DeepSeek R1 — Silicon Valley’s Energy Crisis Moment

DeepSeek R1 isn’t anyone’s Sputnik moment. It is a straightforward example of a constraint-driven start-up quietly building a product until they’re ready to enter a red ocean market. A strategy you’ll find in any number of start-up manuals, and boot camps.

Now that some of the hot air has dissipated, it’s clear that DeepSeek R1 is unlikely to galvanise Big Tech’s to change it’s approach to AI. The reaction of the incumbents has been largely limited to crying foul. Ironic given how much those same incumbents spend on lobbying to keep the market unfettered by regulations that might interfere with their ability to “innovate”.

Instead of a Sputnik moment, this will turnout to be Silicon Valley’s version of the 1970s energy crises. Triggered first by the 1973 Arab-Israeli War, and then by the 1979 Iranian revolution, the shortages of gasoline coupled with the huge prices increases (500%+ between 1973 and 1980) led to businesses and consumers being much more conscious of fuel efficiency. The crises were turning points that shook up the insular Detroit auto manufacturers, and led to US car buyers being ever more open to smaller, more efficient models from German and Japanese manufacturers.

Is DeepSeek R1 better than the Google or OpenAI alternatives?

No. Based on my experiments and the research I’ve read, it’s biases are different, but no more or less pronounced. It writes English in a hypey voice that I don’t like, but it seems no more or less prone to making up “facts” than the Google and OpenAI equivalents. It’s attempts at derivative fiction are more limited, at least in English.

Is DeepSeek R1 based on stolen intellectual property?

Quite possibly, it’s a practice that’s become so widespread in the global tech sector that it’s become part of the folklore. It’s also part of the business model of companies like OpenAI. Just look at the queue of publishers suing for theft of copyright material.

Is DeepSeek R1 a risk to my personal data?

Probably. Chinese companies are no more interested in personal data sovereignty than their American counterparts are: witness JD Vance’s comments in Paris this week. I don’t feel my data is safe with any of them. I might feel safer if my medical data were in the hands of a European headquartered company, but only slightly.

Is DeepSeek R1 a security risk?

Probably. DeepSeek is a Chinese company, and therefore subject to Chinese law. That said, given the terms of US Foreign Intelligence Surveillance Act (FISA) trusting American companies feels foolish especially in the current political climate.

Conclusion

DeepSeek R1 offers a glimpse of an alternative AI future that’s cheaper, less power hungry, and less polluting than the energy hungry AI future Silicon Valley is so keen on selling.

DeepSeek R1’s results are not better that the Silicon Valley alternatives, but they are probably “good enough” for a lot of applications — hence all the fuss.

If its proven that their approach delivers “good enough” results without the need for expensive Nvidia chips, and without need for data centres that are just as polluting as oil petrochemical plants. Then their approach, but not necessarily DataSeek’s products, should be what we base the AI future on.