Chinese AI models come thundering in: One year after the DeepSeek shock

# Chinese AI models surge in: One year after the DeepSeek shock 15 February 2025 was a turning point for the AI market. That was when the relatively unknown Chinese company DeepSeek launched a model that in several areas challenged Western competitors, at a far lower cost and despite US export restrictions on advanced chips. One year later, analysts believe that DeepSeek was not an isolated case, but the start of a new wave of cheap Chinese AI models that could push down prices and shift the balance of power in the market. ## The DeepSeek shock: Why did it rattle the market? When DeepSeek broke through in February 2025, three factors in particular made investors and the technology industry react: - High performance that on many tasks was judged to be on a par with GPT-4 and Claude - Low costs for both training and operation - Development despite US export restrictions on advanced AI chips The reaction came quickly: - Nvidia's share price fell sharply - Western AI companies had to reconsider their pricing strategies - Investors began to question the entire narrative that Western companies had a lasting AI moat It was not just a piece of technology news. It was a signal that AI could become cheaper, more accessible and far harder to control geopolitically than many had imagined. ## One year later: DeepSeek was only the beginning What at first looked like a one-off shock now appears to be the start of a broader trend. Reuters reports that analysts expect a flurry of low-cost Chinese AI models when the Chinese New Year begins on 15 February 2026. In other words: the market may be facing a new wave of models that combine low prices with steadily improving quality. ### Why is the wave coming now? There are several explanations: 1. Annual cycle: Chinese technology companies often launch new products around the New Year. 2. Political timing: It may be smart to show strength ahead of any new US restrictions. 3. Competitive logic: When one company shows that it is possible, others will follow. For Western players, this means the pressure no longer comes from a single challenger, but from a growing ecosystem. ## The paradox: Restrictions on chips, but full speed in the market At the same time as the US has tried to limit China's access to advanced AI chips, the trend is pointing in another direction. Reuters reports that Taiwan has raised its growth forecast for 2026 to 7.7 percent, driven primarily by demand for AI chips. That tells us two things at once: - Demand for AI infrastructure is still extremely high - Chinese companies still appear to be finding ways to obtain the necessary hardware The big industrial winner in this picture is Taiwan, and in particular TSMC, which supplies chips to large parts of the world's AI industry. ## Why have the export restrictions not worked as planned? The US has introduced extensive restrictions on the export of advanced AI chips to China. Even so, Chinese companies have continued to develop competitive models. This is probably due to a combination of several factors. ### 1. Smarter training methods DeepSeek and other players appear to have found ways to get more out of less computing power, including through: - more efficient architectures - smarter use of data - new optimisation methods In practice, this means they do not necessarily need as much state-of-the-art hardware as previously assumed. ### 2. Workarounds and grey areas Chips can still find their way into China via: - third countries - front companies - older generations of hardware that are still powerful enough for many AI purposes It is a classic example of how difficult it is to enforce technology controls in global supply chains. ### 3. China's own chip development China is investing heavily in building up its own semiconductor industry. The country still lags behind players such as TSMC and Nvidia, but the gap may narrow over time. ## The pressure on Western AI companies is increasing The new competition is not just about technology. It is about economics. The New York Times reports that OpenAI hopes to triple its revenue in the coming years, while the company plans to spend tens of billions of dollars. The challenge is obvious: if competitors offer "good enough" at a far lower price, it becomes harder to defend high margins. ### OpenAI's problem: The clock is ticking If Chinese models continue to close in on Western competitors in quality, OpenAI and others may face: - stronger price pressure - weaker margins - greater doubt about how sustainable the business model is in the long run For customers, this could be good news. For the suppliers, it is a warning sign. ## Investors see the shift: Infrastructure may become more important than the models Another clear sign of a market shift is how investors are positioning themselves. Data centre giant Equinix rose 11 percent to an all-time high in mid-February. The logic behind the rise is simple: if models become cheaper and more standardised, more of the value can shift to the infrastructure needed to run them. This points to a new investment thesis: - Model development gradually becomes more commoditised - Data centres, networks and inference capacity retain pricing power - The "picks and shovels" strategy may beat those who hunt for gold directly It is a bit like during a gold rush: those who sell shovels and tents may end up earning more steadily than those who search for gold. ## What does this mean for Norwegian businesses? For Norwegian companies, the question is not just who wins the AI race globally. The most important thing is what this development means for their own costs, choice of suppliers and competitive advantage. ## 1. Should Norwegian companies consider Chinese AI models? The short answer is: maybe, but