AI panic costs more than $1 trillion, raises new questions about safety
On Monday, an unfathomable amount of wealth vanished as AI-related stocks were sold off in a panic, after a Chinese start-up said it can do AI at lower cost, with fewer chips and far less energy.
On Monday, January 27, 2025, NVIDIA stock lost 17.3% of its value, estimated at nearly $600 billion. (At one point, NVIDIA had lost nearly $1 trillion.)
For reference, only 23 countries have annual economic output (all transactions across the entire economy) greater than $600 billion. That leaves around 200 nations that don't exchange as much value across their entire economies, over a 365-day period, as NVIDIA lost in one day.
Among countries with economies smaller than NVIDIA's one-day loss are Sweden, Norway, Belgium, Ireland, the UAE, and Nigeria—a nation of 200 million people.
It wasn’t just NVIDIA that lost big in the DeepSeek panic. Major AI-linked and tech companies, including chip manufacturers, also lost more than 10% or even 15% of their value. Total losses across those affected by the crash could be more than $1 trillion.
There is a lot that is wrong with what happened with tech stocks on Monday, January 27. First, it appears the massive selloff was the result of a claim by a Chinese company to have created a generative AI model that performs similar tasks to leading models, though not as well. The reason for the selloff was that "DeepSeek" appears to have achieved functional generative AI with inferior chips, made more cheaply, and running fewer calculations—meaning less energy consumption.
1) Does this make any sense?
In principle, those companies held the extreme value they did, because share holders were invested in their success. So, if the generative AI systems that have spurred such global technological and economic disruption—and accumulated trillions of dollars for a handful of companies—were so revolutionary, why would their bright future be so destabilized by a single announcement which most people and companies, and nearly all of those selling stocks, have not tested?
The obvious answer, which only raises more concerns, is that this is a bubble, where rank speculation has driven unprecedented sums of wealth into a handful of companies, with little real security about any reliable future value. This is most likely not the first time you have read that companies valued at more than $3 trillion might not be able to sustain and deliver that value over time. What is stunning is that the veil could be pulled back so suddenly from such an avid and committed stock market.
2) How can so much value disappear so quickly?
Second, there is the problem of any entity being able to lose that amount of money in a single day. That this is possible, functionally, is a demonstration of the implicit bargain that allows for rank speculation to drive valuations. This is problematic not only for the integrity of the trading of stocks and other equity investments. It is also problematic, because that unfounded speculative value distorts the entire economy, including what financial support is available to which entities.
If financial institutions' calculations about real-world value are distorted by unfounded speculative stock prices, with no boundaries, then smaller companies that reliably create actual, measurable, everyday nonfinancial value for customers and communities, might find themselves less able to access both investment capital and credit, as financing shifts to these top-heavy giants and their professed ways of creating value.
3) What is lost when markets are so distorted?
Third, we must ask what is lost if valued services cannot compete with entities that profit from unrealizable speculative pricing:
Will local newspapers have an easier time staying in business if generative AI systems can provide artificially "curated" news feeds with a mix of local, national, and global news, at little to no cost?
One question is whether consumers prefer one or the other; another is whether one side of that equation is unfairly advantaged by wildly inflated valuations that cannot be realized and whether that distortion subjects consumers to a lower quality of reporting and fact-finding.
Ill-advised acts of blind speculation are shaping value considerations across the whole economy, while potentially degrading everyone’s access to reliable, relevant, reviewable information.
4) Will investors now favor cheap, lower quality AI systems, owned by authoritarian regimes?
Fourth: The record losses of Monday, January 27, appear to represent a vote in favor of cheaper, lower-quality, less well-structured and less verifiable generative AI systems—even if that lower-quality system is owned and run by a totalitarian dictatorship. There are deep concerns about DeepSeek’s handling of user data, including allegedly consuming and storing user data, even if users decline or delete the app, and there are reports of a major leak of sensitive data.
Traders will say it is all about the trend. Many will expect NVIDIA and other tech companies’ stocks to rebound and recognize they were participating in a kind of rush to the bottom, hoping to get out before losing too much and maybe come back in at a profit. While that might work out for some, it will not change the fact that the market incentive is now shifted.
5) Whose money vanished?
Finally, there is the question of whose money was lost. Not all of this money was owned by day traders and wealthy investors. Tens of billions of dollars were likely shed from pension funds and other long-term portfolio investments held by people who don’t use the stock market to gamble on flashy tech announcements but to build value for their retirement or for their family’s future.
$600 billion is also equivalent to roughly 2.4% of annual US economic output. The wealth lost in the NVIDIA crash alone effectively erases last year’s economic expansion for the whole country. Hedge funds and investment banks will probably find a way to come out ahead, but that loss of wealth will constrain what kind of credit and financing ordinary people and small businesses can access.
Important questions going forward
A few more questions jump out of this week’s AI panic:
Can stock trades centered on unproven emerging technologies be more evidence-based and less speculative?
Do standards need to be adopted, without further delay, for how AI systems are developed, funded, and deployed?
If AI systems will condition decisions about health, environmental protection, and military action, who will pay for quality?
Can the risk of hallucination, fabrication, or information manipulation ever be eliminated from generative AI?
What do legislation and investment strategy look like that favor small, local enterprise over transnational AI giants?
How we answer these questions could have far-reaching implications for the future of advanced information technology, including but not limited to artificial intelligence, and for whether people will have access to facts, evidence, and the opportunity and empowerment benefits of good information and well-designed technology.
Related:
Multidimensional metrics & integrated data systems need to meet strict standards
The Good Food Finance Blueprint for Data Systems Integration includes important insights on standards for data security, confidentiality, intellectual property, and AI safeguards. The report notes:
“Beyond the risk of fabrication or distortion is the risk of excess deference to systems that do not actually make informed judgments, but produce words that state that they have. Surrendering decisions to such systems in the early stages of development can, even without hallucinations, lead to unintended negative outcomes, which might escape detection or fail to be address in a timely manner.”
The report also cites a number of efforts ongoing to shape standards for the responsible development and deployment of AI systems, including that they be “be human-centric, trustworthy and responsible”.