Author: Dr. Andrea Little Limbago, SVP, Applied AI, interos.ai
“DeepSeek R1 is AI’s Sputnik moment” claimed Marc Andreesen following their announcement of the open source reasoning model that rivals those made by Silicon Valley tech giants at a fraction of the compute.
Last week, the Chinese startup launched an open source AI assistant that by Monday had become the number one downloaded app on Apple.
Concerns over this breakthrough instigated a $1 Trillion loss in tech stock shares and have reignited the debate over closed versus open systems.
Whether or not this is a Sputnik moment, there are many prognostications surrounding this release. While it is too early to know the long-term impact of this latest shock to the AI world, it is important to both take this breakthrough seriously while also not jumping to the wrong conclusions.
The discourse is certainly going to evolve, but this is not the time for quick conclusions about strategic priorities that will shape the future of AI and geopolitics.
- Stargate is not necessary: Many are wondering whether DeepSeek will ‘deep-six’ Stargate, President Trump’s $500 B AI project aimed at developing AI infrastructure in the US. For years, technological bifurcation – the splintering of the physical infrastructure that serves as the backbone of the digital world – has created a divergent system between Chinese backed technologies and those produced by the US and like-minded countries. Governments are moving toward greater data sovereignty and tech sovereignty under the auspices of national security. Stargate is a natural progression of this movement. Given recent attacks on physical infrastructure, including on ViaSat as well as underseas cables, maintaining tech sovereignty over the infrastructure that powers AI is part of the broader global splintering of the internet and technological infrastructure into technospheres. If anything, DeepSeek’s announcement will only deepen this growing divide.
- Export controls don’t work: US export controls on the most sophisticated chips inadvertently sparked innovation by requiring Chinese companies to do more with less, or so the argument goes. Export controls will never be perfect, but that does not mean they may not be effective. The array of tech-related export controls have made it much harder for China to develop their own semiconductor industry, and perhaps is better compared to tech containment. Whether its enforcement challenges, or the notion that you can’t ban math – to pull from arguments over encryption bans in recent years – make AI technologies easier to circumvent than other emerging technologies.
- Security as an afterthought: With record-breaking downloads, yet again we are witnessing the flight to new tech with security as an afterthought. Nevertheless, DeepSeek had to halt new registrants due to ‘large-scale malicious attacks’. The security risks not only pertain to DeepSeek, but rather include the broad range of attacks, from data poisoning to model corruption. But wait, there’s more. China’s data policies enable the government to access data from companies located in China. Furthermore, given the tight connection between the government and companies, it is naïve to assume complete separation of the Chinese government from DeepSeek going forward, including manipulation and backdoors.
- Authoritarian regimes have the innovation edge in AI: While this has not been a prime take-away this week, there has been a growing debate about which regime type has the AI edge, largely based on the greater access to all kinds of data by authoritarian regimes. However, it ignores the censorship and propaganda that can poison the AI models. Existing Chinese GenAI models have already demonstrated censorship and disinformation, and initial research shows DeepSeek’s AI suffers the same problem. Garbage in, garbage out may give democracies the edge, including in reasoning models.
Over a decade ago, Edward Snowden’s revelations helped deepen the divide between Silicon Valley and DC. DeepSeek may finally be the impetus that brings the two group together. Is this the spark that rejuvenates the close allegiance between the tech sector and the US government, similar to Hewlett Packer, Texas Instruments, and IBM from the early days of the Cold War?
Will the AI-focused export controls passed earlier this month target DeepSeek before it becomes a TikTok regulatory problem? If so, how will China retaliate?
The only certainty is that we are still in the early days of this generation-defining technology. AI is more than the technology, but must be viewed through the regulatory, national security, and social systems lens in which it is deeply intertwined.
Go Deeper: What’s in Store for 2025
You can view our take on the 5 trends to look out for in 2025 in our latest report, including our breakdown on the need for Secure AI and the larger Bifurcation of technology along geopolitical fault lines: