January 31: DeepSeek, Sputnik, and the vagaries of technological arms races

In which, China’s latest and greatest AI model draws comparisons to the Space Race that may be more apt, but less alarming that most realize.

DeepSeek, Sputnik, and the vagaries of technological arms races

The Chinese built AI model DeepSeek exploded onto the AI scene this week indicating that China isn’t as far behind the US on AI as many believed. DeepSeek demonstrated sophistication not quite at the level of some of the more advanced US models, but close enough to send US markets into a tailspin as confidence in America’s tech superstars faltered.

Why would the emergence of a competitive AI model have that kind of a market effect? Well, it’s because DeepSeek isn’t just as good, or almost as good, as some US models, but that it is open source… and that it was purportedly made for less than $6 million USD!

It’s that last data point, in particular that freaked out the markets. The “tech bubble” talk has been ongoing, especially as industry leader OpenAI continues to hemorrhage money.

Could the American startup ethos be losing its innovation and economic edge? Is it really that cheap to produce an AI model? Such possibilities are what drew the allusions to DeepSeek being a “Sputnik moment” for the US. The allusion to the Soviet Union’s sudden jump to an early lead in the 1950s Space Race certainly captured the panicky alarmism of the first few days post DeepSeek’s unveiling. However, there are several elements of the Space Race that would render the Sputnik reference more favorable to the US. For example, America eventually won the Space Race and put a man on the moon first. In other words, the “Sputnik moment” infused America’s space program with a new purpose that drove major innovation that quickly erased whatever early lead the USSR had. Eventually, that basic fact that the Soviets could get flashy early leads in technology or economic races but couldn’t hold them would be the key indicator that the Soviet system as a whole couldn’t keep up with the US.

Back to China and DeepSeek. China is bigger than the Soviet Union ever was, but its economic model is still subservient to the command and control of the CCP much like the Soviet model, meaning that it cannot grow and innovate like the American economy. “But, Tim,” you say, “This is AI world, this is different.”

There’s still plenty about DeepSeek’s claims that remain unclear, its price tag being one of the biggest. Did it cost just $6 million? Sure, it could’ve, but that doesn’t mean that’s all it cost. In a Chinese economy where the CCP chooses industry winners and losers, it stands to reason that DeepSeek is heavily subsidized, so $6 million could just be referring to what its parent company put up. It may not refer to what the CCP has added. Additionally, training an AI system requires massive amounts of human input. I haven’t seen any evidence yet, but the CCP is not above deploying slave labor to keep costs down and advance a project on a fixed timetable. So, factor in slave labor and not reporting government subsidies (a common occurrence with Chinese companies) and yeah, you could probably get a $6 million AI, but no, you probably haven’t demonstrated superiority to the American models, especially if you’re piggybacking off of them as OpenAI seems to think.

Jury’s still out on this, but those are my initial thoughts/suspicions.