The launch of R1, the newest AI model by Chinese AI firm DeepSeek, quickly captured global attention.
R1's launch, which temporarily erased nearly $1 trillion from US markets, has been hailed as a Sputnik moment for the United States, a triumph for Chinese AI firms, and revealed how much closer the global AI race is than previously thought.
But what's less discussed amid the hype around DeepSeek is the question: what does DeepSeek's launch mean for Australia and other middle powers in AI?
A good place to start is to first understand the technical fundamentals of DeepSeek.
Like other Chinese AI firms challenged by US export controls, DeepSeek has faced a shortage of the computational resources or compute (the semiconductor chips and software) needed for advanced AI systems.
To maximise their limited resources, the DeepSeek team effectively deployed various algorithmic and computational techniques to optimise the model's computational efficiency.
Many of these techniques, like the Mixture-of-Experts (MoE) architecture, are not new.
But DeepSeek managed to deploy these older techniques to create a more computationally efficient model, thus contributing to the model's shockingly low cost.
In the weeks ahead, it is likely that American laboratories will deploy many of the same techniques as DeepSeek to achieve similar efficiency gains.
What implications will it have for Australia?
This new focus on computational efficiency - and the DeepSeek moment more broadly - has several major implications for middle powers, the first of which is economic.
Many have argued that middle powers must find their place in the AI value chain through contributing some input key to the technology's development. For example, Australia is a major supplier of the critical minerals that are integral to the compute used to train and run AI models.
At first glance, some speculated that DeepSeek R1's computational efficiency gains would mean that future AI models would require less computing power, depressing demand for Australia's critical minerals and harming other middle powers in this supply chain.
But in reality, Jevon's paradox suggests the opposite may occur. Jevon's paradox argues that as a good gets cheaper, its use becomes more widespread, which raises demand.
When applied to AI, this paradox implies that as computational efficiency increases, AI use would become more widespread and demand for advanced chips would go up.
This would likely raise demand for Australia's critical minerals and benefit other middle powers in the computing supply chain.
Therefore, it's possible the DeepSeek moment may strengthen the economic relevance of some middle powers, rather than harm them.
Australia would benefit from further investing in capturing parts of the global computing supply chain.
Could Australia get caught in the middle?
The second key implication is geopolitical. With the DeepSeek moment, the AI race between the United States and China is more competitive than ever before.
Due to the affordable cost of DeepSeek's R1, it is possible that firms across the world, including in middle powers, will increasingly opt to use the model, in spite of concerns about censorship, privacy, and national security risks.
American firms will also be eager to launch their own low-cost offerings to counter DeepSeek.
Middle powers may thus soon find themselves caught in the middle, as American and Chinese firms increasingly compete for customers. The result may be that middle-power countries benefit from access to cheaper, more competitive models.
An opportunity awaits
The third, though likely most important implication, however, is entrepreneurial.
DeepSeek spun out of the Chinese hedge fund High-Flyer Quant, and famously mainly hires graduates from domestic Chinese universities.
This, combined with the fact that DeepSeek claimed to use fewer chips and spend much less money to train and develop R1 (even if these claims are not true in reality), have created a global perception that the barriers to entry for developing advanced AI models are lower than previously thought.
This perception has explicitly caused AI leaders in countries like India to suggest they build their own foundation models.
For middle powers like Australia, this leads to important questions. Even in the wake of DeepSeek, it is not likely that middle powers like Australia will produce the world's highest capability frontier foundation models.
However, many of AI's greatest opportunities lie in the application layer-building tools which sit on and use advanced AI models.
This, combined with DeepSeek's domestic talent pool, suggests that Australia and middle powers like it should, as I have outlined before, work to promote domestic entrepreneurship.
Australian universities produce tremendous technical graduates. As DeepSeek shows, these domestic graduates can, with the right support, produce successful homegrown firms, especially at the application layer.
Australian government policies, as well as those of other middle powers, should therefore work to promote this kind of domestic entrepreneurship. Doing so will enable these middle powers to capture a bigger stake in the global AI value chain.
The DeepSeek moment represents an important distance marker in the global AI race. Middle powers like Australia should be aware of both the benefits and risks of these moments, and work to seize the opportunities they create.
