The mass delusion of insatiable AI economics
I’m not an economist. I do not invest in the stock market. I find accountancy largely confounding. But the current hype surrounding AI, and the unhinged speculation it is inspiring among investors, has left me dumbfounded. The astounding numbers simply do not add up, even to an economic layman, and it is time we called that out.
The sheer unbridled craziness is exemplified by OpenAI, the makers of ChatGPT, whose recent spending spree is unlike anything we have ever seen. By the admission of its own CEO, Sam Altman, the Silicon Valley startup has committed to spend $1.4 trillion on data centre infrastructure by 2033. Yet with current annual revenues hovering around $20 billion, and no clear roadmap to profitability, the figures cannot be easily reconciled. (1)
Globally, McKinsey estimates that, by 2030, $6.7 trillion will need to be invested in data centres to meet surging demand for AI compute. (2) And with murky mixed messages on the profitability of products and services spawned by that architecture, a nonsensical thread runs through the entire AI ecosystem.
Fundamentally, the stark discrepancy between input and output rests on the insatiable costs of running AI products and platforms. To generate a 100-word email, ChatGPT-4 uses more than a 500ml bottle of water and enough electricity to power 14 LED lightbulbs for an hour. Multiply that example by 2.5 billion daily prompts, and every 24 hours, ChatGPT consumes enough electricity to run the Empire State Building for 540 days. Extrapolated over a year, that electricity consumption outstrips 117 countries. (3) (4)
OpenAI’s combined data centre commitments will, in theory, produce 17 gigawatts of capacity – equivalent to 17 nuclear power plants, nine Hoover Dams, or the total energy consumption of Switzerland and Portugal combined. (5) (6) Meanwhile, Nvidia, the main manufacturer of GPU microchips that enable AI processing, is on track to use up to 134 terawatt hours of energy each year by 2027 – up to five times the electricity consumption of Ireland. (3)
“The number just don’t make sense,” Derek Thompson wrote recently. “Tech companies are projected to spend about $400 billion this year on infrastructure to train and operate AI models. By nominal dollar sums, that is more than any group of firms has ever spent to do just about anything. The Apollo program allocated about $300 billion in inflation-adjusted dollars to get America to the moon between the early 1960s and the early 1970s. The AI buildout requires companies to collectively fund a new Apollo program, not every 10 years, but every 10 months.” (7)
The committed capital expenditure is mind-boggling, but the fact that it is underpinned by a chronic lack of profitability – and the mere fantasy of these products ever generating more than they cost – makes this moment even more psychotic.
OpenAI is not a public company, so analysis of its financial performance is beholden to projections and research rather than stringent self-reporting. Still, the profit-loss estimates are gnarly, regardless, with some reports predicting a $27 billion net annual loss in 2025. (8)
Altman contests those figures, but his own arithmetic is similarly dire. Internal documents show OpenAI plans to record major annual losses – as much as $74 billion – through 2028, then suddenly turn a profit by 2030. To do so, and remain solvent through those 2033 infrastructure spend commitments, the company would need to grow revenues by 6,900% over the next eight years – unless major cost reductions are found. (9) No company has ever recorded $1 trillion in annual revenue, but assuming expenditure remains consistent, Altman is convinced OpenAI can break that threshold in less than a decade from an underwater start. That smart people take him at his word, without minimal proof of concept, is absurd.
AI promises to effortlessly yield business efficiencies and turbocharge the revenue of adopting enterprises by automating repetitive tasks and reducing the costly workforce. Those promises have whipped speculative investors into a frenzy, inflating stock prices despite questionable evidence of long-term scalability. The irony that OpenAI is itself deeply unprofitable seems strangely inconsequential to most investors. To me, however, OpenAI selling financial freedom is like ExxonMobil selling environmental sustainability. The means are unmoored from the ends.
