Launching on the day of Trump's inauguration, few anticipated the impact Deepseek would have. What has followed so far is not the robot apocalypse many have feared, but something perhaps more consequential: the opening salvo in what analysts are now calling the AI Cold War.
The AI war I pictured is somewhat differently from the reality. I pictured Arnie, robots, and lots of machines taking over the world and wiping us out. The reality (at the minute) is quite different. We have the 'traditional' LLM wars, with the big tech companies apendage measuring, with updates and now reasoning coming thick and fast every month.
Instead of killer robots, we face a geopolitical contest centered on DeepSeek, a relatively unknown Chinese AI company that reportedly trained for just $5.6 million. Funded by a former quant fund manager, it was purportedly launched for a fraction of what Western companies spend. Accusations have been flung by OpenAI, Andreessen has described it as a "Sputnik moment." And there you have it, the AI Cold War between the US and China.
DeepSeek has been banned or put on the watchlist left, right, and centre. Its successful launch wiped billions off tech stock markets almost overnight, creating real panic among Western investors and governments.
The international response has been decisive. Taiwan's Ministry of Digital Affairs banned government agencies, stating DeepSeek "endangers national information security." Australia banned all government devices, citing "unacceptable risk" to national security. Italy was the first European nation to implement a complete ban following GDPR investigations, while South Korea restricted government usage after security assessments. In the United States, multiple federal agencies including the Pentagon, Navy, and NASA implemented immediate bans, with states like Texas, New York, and Virginia enacting comprehensive bans on government devices.
European regulators have continued the pressure, with France, the Netherlands, Belgium, and Ireland launching investigations into DeepSeek's data practices, particularly regarding transfers of European citizens' data to China.
The fact that Deepseek exists is a problem for US dominance of AI. Floating the US export controls of advanced semiconductors. DeepSeek has apparently achieved comparable results to Western AI leaders while operating with significantly less computational power and investment, effectively getting around US attempts to contain Chinese AI development.
Security researchers have uncovered outdated encryption algorithms, hardcoded security keys, and data flows to entities with connections to Chinese state interests. Testing has revealed DeepSeek's safety mechanisms failed against more than half of standard jailbreak attempts, suggesting the model could be easily manipulated to bypass safety guardrails.
I originally pictured a world where AI in China would evolve very differently from the AI in the US. Now I'm not so sure. With both countries vying for global supremacy, is it just a case of the famous Thai saying "same same but different"? It's becoming like choosing between a Cisco system or a Huawei system – very similar in what they do, but with completely different implications for who has access to your data.
While Western nations focus on security concerns and regulatory compliance, a crucial factor remains underappreciated: cost. In this post pandemic world of cost-cutting, PRICE MATTERS. And whilst some European diehards/GDPR purists will resist, the REST OF THE WORLD WILL PICK THE CHEAP ONE. This economic reality means that Africa, South America, and the vast majority of Asia (POSSIBLY excluding India) will be running on Chinese AI.
This threatens to create an even more divided AI world with implications for global standards, governance, and security.
The core anxiety driving Western responses centres on DeepSeek's data handling policy. All user data is stored in China, where local laws require organisations to share data with intelligence officials upon request. DeepSeek's data collection reportedly encompasses an extensive range of user information: chat and search query history, device information, keystroke patterns, IP addresses, internet connection details, and even activity from other apps on users' devices.
The AI landscape is likely to fragment along geopolitical lines. This fragmentation points to a split into competing US and Chinese AI spheres of influence, creating difficult choices for the rest of the world.
This wasn't the robot uprising many feared, but perhaps something more consequential: the beginning of a profound technological divide that will reshape global power dynamics for decades to come. The AI war is here, but not the way I pictured it.