You can call it a Sputnik moment, when an unexpected Soviet breakthrough in the space race in 1957 spurred the Americans to put a man on the moon.
Or perhaps it’s another China shock, a reference to China’s entry to the global market that made goods cheap and laid waste to the manufacturing sectors in many other countries.
Either way, the earthquake that a little-known Chinese AI startup, DeepSeek, has caused in the past few days is a collective wake-up call for those who thought they were leading in AI. In many ways, we’ve seen this movie before.
What DeepSeek has done is essentially rip up the script for AI. Instead of pumping more computing hardware (in the form of Nvidia graphics processing units) to train massive AI models, DeepSeek basically just trains on smaller models to get enough accuracy to do the job.
This means far less computing hardware needed, something that the United States-China trade war has focused on. Yet, the DeepSeek AI is said to match the best of OpenAI and Meta by simply being more efficient.
While some of its claims are yet to be verified, the importance is that they can be verified, since DeepSeek AI has opened up its recipe book, unlike its American rivals.
The other implication for this is that more businesses can possibly take advantage of the most advanced AI in the near future. Developers can take DeepSeek’s open-source model to create new apps, without relying on AI that is being tightly held by a few Big Tech firms in Silicon Valley.
Already, Wall Street is experiencing the aftershocks. Fickle investors have dumped technology stocks in a massive sellout that wiped US$600 billion off Nvidia’s market value.
You can bet the engineers in OpenAI, Meta and Google are feverishly looking at the open-source DeepSeek engine and seeing what they can learn from it.
And that is a good thing. The result may be a big breakthrough in AI since the doors are now open to a possibly more efficient way of doing things.
Once again, this is a reminder of how technology has evolved in the past. Seldom has it taken the trajectory that is expected of it – instead, some of today’s most important innovations have developed in moments that were like life-altering revolving doors.
Think of the first PCs as many know them today. In 1980, a small startup in Seattle called Microsoft was tasked to create an operating system for computing giant IBM.
To do so, Microsoft had to buy a small developer and its little-known software called 86-DOS (DOS was for disk operating system). This was renamed MS-DOS, yes, after Microsoft, and used on IBM PCs.
Bill Gates and company were smart to not tie themselves to IBM. This meant Microsoft could sell its software to other PC makers that were just springing up from garages and other small offices in the still-nascent Silicon Valley. Think of HP, for example.
While IBM PCs were the leaders in the early 1980s, they would be eclipsed in just a few years by many other PC makers that soon flooded the market. Today, IBM no longer makes PCs.
Perhaps a closer analogy to today’s AI development is the dot.com era. During the gold rush of the late 1990s and early 2000s, every startup that wanted to sell books or CDs or do anything online looked to a server made by Sun Microsystems.
These machines were expensive but they were fast enough to handle the traffic coming to your website. Much investor money was thrown into buying these Sun servers, so much that Sun declared itself “the dot in .com”.
Are people rushing in and investing in Nvidia and its hardware the same way now? You know, the picks and shovels for today’s AI gold rush? The similarities are obvious, though this new chapter is still being written.
For Sun, unfortunately, the cloud came along after the dot.com bust. These new cloud offerings were built with many low-cost Intel machines that were joined to become a unified whole, offering great performance to businesses without the monolithic and costly servers sold by Sun.
With the cloud, businesses could also buy computing services at subscription rates, scaling up and down, without investing upfront in expensive hardware.
This spelt the end for Sun, which was sold to Oracle in 2009. Sun’s former campus in Menlo Park, California was also sold. In its place now is Meta.
