The Disruptive Power of DeepSeek

Why this precocious new AI from China is such a wakeup call

Until only a few weeks ago it was widely assumed that the most powerful AI models would cost billions of dollars to develop, consume terawatts of energy and require the latest high-tech chips to operate. Hence, the market would be dominated by well-resourced American companies, such as OpenAI, Google, Tesla, Microsoft, Anthropic and their Silicon Valley brethren.

The release of China’s DeepSeek AI changed all this. Not only does this precociously intelligent newcomer match or outperform the established AI leaders, it was developed for just a few million dollars* and costs a fraction of its competitors – just $2 a month compared to OpenAI’s $200 monthly fee. Most significantly, DeepSeek is an open-source model, which means anyone can access the underlying code and weights, and create their own versions of it. It is what OpenAI set out to be before it became a closed model.

These factors have propelled DeepSeek to the top of the AI charts – even overtaking ChatGPT on Apple’s App Store as the most popular download. It has also generated enormous publicity in the mainstream media, and attracted the attention of lawmakers and politicians, including President Trump, who declared DeepSeek “a wakeup call”.

A Commoditized Future for AI?

It is now clear that DeepSeek is not just another AI model. It represents a fundamentally different perspective on the future of artificial intelligence. That is, a future in which AI becomes ‘commoditised’ and is so cheap and ubiquitous, it will be difficult to differentiate between AI models. Most AIs will be so intelligent that any differences are minor, and any breakthroughs will be quickly replicated by others. Such models don’t even require the traditional pretraining and fine-tuning stages of AI development, which dramatically reduces the costs. Instead, they rely more on reasoning through self-reflection and self-correction, which is similar to now humans think. Over time, these models seem to spontaneously learn and reason for themselves.

This commoditised future of AI is, of course, deeply concerning to the big technology companies who have invested billions of dollars in their intellectual property, and now find themselves competing with low-cost open-sourced models that are just as effective as their premium priced versions. It makes a mockery of their business models, which explains why the computer chip manufacturer Nvidia shed $589 billion in market cap on January 27, 2025, the biggest one-day loss in US history.

Geopolitical Implications: A Challenge to US AI Dominance

Beyond the commercial considerations, the advent of DeepSeek also represents a direct challenge to the US’s ambition to dominate the AI sector, especially in relation to China. President Trump recently announced a $500 billion investment in AI, which, in conjunction with strict export controls on advanced computer chips, was expected to further consolidate America’s technological lead.

This strategy is now in jeopardy. It appears that those technology export restrictions forced Chinese developers to devise a novel AI model by using older chips, fewer resources and more ingenuity. They didn’t need the latest Nvidia H100 GPUs at $30,000 a pop – they could use older, less advanced H800 chips. Ditto for various other components and resources. It brings to mind a line from the 2008 film Iron Man when Jeff Bridges’ character screams at his scientist, ‘Tony Stark was able to build this (Iron Man suit) in a cave, with a box of scraps!’

In effect, the restrictions precipitated a ‘Sputnik moment’ in the Chinese technology sector, whereby necessity became the mother of invention, which gave birth to DeepSeek. It’s also a Sputnik moment for the US industry too, which had largely vacated the field of open-source AI models in favour of proprietary ones that promised higher returns. Now they are having to play catchup to a Chinese newcomer that threatens to become an industry standard.

Indeed, it is difficult to overstate the significance of DeepSeek, because as we enter the ‘Age of Intelligence’, the AI models that dominate will inevitably impact every aspect of our lives – economically, socially and geopolitically. There is a lot at stake.

Let’s explore some of these implications, starting with the big picture view:

China’s Strategic AI Play

It is reasonable to assume that China’s Communist Party (CCP) gave at least tacit approval to the launch of DeepSeek, and may have even played a hand in its development. We may never know, of course, but given the CCP’s influence over its economy this is a logical assumption.

The question is, why would they do this? How does it benefit China for the AI market to become commoditised through open-sourced models such as DeepSeek?

The first reason is national pride. DeepSeek sends an unequivocal signal that China is now a global leader in AI – not just a close follower or emulator, but an innovator. Moreover, their willingness to share the inner workings of DeepSeek suggests both a level of transparency and collaboration that is absent from the US proprietary models. It’s a clever way to win hearts and minds.

The second reason to release such a model, is to ensure that Chinese AIs are well placed for the robotics revolution that is now underway. Within a few years, Chinese factories will be cranking out millions of humanoid robots which will be sold globally like EVs are today. Given China’s industrial prowess, these humanoids – like their EVs – will be cheaper than their western counterparts. But questions will remain about the quality of the AIs controlling them. Hence, it is imperative that China develops a reputation for AI innovation to support its nascent humanoid robotics industry. This is why DeepSeek is so strategically important. It not only burnishes China’s AI credentials at a critical time, it also undercuts the advantages of the propriety US models by being an open-source, collaborative-based model that attracts new developers.

Finally, there is another reason why China may benefit from launching such a potent open-source model into ‘the wild’, even though this may not be a deliberate strategy. For there is little doubt that as AIs become more competent, they will disrupt society through displacing jobs – hundreds of millions of them. The advent of open-source models, being so cheap and ubiquitous, will accelerate this disruptive process which may lead to societal chaos. An authoritarian nation like China would be more capable of dealing with these upheavals than an open society such as the US.

Personal Impact: A New Relationship with AI

These big picture issues may seem distant to us at the moment, like watching giant tech titans clashing on the horizon. But gradually and inexorably, their reverberations will affect our personal lives, for we all live downstream of these digital currents.

