The tech news of the decade blew up on Monday January 27, 2025.
Nvidia, the darling stock of the AI revolution dropped six hundred billion (17%) in market capitalisation in one day. This is the biggest one day loss in stock market history. It sparked a selloff of other tech stocks, leading to a sector drop of 5.6%.
Has the bubble burst, or is it just the theories of Clayton Christianson writ large, again?
The spark was the recognition of the impact of the Chinese AI architecture represented by DeepSeek R1 by the technical wizards and stock analysts.
Surprisingly, DeepSeek released a research paper outlining their approach to AI training. This details an architecture that dramatically reduces cost and complexity of training LLM’s while delivering results at least as good as OpenAI and comparable models. It took a week or so for the described technology and results to be absorbed and understood, culminating in Mondays panicked sell-off.
Is this a bubble bursting or just a sensible reordering of expectations?
Two factors outside corporate malaise have dogged my innovative efforts over the years, both of which are in play here:
- The notion that innovation takes place in an environment of constraints. While history demonstrates the truth of this, the stories we tell ourselves celebrate what appears to be great innovation emerging as a result of chaos. In this case, the restrictions placed on China getting the existing technology created restrictions they have beaten.
- What I call the ‘Christianson effect’, better known as the Innovators Dilemma, after Harvard professor Clayton Christianson is proven accurate time after time, after time. Again, Christianson accurately saw that a high cost solution to a problem would eventually be replaced by a much lower cost solution to the same problem. DeepSeek is just another example of the power of his observation.
The US under the Biden administration for security reasons put export bans on Nvidia chips, chipmaking tools, and development software. These bans covered US allies in an effort to isolate China from the Intellectual capital as well as the means to bridge the technology gap that suddenly appeared. It would appear that rather than accepting the ban and going home, the Chinese reacted by using the ban as a motivator to rethink the engineering of the guts of AI systems, and come up with a solution that addressed the two hurdles facing current AI:
- The enormous amounts of data required to train the models.
- The huge drain on power required to process even modest requests to the models for a response.
Both it would seem, are gamechangers, as the cost reduction probable for AI platforms is enormous.
The real question for those who run businesses that use this technology, or are starting to use it more generally in our lives, which is all of us, is what comes next?
Here is what I think, assuming the initial hype is close to the mark, and not another chimera like the Theranos scam.
- The huge allocations of capital being made by the big US companies, Microsoft, Google, Amazon, and Meta, will be put on ice. Nvidia has hundreds of billions of dollars in orders from these giants that it cannot currently adequately fill. Some if not many will be quietly cancelled.
- More billions allocated to build the infrastructure to accommodate the models, big chunks of expensive land, and power sources will also be slowed down. For example, the project called the ‘Stargate project’ triumphantly announced last week by the president involving a 500 billion dollar investment by the government will become just another Trump press release consigned to the round file. The project as outlined is a JV with Oracle, Microsoft, Softbank, and others to build AI capability in the US. It represented an equity investment by the government in the commercial leveraging of emerging technology, a first. I also speculate that the proposal to fire up a mothballed nuclear reactor at 3 Mile Island by Microsoft will require a rethink, although it may have just been at best, a thought-bubble.
- The disruption created by the DeepSeek technology will redirect the tsunami of capital towards Chinese technology, until the next innovation iteration comes along. This will both geometrically accelerate the rate of adoption necessary by business if they want to keep up with competitors, and make the current security concerns surrounding Tik Tok look trivial by comparison.
- The disruption might ‘democratise’ the use of AI in the sense that it will be more widely available once the costs are dramatically reduced. Alternatively, it may mean that the existing ‘moat’ controlled by the current crop of AI platforms, all American, will be replaced by a Chinese moat.
- Regulating AI in some way has been a topic of frantic debate since OpenAI launched Chat. To observe that regulators have no idea would be accurate. Now, instead of regulators being caught with their pants around their ankles, it is apparent that their pants, if they own any, are secure in the wardrobe. In a regulatory and geopolitical sense, we are spinning out of control.
- The rate of development of systems that enable humans to expand the reach and depth of the intelligence we evolved to have will be extended at a rate that is further accelerated by the huge reduction in cost that appears probable as a result of this Chinese breakthrough. We had better all start learning Mandarin.
As the old Chinese saying goes ‘We live in interesting times’
