International

By Michael Roberts

Most readers will know the news by now. DeepSeek, a Chinese AI company, released an AI model called R1 that is comparable in ability to the best models from companies such as OpenAI, Anthropic and Meta, but was trained at a radically lower cost and using less than state-of-the art GPU chips. DeepSeek also made public enough of the details of the model that others can run it on their own computers without charge.

DeepSeek is a torpedo that has hit the Magnificent Seven US hi-tech companies below the water line. DeepSeek did not use the latest and best Nvidia’s chips and software; it did not require huge spending on training its AI model unlike its American rivals; and it offers just as many useful applications.

DeepSeek delivers a cheaper AI solution

DeepSeek built its R1 with Nvidia’s older, slower chips, which US sanctions had allowed to be exported to China. The US government and the tech titans thought they had a monopoly in AI development because of the huge costs involved in making better chips and AI models. But now DeepSeek’s R1 suggests that companies with less money can soon operate competitive AI models. R1 can be used on a shoestring budget and with much less computing power. Moreover, R1 is just as good as rivals at ‘inference’, the AI jargon for when users question the model and get answers. And it runs on servers for all sorts of companies so that they need not ‘rent’ at huge prices from the likes of OpenAI.

Most important, DeepSeek’s R1 is ‘open source’, namely that is coding and training methods are open to all to copy and develop. This is a real blow to the ‘proprietary’ secrets that OpenAI or Google’s Gemini lock away in a ‘black box’ in order to maximise profits. The analogy here is with branded and generic pharmaceuticals.

Capital expenditure by the big tech companies

The big issue for the US AI companies and their investors is that it appears that building huge data centres to house multiples of expensive chips may not be necessary in order to achieve sufficiently successful outcomes. Up to now, the US companies have been ratcheting up huge spending plans and trying to raise mega amounts of funding to do so. Indeed, on the very Monday that DeepSeek’s R1 hit the news, Meta announced another $65bn of investment, and only days earlier President Trump announced government subsidies of $500bn to the tech giants as part of the so-called Stargate project. Ironically, Meta chief executive Mark Zuckerberg said he was investing because “We want the US to set the global AI standard, not China.” Oh dear.

Now investors are concerned that this spending is unnecessary and, more to the point, that it will hit the profitability of the American companies if DeepSeek can deliver AI applications at a tenth of the cost. Five of the biggest technology stocks geared to AI — chipmaker Nvidia and so-called ‘hyperscalers’ Alphabet, Amazon, Microsoft and Meta Platforms — collectively shed almost $750bn of their stock market value in one day. And DeepSeek does threaten the profits of the data centre companies and the water and power operators which expect to benefit from the huge ‘scaling up’ by the Magnificent Seven. The US stock market boom is heavily concentrated in the ‘Magnificent Seven’.

So has DeepSeek punctured the massive stock market bubble in US tech stocks? Billionaire investor Ray Dalio thinks so. He told the Financial Times that “pricing has got to levels which are high at the same time as there’s an interest rate risk, and that combination could prick the bubble … Where we are in the cycle right now is very similar to where we were between 1998 or 1999,” Dalio said. “In other words, there’s a major new technology that certainly will change the world and be successful. But some people are confusing that with the investments being successful.”

But that may not be the case, at least not just yet. The AI chip company Nvidia’s stock price may have dived this week, but its ‘proprietary’ coding language, Cuda, is still the US industry standard. While its shares dropped nearly 17%, that only brings it back to the (very, very high) level of September.

Much will depend on other factors like the US Fed keeping interest rates high because of a reversal in the fall in inflation and on whether Trump proceeds big time with his tariff and immigration threats that will only fuel inflation.

The tech oligarchs sucking up to Trump

What must enrage the tech oligarchs sucking up to Trump is that US sanctions on Chinese companies and bans on chip exports have not stopped China making yet more advances in the tech and chip war with the US. China is managing to make technological leaps in AI despite export controls introduced by the Biden administration intended to deprive it of both the most powerful chips and the advanced tools needed to make them.

Chinese tech champion Huawei has emerged as Nvidia’s primary competitor in China for ‘inference’ chips. And it has been working with AI companies, including DeepSeek, to adapt models trained on Nvidia GPUs to run inference on its Ascend chips. “Huawei is getting better. They have an opening as the government is telling the big tech companies that they need to buy their chips and use them for inference,” said one semiconductor investor in Beijing.

This is a further demonstration that state-led planned investment into technology and tech skills by China works so much better than relying on huge private tech giants led by moguls. As Ray Dallo said: “In our system, by and large, we are moving to a more industrial-complex- type of policy in which there is going to be government-mandated and government-influenced activity, because it is so important…Capitalism alone — the profit motive alone — cannot win this battle.”

