Nvidia is worth more than every company listed on the Indian stock market combined. But on a single day in January 2025, it lost roughly $600 billion, marking the biggest single-day wipeout in US stock market history. The driving factor behind this loss was that a Chinese lab had done for $6 million what top US labs had spent hundreds of millions attempting.Throughout history, scarcity has been the catalyst for innovation, and that is what DeepSeek-R1 [arXiv] demonstrates: building smarter, not bigger. The US had blocked the sale of NVIDIA’s most powerful AI chips, such as the H100 and B200, to Chinese companies. This restriction is the exact reason why DeepSeek had limited compute power and had to train their massive models on the less capable H800 graphics processing units (GPUs).Rather than accept those constraints, the DeepSeek lab refined both architecture and infrastructure in tandem to train massive models far more efficiently, while keeping in mind their hardware realities. They proved they could bypass these hardware restrictions and train world-class models on older chips at a fraction of the cost.The immediate impact was panic. Stocks for Nvidia, ASML, Google and Microsoft all plummeted the week DeepSeek launched as investors feared a massive shift. If DeepSeek could achieve frontier performance with such lean resources, the multi-billion-dollar demand for massive GPU clusters might simply evaporate.Geopolitically, it shattered the illusion that the US was at the forefront of AI research. This forced the existing giants to pivot away from simply throwing more compute at models and hoping for better results. Instead, they began adopting an efficiency-first framework. China has since shut Nvidia out of its market, citing security concerns. They want AI sovereignty, with a homegrown ecosystem using Huawei chips.Sixteen months on, Nvidia, as well as other companies focusing heavily on AI are doing better than ever. An ever-increasing demand for compute obliterated the investors’ apprehensions. This phenomenon predates modern technology, stretching back to the steam engine.The English economist William Stanley Jevons in his 1865 book The Coal Question, wrote: “It is a confusion of ideas to suppose that the economical use of fuel is equivalent to diminished consumption. The very contrary is the truth.” He argued that more efficient technology would accelerate the depletion of England’s coal reserves, not slow it.AI compute is the latest example of this. The demand for it has skyrocketed, when intuitively, efficiency should have made massive compute obsolete.DeepSeek-R1 was also the first open-source reasoning model. For closed-source, proprietary models, the benefits primarily flow to the model creator and the cloud provider running those workloads. OpenAI and Anthropic are the clearest examples of this. Open-source models, however, benefit startups, individual developers, enterprises and entire ecosystems of service providers.The US strategy to maintain the lead in the AI race relied on the assumption that gatekeeping compute would render China unable to catch up. As Liang Wenfeng, DeepSeek’s founder, told 36kr: “We didn’t intentionally become a [disruptive force], we accidentally became one.”DeepSeek overthrew this assumption. What followed was a commoditisation shock. These models were now cheaper and more accessible than ever, drawing far more users. The policy designed to kneecap China may have handed the world cheaper AI.The US set out to win an AI race by controlling the hardware. Instead, it may have accelerated the moment when hardware stopped mattering. But what does it even mean to “win”?Having the best model? No, because every advance gets reproduced by competitors almost immediately. DeepSeek proved that a $500 million “lead” could be replicated in weeks. Market valuation? Also no. The market reflects investor sentiments, not the technological realities. As investor Ray Dalio says in The All-In Podcast, betting on the company and buying its stock are not the same thing.Being the first to build artificial general intelligence? Maybe – at least that is what the tech giants seem to be focusing on. But how durable is the first mover advantage? According to an article in the Harvard Business Review, “a rapid pace of market evolution makes long-term dominance unlikely”. The authors suggest a ‘quick in, quick out’ approach for companies in spaces where both the market and the technology are developing rapidly. AI fits that description exactly.Nvidia’s CEO Jensen Huang contradicted himself recently after giving a Financial Times interview. Although he told the newspaper that “China is going to win the AI race”, he quickly issued a statement afterwards saying “As I have long said, China is nanoseconds behind America in AI. It’s vital that America wins by racing ahead and winning developers worldwide.” The CEO of the company that’s benefitting the most from the hype can’t give a straight answer on who is winning.The Huangs and the Altmans built the narrative of “we can’t fall behind”, and that “compute is everything”. DeepSeek broke the story down. Despite that, money kept flowing. The narrative collapsed, but the valuation didn’t.Compute and efficient algorithms are power. Right now, access is getting cheaper by the month. The only question is whether the people who built the wall will find a way to rebuild it. The best weapon against it is the one DeepSeek already handed everyone: open source. The next DeepSeek moment won’t come from a well-funded lab in San Francisco. It will come from wherever the next constraint is tightest.There will be more shockwaves.Avisha Mathur is an electrical engineering undergraduate at IIT Madras.