What comes after DeepSeek?

Is this the turning point where the AI model race "ends" and the era of open-source efficiency begins?

By Theodoros Evgeniou , Andrew Shipilov and Peter Zemsky

Published: Feb 18, 2025 10:45:00 AM IST

Updated: Feb 18, 2025 10:52:56 AM IST

DeepSeek and OpenAI’s latest model show that “thinking aloud” is becoming more central. Image: Shutterstock

Ever since DeepSeek sprung onto the scene, the hot takes have been flooding in. Misinformation, controversy, hype and panic are swirling – but maybe that was the desired intention.

DeepSeek has made significant technical advances, but did it really change the game? In a recent Tech Talk X webinar, we explored the broader implications and what comes next.

The AI model race is over – time for AI adoption at scale

Theodoros Evgeniou, Professor of Technology and Business at INSEAD

OpenAI’s closed-model strategy, which set the industry standard, has been – at least for now – hammered by open-source. The DeepSeek story shows that smaller players can compete, and this is just the beginning. We should expect an explosion of innovation from around the globe, and it’s likely that DeepSeek itself won’t remain the leading open-source model for long.

This shift also spotlights the growing importance of inference – how AI processes information in real-time – rather than just the compute power used in training. While true reasoning may still be beyond current AI capabilities, both DeepSeek and OpenAI’s latest model show that “thinking aloud” is becoming more central. In addition, the cost of using AI is also growing with its “thinking time”. Compute needs are moving mostly from AI training to AI usage.

The day DeepSeek launched may indeed be remembered as the moment the corporate AI model race effectively ended. While open-source and smaller players will continue to compete, we are now at the start of the AI adoption race. Open-source models, which are both cost-effective and energy-efficient, will drive widespread AI adoption and productisation. With that, compute and energy needs will explode, and energy will be squarely part of the AI technology stack.

This creates an opportunity for Europe, and other latecomers who fell behind in the AI model race. Now, if they get AI adoption right, they have a fighting chance to catch up. Talent is key.

This is a hype game

Peter Zemsky, Professor of Strategy at INSEAD

The DeepSeek development could shake up the AI industry, changing both how value is created and captured. Lower costs, especially in training, could grow the market, but who will reap the rewards? In AI, having access to huge numbers of Nvidia processors has, until now, been the key advantage. If DeepSeek truly reduces the need for all that expensive hardware, it could upend the current AI power structure and make Big Tech scramble.

Also read: DeepSeek: Chinese AI firm making waves in Silicon Valley

Still, it’s worth being skeptical. We don’t know exactly how much DeepSeek still depends on Nvidia chips, and hype plays a big role in AI. Tech companies benefit from making bold claims because rapid adoption helps them gather valuable data, as OpenAI has demonstrated. DeepSeek and similar players have strong reasons to emphasise their independence from traditional hardware, but the reality may be more complicated. Meanwhile, Big Tech firms will likely double down on research to keep AI advancements tied to hardware, where they have an edge.

As long as AI needs powerful chips, Nvidia holds pole position. But if the focus shifts from training (where Nvidia leads) to inference (less certain), it will be interesting to see if new players come the front.

Data is the new oil

Andrew Shipilov, Professor of Strategy at INSEAD

Disruption is inevitable. Lower-cost, more efficient competitors will always emerge. We’ve seen this in the automotive and semiconductor industries, and the same holds for AI. The DeepSeek example signals a shift towards efficiency, something we have not seen much of before. As new models crop up, investors and stakeholders will increasingly focus on how efficiently they operate and look for clear ways to measure and compare such efficiency.

DeepSeek is becoming a core player in China’s GenAI (generative AI) ecosystem. It will be interesting to see which firms it partners with to orchestrate its own ecosystem, and which will act as complementors. It has multiple pathways for monetisation, for example, by providing low-cost API (application programming interface) services and customised AI solutions with lightweight, distilled models. Given that it is funded by a hedge fund, DeepSeek can continue receiving financial support without immediate pressure to find new sources of revenue.

It seems that AI’s next competitive edge lies in customisation, such as training models for specific tasks, integrating AI into workflows and leveraging proprietary data. Indeed, data is the new oil – a clear competitive advantage.

Success will also depend on seamless integration and user experience – making AI easily accessible, almost “on demand”. AI systems will have a competitive edge if they can efficiently access specialised knowledge – much like DeepSeek, which retrieves insights from multiple sources rather than storing everything in a single model.

In other words, being intelligent doesn't mean knowing everything about everything. It means knowing who to ask. Human societies work similarly: Not everyone's an expert, but by knowing where to find help, we can still learn and grow.

[This article is republished courtesy of INSEAD Knowledge, the portal to the latest business insights and views of The Business School of the World. Copyright INSEAD 2024]