Take Sora 2, OpenAI’s latest video generation model, which is burning through $15 million per day to service user-generated AI slop in a TikTok-style feed. Remarkably, it costs OpenAI around $1.30 to generate a single 10-second Sora 2 video, yet the app is available for free. And with 11.3 million videos generated per day, the astoundingly bad business model strains credulity. (9)
Selling products or providing services at a loss to gain market share is cliché in Silicon Valley, of course. Tying people in at a low initial cost, then boosting prices later, is a well-worn strategy within tech startups. It took Twitter 12 years to turn a quarterly profit. Amazon required seven years and almost ran out of cash. Uber, meanwhile, spent 14 years in the red. So, by making access to its products free or cheap (in the case of $20 per month subscriptions), OpenAI is seeding dependence among users who will likely be hit with hiked prices and paywalled features down the road. It also has designs on proprietary devices and robotics, with ads and sponsored links a foregone conclusion. Meanwhile, AGI or super-intelligence lurks as a shimmering – if perhaps hallucinogenic – panacea. Still, none of those developments seem likely to catalyse the required step shift in scalable profit-generating AI adoption. Not anytime soon, at least, and perhaps not before investors run out of patience.
Failing to deliver on AI’s grand promise is not a problem unique to OpenAI, though. Reporting by Ed Zitron found that, as of August 2025, Microsoft had just 8 million active licensed users of Microsoft 365 Copilot – a conversion rate of just 1.81% from the subscriber base. Assuming – generously – that each licensee pays the maximum $30 per month subscription fee, that equates to $2.88 billion in annual revenue, a drop in the ocean of Microsoft’s $80 billion annual expenditure on AI. (11) (12)
Elsewhere, AI rollouts have failed spectacularly. Consider Taco Bell’s AI drive-thru attendant adding 18,000 cups of water to an order, or McDonalds’ equivalent adding 260 chicken nuggets to one meal while dropping bacon on ice cream in another. (13) (14) Sure, these are extreme, isolated examples, but they at least gesture towards a prevailing empirical truism: that AI feels a bit – well – crap, and that the guaranteed efficiencies are – well – hard to decipher.
To that end, in August 2025, MIT published a damning report that said 95% of ‘enterprise AI solutions’ – in other words, projects where companies try to integrate AI – fail. MIT interviewed 500 people and studied 300 AI pilots – only 5% of which reported a tangible return on investment. (15)
In fairness, much of that MIT report has been taken out of context by clickbait merchants, and the sample size is nowhere near large enough to draw definitive conclusions. Moreover, AI is useful, I contend. It can do in seconds amazing things that took hours or days just five years ago, such as generating passable marketing copy. This is not a polemic on AI, then, nor is it a dystopian projection of AGI aliens taking our jobs then killing our families. Rather, my point is simply about the absurd economics tethered to the blind and bewildering hype engulfing a homogenous – and wilfully fantastical – vision of what AI might become. It is a shared mass delusion, and I’m astonished by the entire global economy resting upon it.
To me, right now, AI is a groundbreaking tool whose best future may be as an expensive, boutique luxury with discreet use cases worth the spend to a narrow stratum of society and commerce. There is enough evidence to surmise that AI might just be an unprofitable technology at scale, but one that is very worthwhile in focused contexts – akin to 3D printers, say, or thermal imaging cameras and ultrasound scanners.
Drinking the Altman Kool-Aid, however, investors have gambled recklessly on AI becoming an omniscient, all-encompassing, paradigm-shifting force – a ubiquitous phenomenon godlike in its elusive yet enswathing mystique. The potential market for such a power is every person and every company on earth, producing limitless revenue and – eventually – profit. This modality views AI as a modern ancestor of electricity or water, and the opportunity to get in on the ground floor of such a gravy train cannot be missed. In fact, it is even worth burning copious resources to service Sora 2 slop and chatbot chunder as a necessary step on the road to financial nirvana.