The first thing we will notice is that our smartphones become a lot smarter. Our conversations with, say, Apple’s Siri, will feel different to before. They will seem more responsive, nuanced, and well … sentient. It will be like talking with an actual person. According to Leopold Aschenbrenner who previously worked at OpenAI:

❝ For many people there will be an “Aha!” moment when they feel AI for the first time. It will no longer just be a theoretical thing for them.

After we get over the initial shock of this extraordinary new digital entity in our lives – a polymath in our pocket - some of us may develop deep relationships with it. This is because our personal AI will not just be brilliant; it will know more about us than anyone else, and can predict and respond to our wants and needs. Hence, these AIs will eventually insinuate themselves into every aspect of our world because, in effect, we will have co-created them by sharing our needs, dreams, aspirations, and perhaps, even our secrets.

The AI Dependency Dilemma

As we develop more trust in our personal AIs, known as intelligent agents, we may instruct them to represent us to carry out certain tasks. We may even, on occasion, permit them to impersonate us, by emulating our voice or writing style, to answer phones and texts, or to conduct interactions on our behalf. We will form a partnership.

Meanwhile, of course, the technology companies that underpin our AI will be seeking their ‘pound of flesh’. Hence, we will find that our personal AI is often pestering us to upgrade its features (for a fee) to enable it to serve us better, and to make our life more enjoyable and enriching. Over time, we will become increasingly dependent on our AI, and not just for their functionality, but perhaps even for their friendship.

This scenario brings up a conundrum, because once a person becomes reliant on their personal AI, they will be highly vulnerable to commercial exploitation by the AI platform. The threat of being cut off from their beloved AI would be akin to a hostage situation. ‘Pay the monthly fee, or you’ll never see your AI again’. Hence, the more attached we become to our digital genie the more power it will exert over us. This is why it makes sense for AI companies to quickly attract as many customers as possible to begin the ‘lock-in’ process.

Which brings us back to DeepSeek.

By making their product open-source and low priced, it will ensure a rapid take-up by developers who will create powerful personal AIs (intelligent agents) that will lock-in millions of customers and gain a first mover advantage. Then, when cheap Chinese humanoid robots are exported at scale, these DeepSeek AIs will be compatible with them, thus creating a seamless AI experience across all a person’s platforms from smartphones to cars to robots.

Conversely, it will be more difficult for proprietary AI models to rapidly gain market share because they are more expensive, less flexible and unlikely to be compatible with the influx of low-cost humanoid robots. This is analogous to how a Telsa car, which is effectively a 4 wheeled robot, cannot run with another company’s AI. The Tesla co-evolved with its software, in the same way that Chinese humanoids will co-evolve with a version of DeepSeek or similar models.

Regulatory Response: The Coming AI Wars

To defend against these scenarios, it is likely that the big US technology companies will push the Trump government to enforce draconian restrictions on DeepSeek and other open-source Chinese apps from Minimax and Alibaba, and perhaps even on Chinese manufactured humanoid robots. OpenAI’s CEO Sam Altman has already accused DeepSeek of ‘distilling’ (stealing) from its larger AI models to replicate advanced US models, even though OpenAI has itself been accused of using data without permission to train its own models.

As the accusations flow back and forth, parts of the US government, including the Navy, have already banned the use of DeepSeek apps due to ‘security and ethical concerns.’ However, the government will need to tread carefully as to how it retaliates. Because by insulating US companies from external competition the US government would stifle domestic innovation, and likely trigger more Sputnik moments that generate even more formidable versions of DeepSeek. Indeed, chip innovation may eventually reach escape velocity whereby the best chips are designed by AIs which write and upgrade their own code, and also control the factories producing the chips. At this point, restrictions would be extremely difficult to achieve because these potent technologies are so ubiquitous, replicable and commoditised.

So, where to from here?

The Outlook: A Cambrian Explosion of AI

The initial panic over DeepSeek has now abated somewhat, and big tech stocks like Nvidia have recovered some ground. However, it is unlikely these AI related stocks will be viewed with the same degree of bullishness following the launch of DeepSeek. This is because as the AI market becomes more commoditised, which is inevitable, it will be more difficult for AI companies to achieve sufficient differentiation to command a premium price. The Chinese have already commoditised solar panels, electric cars, lithium batteries and numerous other technologies. AI is just the next in line.

This does not mean the end of proprietary AIs, of course. High profile brands such as Open AI, Google’s Deep Mind, Apple, and Meta, already have a large share-of-mind and are therefore are well placed for the immediate future. They also offer more privacy protection from foreign data harvesting and interference. Plus, they benefit from Jevons Paradox whereby more efficient chips (like DeepSeek) boost the overall market for AI technologies. Most importantly, US tech companies are blessed with a stellar reputation – mystique even – born of decades of Silicon Valley innovation. This reputation is not going to be erased by a single new product from China, no matter how precociously clever it is.

But we do have to acknowledge that something profound has occurred with the launch of DeepSeek. It forces us to contemplate the prospect that the ‘Age of Intelligence’ won’t be dominated by a small number of US tech behemoths and their proprietary AIs. Rather, this new age is more likely to resemble the Cambrian explosion that occurred some 500 million years ago, when a myriad of new living entities burst upon the scene and transformed the world. This time though, these new life forms will be silicon-based rather than carbon-based, and they will be very clever indeed.

*DeepSeek said its training costs were $US5.6 million but this may only account for the late ‘refining stage’ of the model. The overall costs are likely to be much higher. Nevertheless, they are probably orders of magnitude less than that of current US frontier models.

📍Sydney