Computing power is being used up exponentially

Nevertheless, the AI titans are not yet the titanic. They are going ahead with ‘scaling up’ by ploughing yet more and more billions into data centres and more advanced chips. This eating up computer power exponentially.

And of course, there is no consideration of what mainstream economists politely like to call ‘externalities’. According to a report by Goldman Sachs, a ChatGPT query needs nearly 10 times as much electricity as a Google search query. Researcher Jesse Dodge did some back-of-the-napkin math on the amount of energy AI chatbots use. “One query to ChatGPT uses approximately as much electricity as could light one light bulb for about 20 minutes,” he says. “So, you can imagine with millions of people using something like that every day, that adds up to a really large amount of electricity.” More electricity consumption means more energy production and in particular more fossil-fuelled greenhouse gas emissions.

Google has the goal of reaching net-zero emissions by 2030. Since 2007, the company has said its company operations were carbon neutral because of the carbon offsets it buys to match its emissions. But, starting in 2023, Google wrote in its sustainability report that it was no longer “maintaining operational carbon neutrality.” The company says it’s still pushing for its net-zero goal in 2030. “Google’s real motivation here is to build the best AI systems that they can,” Dodge says. “And they’re willing to pour a ton of resources into that, including things like training AI systems on bigger and bigger data centers all the way up to supercomputers, which incurs a tremendous amount of electricity consumption and therefore CO2 emissions.”

Silicon valley tech companies taking control of water supply

Then there’s water. As the US faces droughts and wildfires, the AI companies are sucking up deep water to ‘cool’ their mega data centres to protect the chips. More than that, Silicon Valley companies are increasingly taking control of water supply infrastructure to meet their needs. Research suggests, for instance, that about 700,000 litres of water could have been used to cool the machines that trained ChatGPT-3 at Microsoft’s data facilities.

Training AI models consumes 6,000 times more energy than a European city. Furthermore, while minerals such as lithium and cobalt are most commonly associated with batteries in the motor sector, they are also crucial for the batteries used in datacentres. The extraction process often involves significant water usage and can lead to pollution, undermining water security.

Sam Altman, the previous non-profit hero of Open AI, but now out to maximise profits for Microsoft, argues that yes, unfortunately there are ‘trade-offs’ in the short term, but they’re necessary to reach so-called AGI; and AGI will then help us solve all these problems so the trade off of ‘externalities’ is worth it.

AGI – the holy grail

AGI? What’s this? Artificial generalised intelligence (AGI) is the holy grail of AI developers. It means that AI models would become ‘superintelligent’ way above human intelligence. When that is achieved, Altman promises, its AI won’t just be able to do a single worker’s job, it will be able to do all of their jobs: “AI can do the work of an organization.” This would be the ultimate in maximising profitability by doing away with workers in companies (even AI companies?) as AI machines take over operating, developing and marketing everything. This is the apocalyptic dream for capital (but a nightmare for labour: no job, no income).

That’s why Altman and the other AI moguls will not stop expanding their data centres and developing yet more advanced chips just because DeepSeek has undercut their current models. Research firm Rosenblatt forecast the response of the tech giants: “In general, we expect the bias to be on improved capability, sprinting faster towards artificial general intelligence, more than reduced spending.” Nothing must stop the objective of super-intelligent AI.

Some see the race to achieving AGI as a threat to humanity itself. Stuart Russell, professor of computer science at the University of California, Berkeley, said “Even the CEOs who are engaging in the race have stated that whoever wins has a significant probability of causing human extinction in the process, because we have no idea how to control systems more intelligent than ourselves,” he said. “In other words, the AGI race is a race towards the edge of a cliff.”

Human intelligence is different to machine intelligence

Maybe, but I continue to doubt that human ‘intelligence’ can be replaced by machine intelligence, mainly because they are different. Machines cannot think of potential and qualitative changes. New knowledge comes from such transformations (human), not from the extension of existing knowledge (machines). Only human intelligence is social and can see the potential for change, in particular social change, that leads to a better life for humanity and nature.

What DeepSeek’s emergence has shown is that AI can be developed to a level that can help humanity and its social needs. It’s free and open and available to the smallest user and developer. It has not been developed at a profit or to make a profit. As one commentator put it: “I want AI to do my laundry and dishes so that I can do art and writing, not for AI to do my art and writing so that I can do my laundry and dishes.” Managers are introducing AI to “make management problems easier at the cost of the stuff that many people don’t think AI should be used for, like creative work….. If AI is going to work, it needs to come from the bottom-up, or AI is going to be useless for the vast majority of people in the workplace”.

Rather than develop AI to make profits, reduce jobs and the livelihoods of humans, AI under common ownership and planning could reduce the hours of human labour for all and free humans from toil to concentrate on creative work that only human intelligence can deliver. Remember the ‘holy grail’ was a Victorian fiction and later a Dan Brown one as well.

From the blog of Michael Roberts. The original, with all charts and hyperlinks, can be found here.