $230 billion in 2024. Amazon leads the pack with plans to invest over $100 billion, mainly in AI-driven cloud infrastructure. Microsoft and Meta have also allocated over $70 billion to develop AI data centres, with Meta calling 2025 a 'defining year for AI.'

Despite these record investments, cloud revenues have been weaker than expected due to supply constraints. Analysts predict these challenges will ease in the second half of 2025. However, for investors, understanding how much money is needed to develop and train AI is crucial for future bets.