Intro
On April 26 2026, OpenAI shut down its SORA video generation model. But wasn´t Sora supposed to take on Hollywood? Yes. When it was released in late September 2025, the ability of Sora 2 to generate short videos sparked significant concern among major movie studios and creators more generally. But now, only a few months later, OpenAI says it can no longer justify the computing expenses. As it turns out, generating silly videos of OpenAI CEO Sam Altman is computationally very demanding.
However, the shutting down of Sora is emblematic of a wider problem. From recently leaked financial data[1], it seems that OpenAI lost well over 30 billion USD in 2025, much more than the already large losses of 5 billion USD in 2024. To be fair, revenues have increased as well, but not by nearly enough to compensate. In other words, OpenAI´s trajectory is not sustainable.
Is this indicative for trends in the broader AI industry? And if so, what does it mean for the tech moguls` predictions of jobless futures and a white-collar massacre? Let´s have a look.
There is a bubble, but the technology will stay
In 2025, big tech firms spent approximately 400 billion USD on AI infrastructure[2], a figure set to rise significantly in 2026. Are these investments inflating a bubble? Yes. Like in the late 90s dotcom bubble, or the subprime mortgage crisis, the financial pyrotechnics are on full display. To give but one example, AI chip provider Nvidia is using its enormous stock market valuation to further boost demand for its chips, by financing its own customers, AI compute providers. In the words of Rogé Karma, “CoreWeave (a big AI compute provider – JN) is using Nvidia´s money to buy Nvidia´s chips and then renting them right back to Nvidia”.[3]
Apart from Nvidia, everyone is making losses, in the hope of future gains. But big tech firms like Microsoft, Amazon, and Google can use money from their profitable operations (e.g. selling software and online ads), to fuel their (generative) AI ambitions. Many other firms in this space; however, have unsustainable levels of debt, little revenue, and are massively dependent on big tech firms, either for the supply of chips (Nvidia) or for reaching customers (Microsoft, Google, Amazon). These firms may well collapse at the first sign of trouble. This could for instance be a disappointing IPO of either OpenAI or Anthrophic – both of which are rumoured to go to the stock market later this year and aim at a valuation of at least 1 trillion USD.
Yet the near-certainty that there is an AI bubble does not mean all the workers negatively affected by the technology – from voice actors and creative writers, to game developers and translators, can stop worrying. The dotcom bubble also launched Google, Amazon, and Facebook into dominant positions for the following quarter century. And generative AI does work: with the help of such tools, engineers code faster and customer agents handle more calls. In addition, hundreds of millions of users have experienced the tech and are used to it. Therefore, the bursting of the bubble may simply determine which firms will capture the value of AI.
What´s more, after a long period of sluggish economic growth, especially in the EU, politicians are very much invested in the success of AI. For instance, the European Commission announced the ambition to turn the EU into an “AI Continent”, with a tripling of data centre capacity in 5 to 7 years, and rapid adoption across all sectors, from media and culture to healthcare and the public sector more broadly (the “Apply AI Strategy”). So even if private capital may run up against limits, the public sector is coming in to provide additional investment and to boost demand.
The effects on labour are contingent on our institutions
Making predictions is of little value for AI. Unlike previous digital technologies, the impacts of generative AI are diffuse, and various sectors, especially white-collar, are likely to be affected. However, a useful heuristic is that if your job can be done fully from home via a computer, then the introduction of generative AI may significantly affect the work you do, up to and including full automation.
But this is far from automatic or inevitable. First, much depends on the price of AI relative to human labour. Until now, the use of AI has been subsidised, with a 200 USD monthly subscription to Claude or ChatGPT allowing you to burn through tokens of a value of up to 8,000 and 14,000 USD respectively[4]. That cannot go on forever though. Already, AI firms are shifting to token-based payment models, resulting in price increases for customers. That said, a bursting of the AI bubble could (temporarily) dampen the demand for compute, leading to lower prices.
Second, the impact of AI will not be determined chiefly by the technology itself, as many neoclassic economists still seem to think. The trajectory of AI will be determined by institutions – from employment, copyright, data protection and competition laws, via the degree of labour organising and organisation culture, to corporate governance, public procurement and taxation regimes. At the moment, there are significant battles being thought on different fronts, and although labour interests may lose out in the process, this is not a given. The 2023 Hollywood strike remains a good example of what labour organising can achieve to affect the use of AI.
Finally, and crucially, the battle over AI is also fought at the level of narrative and ideology: many tech founders and funders have a dim view of workplace democracy and labour organising. They present technology as some alien force beyond anyone´s control, while explicitly building AI to eliminate labour and more strictly control workers. If their authoritarian view prevails, then inequality between capital and labour is indeed set to further increase. Only if organised labour and policy makers can articulate and effectively advance a vision for pro-worker AI, can we expect reduced working hours, an increased labour share of the economy, and better working conditions.
Even though it is likely informed by tactical considerations, it is promising that both the CEOs of Anthropic and OpenAI have recently dropped their rhetoric of an impending “job apocalypse” and “white-collar massacre”, and started talking about how AI can augment human efforts instead.
Footnotes:
[1] Ed Zitron, https://www.wheresyoured.at/exclusive-openai-financials/, 15 June 2026.
[2] Meghan Bobrovsky, https://www.wsj.com/tech/ai/big-tech-is-spending-more-than-ever-on-ai-and-its-still-not-enough-f2398cfe, 30 October 2025.
[3] Rogé Karma, https://www.theatlantic.com/economy/2025/12/nvidia-ai-financing-deals/685197/, 10 December 2025.
[4] SemiAnalysis, https://x.com/SemiAnalysis_/status/2064815045767213400, 10 June 2026.