Overwhelmingly, then, as opportunists flock to get involved, the stock market seems to be harbouring an AI bubble reminiscent of the dotcom crash of 2000. Back then, hyped internet startups devoid of profit attracted huge investments based on the delusional promise of future wins, only for skittish supporters to pull out when terrible business models produced no such riches. The ‘fake it ‘til you make it’ fever dream eventually popped, wiping out $5 trillion in market value. (16) (17)
The dotcom bubble began when startups and ambitious businesses tacked ‘dotcom’ onto the end of everything to sound innovative. Pets.com became the totemic exemplar, selling kibble for less than it cost to ship, but everyone played the game, and investors ate it up. When Active Apparel announced the launch of an ecommerce site amid the boom, for instance, its stock price soared more than 1,000% in two days – a microcosm of the mindless speculation. (18)
A quarter century on, the same thing is happening again. ‘AI driven’ is the corporate buzzword de jure, connoting cost savings, expedited processes, instantaneous results and – ultimately – the removal of expensive humans. As such, in the endless rush to follow the curve and virtue signal cutting edge trend adoption, companies are ploughing billions into AI pilots because – well – it is the next fashionable acronym. Whether the thing actually works seems secondary to sanctimonious preening.
Within the dotcom bubble, the aim of many startup founders was not to create a viable business model, but simply to stand up something buzz-worthy, generate hype using cash from seed investors, then IPO to cash-out. Whatever happened next was largely irrelevant – especially for the venture capitalists – because those who mattered got paid at the expense of the plebs who did not. Unsurprisingly, OpenAI plans to IPO within the next few years, and sales of primary shares will provide a cash injection, but floating on the stock exchange will not correct a defective business model unilaterally.
Besides, Wall Street sharks are already displaying jitters regarding AI, which has accounted for 80% of American stock market gains in 2025 – fuelled almost entirely by infrastructure spend rather than end-user enthusiasm. (19)
Michael Burry, famously depicted in The Big Short, recently closed his hedge fund, Scion Capital, because he believes the market is completely detached from financial fundamental. (20)
Peter Thiel, the Big Tech bellwether, dumped his entire stake in Nvidia in the third quarter, despite its valuation breaching $5 trillion. (21) (22)
Meanwhile, AI doyens have themselves broached the reality of a bubble, often saying the quiet part out loud.
“No company is going to be immune, including us,” says Google chief Sundar Pichai. (23)
“I’m deeply uncomfortable with these decisions being made by a few companies,” adds Dario Amodei, CEO of Anthropic. (24)
AI is in an ‘industrial bubble,’ concedes Amazon grandee Jeff Bezos. (25)
“Are we in a phase where investors as a whole are overexcited about AI?” asks Altman the Almighty. “My opinion is yes.” (26)
Too often, the commentariat gets bogged down asking if a bubble exists and whether we are currently within one, as if Bloomberg will declare it quantifiably with a flashing chyron. Such debates and prognostications veer towards abstract generalisation, while the truly startling aspect of this moment lies in the sheer absurdity of the economics – simple spend versus revenue – whether a bubble exists or not.
In my estimation, if consumers continue to be underwhelmed by AI products, seeing the grand promise as a far-flung mirage, revenue will stagnate and profit will remain a fantasy. Eventually, investors will wake up, smell the coffee, and ask serious questions. Many will cut their losses, causing stock prices to drop, market delistings to grow, and bankruptcies to mount. Devoid of the promised AI efficiencies, companies will retreat to more traditional methods when faced with existential threats, tipping the first dominoes in a wide-ranging crash.