with great caution. ### Possible advantages - significantly lower costs - comparable performance in many use cases - less dependence on a few Western suppliers ### Clear risks - data sovereignty: Where are the data stored and processed? - GDPR and compliance: Are the suppliers compatible with European requirements? - geopolitical risk: What happens if the regulations are tightened? - security and trust: Can the business rely on the infrastructure? For a Norwegian company that uses AI to summarise public product descriptions or generate marketing drafts, low costs can be tempting. For health, finance or the public sector, the picture is entirely different. ## 2. What happens if AI models become a commodity? If models become cheap and available to everyone, the competition moves elsewhere. Then the value increasingly comes from: - your own data - industry knowledge - good user experiences - integration into workflows - the ability to deliver a concrete result, not just access to a model It also means it can be risky to build your entire strategy on owning "the best model" if the market price of model capacity keeps falling. ## Examples: How Norwegian companies may be affected ### MediVox AI: Medical record keeping For a company working on medical documentation, lower AI costs can in isolation be positive. It can make the technology cheaper to offer and easier to scale. But medical data is among the most sensitive data that exists. That is why the question is not just price, but also: - where the data is processed - how privacy is safeguarded - whether the supplier meets strict regulatory requirements In practice, this may mean that Chinese models could be used for development, testing or non-sensitive tasks, while production remains on Western or European infrastructure. ### Eir Tech: ADHD diagnostics For companies working with medical diagnostics, the requirements are even stricter. Here, Chinese models will probably be of little relevance in production, but cheaper compute globally may still contribute to faster research, prototyping and experimentation. ### Norwegian startups in general For start-ups, the picture is twofold: - Lower AI costs lower the threshold for building new products and MVPs - At the same time, it becomes easier for more competitors to do the same The winners are likely to be those who have something others cannot easily copy: their own data, strong distribution, industry understanding or regulatory advantages. ## Geopolitics: Three possible paths forward How this develops further depends not just on technology, but also on politics. ### Scenario 1: Open competition Chinese and Western models compete relatively freely in global markets. Prices fall, and innovation increases. ### Scenario 2: Two parallel AI ecosystems Western countries standardise on Western models for reasons of security and compliance, while China and partner countries build a parallel ecosystem. The result could be a kind of "AI cold war". ### Scenario 3: Chinese dominance in the volume segment Chinese models win in broad, price-sensitive markets, while Western companies concentrate on premium, enterprise and specialised solutions. ### What is most likely? The most realistic outcome is probably a hybrid. Europe and the US will likely prioritise local or Western models in critical infrastructure such as health, finance and defence. At the same time, Chinese models may take market share in less regulated and more price-sensitive segments. ## Advice for Norwegian businesses ### 1. Build a multi-model strategy Avoid locking your business to a single supplier. A more robust strategy may be to use different models for different purposes: - OpenAI for critical and complex tasks - Anthropic (Claude) for tasks where reasoning and compliance matter - DeepSeek or other Chinese models for high-volume tasks with little sensitive data ### 2. Focus on data and distribution, not just the model The real value often lies in: - domain-specific data - fine-tuning for a specific niche - workflow and integration - user experience - trust and regulatory compliance ### 3. Prepare your business for further price falls AI will probably become cheaper. Your business model should therefore be able to withstand technology itself becoming less of a differentiator. Build instead on: - value creation for the customer - network effects - proprietary data - regulatory or operational advantages ### 4. Follow the regulations closely The EU may in the future tighten its stance on Chinese AI services. Norwegian companies should therefore have a backup plan in case access, use or data processing becomes more regulated. ## Conclusion: AI is getting cheaper, and it changes everything One year after the DeepSeek shock, the picture is clearer: - AI models are becoming cheaper and more accessible - Chinese players have become real competitors - Western export restrictions are proving less effective than many hoped - Infrastructure and operations may become more valuable than the models themselves For Norwegian businesses, the message is simple: Use the cheap AI where it is responsible to do so, but build your competitive advantage somewhere else. AI is beginning to resemble electricity more than an exclusive product: essential, widely available and gradually cheaper. In that case, it is not necessarily the one who sells the electricity who wins, but the one who uses it best. --- Sources: - Reuters: "A year on from DeepSeek shock, get set for flurry of low-cost Chinese AI models" (Feb 12, 2026) - Reuters: "Taiwan hikes 2026 economic growth forecast to 7.7% on AI demand" - New York Times: "OpenAI's Biggest Challenge Is Turning Its A.I. Into a Cash Machine" (Feb 11, 2026) - Financial markets data: Equinix stock performance Published: February 2026 Category: Economy, Technology, Geopolitics