Seemingly out of nowhere, a new global superpower has emerged in the ongoing battle for GenAI supremacy. China’s DeepSeek created a significant disruption last week (20 January) with the release of its state-of-the-art DeepSeek R1 family of models. These models have skyrocketed to the top of global comparison charts, forcing every US-based research lab and LLM provider to sit up and take notice.
The DeepSeek R1 model is undeniably impressive. Our testing confirms that it holds its own against the best offerings from OpenAI, including the much-vaunted o1 reasoning model, as well as top-tier models from Anthropic, Google, Meta, and others. To have surfaced seemingly from nowhere with such advanced capabilities is a remarkable achievement in itself.
Go deeper with GlobalData
Yet, what makes DeepSeek’s feat even more extraordinary are two other standout characteristics: open-source commitment and low-cost innovation. The announcement included open weights for the full model, a move that mirrors Meta’s approach and starkly contrasts with the proprietary nature of other major US players DeepSeek reportedly trained its model on a remarkably small budget using older GPUs, necessitated by US export restrictions on advanced chips.
While debates about the actual cost (factoring in infrastructure, salaries, and other factors) are ongoing, the optics of achieving on-par capabilities with leading US labs at a fraction of the cost have caused a stir.
US companies have been understandably rattled, as this release challenges their current strategy of pouring billions into training increasingly complex models. For example, OpenAI’s recent Stargate announcement – a staggering $500bn investment plan – is predicated on ever-escalating costs to develop new capabilities.
DeepSeek’s success, however, suggests that similar results can be achieved far more economically. What’s also truly groundbreaking are the technical innovations underpinning DeepSeek R1. The accompanying paper, which has been widely read in Silicon Valley, describes a novel reinforcement learning-centric training pipeline that enables emergent reasoning capabilities. This breakthrough eliminates the need for exorbitant hardware budgets and has positioned DeepSeek as a serious contender in the GenAI arena.
How well do you really know your competitors?
Access the most comprehensive Company Profiles on the market, powered by GlobalData. Save hours of research. Gain competitive edge.
Thank you!
Your download email will arrive shortly
Not ready to buy yet? Download a free sample
We are confident about the unique quality of our Company Profiles. However, we want you to make the most beneficial decision for your business, so we offer a free sample that you can download by submitting the below form
By GlobalDataThis development also has broader implications. China has now placed itself in a strong position to challenge US GenAI dominance, leveraging a lean, cost-efficient approach that renders the massive investments of US labs potentially excessive. For US companies, this marks the beginning of a new phase of competition. On one side are the US-based tech giants that pioneered GenAI and have invested hundreds of billions of dollars into its advancement. On the other is a rising global competitor that has redefined what’s possible with a modest budget and older technology.
In response, we expect the US players will take decisive action quickly. New, more advanced models such as OpenAI’s o3, Google’s Gemini 2.0, and Meta’s Llama 4 are on the horizon. However, these models have been developed using the traditional, high-cost paradigm. Unless they offer substantial improvements over DeepSeek R1, their reception could be lukewarm. It is also likely that US companies will study DeepSeek’s training innovations and adapt them to enhance their own pipelines, leveraging their economies of scale and the ability to serve millions of clients globally—areas where DeepSeek currently is not a major player.
For now, DeepSeek has established itself as a leading force in the GenAI space and the first non-American company to compete at this level. The landscape has shifted dramatically, and this is only the beginning of a fascinating new chapter in GenAI competition. Interesting times, indeed.