Perhaps the most worrying aspect of such a burgeoning meltdown is the vast interconnectedness of the AI industry. You see, a cadre of tech giants – OpenAI, Nvidia, Meta, Microsoft, Google, Intel, Oracle, SoftBank, AMD, Anthropic – are entangled in complex circular webs of finance, investing in and purchasing resources from each other to the tune of trillions. (27)
“Nvidia invests in OpenAI, OpenAI rents compute from Oracle, Oracle buys Nvidia’s hardware, and Nvidia records huge sales,” Grace Blakeley wrote recently. “Everyone in the circle reports rising revenue, rising valuations, and surging share prices, but the cash is largely moving in circles.” (28)
Other parts of the circular AI economy include OpenAI buying chips from AMD; Microsoft and Nvidia backing OpenAI and Anthropic; Nvidia investing in CoreWeave and Intel; and Google and Amazon injecting cash into Anthropic. Confused? Well, maybe that is by design, to obfuscate and bamboozle as Monopoly money fires back and forth as ephemeral digits on a screen.
Diversifying income is Economics 101, yet the world’s most celebrated companies have largely centralised their fiscal flows. That, to me, creates stark areas of vulnerability. And if one link in the circular chain breaks – if one of the megalodons defaults on its make-believe commitments – the entire house of cards may come tumbling down. And whoever carries the bag when the music stops will be saddled with the debt of a cliquey industry.
Some say data centre deals can be easily renegotiated, rewritten or creatively mothballed, but such possibilities only add to the gross uncertainty and instability of this entire moment. There is a stultifying unreality to the AI funding ecosystem, and the prospect of merely writing off billions of dollars in commitments undermines excitement for their announcement in the first place. As consumers, why should we care about yet another deal where Nvidia ploughs yet more billions into yet more companies that will buy yet more of its chips? The whole thing feels dumb, as a few aloof guys redraw our future in their own wonky vision. People are tired of the glistening phantasmagoria.
Moreover, the most likely force to tip the house of cards is consistently overlooked. The depreciation of computing infrastructure – i.e., the way GPUs eventually burn out – seems to be a wilfully neglected flaw in the entire AI economy. History tells us that compute becomes cheaper as efficiencies and breakthroughs gird scalability. Therefore, if better GPUs will soon be available, as so readily promised, spending vast sums of cash on worse chips now makes little sense. Waiting can be more expensive than buying, in the sense of lost opportunity and vanishing market share amid a frantic race, and AI models need to be trained now to develop in future, but at what point are we crushed to death by the expanding graveyard of obsolete data centres? Maybe we will run out of resources – money, water, electricity, physical space – before the AI promise can ever be delivered.
Warning signs have already flashed. In January, when the Chinese-made DeepSeek chatbot flaunted similar outputs to ChatGPT despite being up-to-40-times cheaper, Nvidia’s share price dropped roughly 17%, wiping almost $600 billion off its market value in one day. (29) (30) If another DeepSeek Moment comes out of the blue, then, with China or another outsider unleashing cheaper, more capable GPUs, the entire circular apparatus could face massive redundancy overnight.
Just how bad could the damage be? Well, according to calculations by Gita Gopinath, former chief economist of the International Monetary Fund (IMF), $35 trillion in global wealth could be wiped out if the market crashes amid AI regret. (31) For context, that is more than the United States’ entire gross domestic product (GDP). (32)
Even if the bubble bursts, however, there will be winners, just as juggernauts rose from the dotcom rubble. While considering any company ‘too big to fail’ is naïve, four or five behemoths will probably find a way to survive, reshaping corporate structures here and smudging ledgers there while hoovering up deserted infrastructure abandoned by liquidated minnows.
OpenAI has already broached the concept of government backstops to guarantee its mammoth infrastructure spend. Sarah Friar, the company’s chief financial officer, floated the idea during a recent Wall Street Journal panel, only for Altman to downplay such ambitions amid a firestorm. (33) However, Altman has evoked federal backstops himself in the past, referring to the government as ‘the insurer of last resort.’ (34)
Barring such a controversial bailout, OpenAI may be undone by its porous business model. Meanwhile, Nvidia could be undercut by cheaper upstarts. Ultimately, though, many Big Tech giants – Alphabet, Amazon, Apple, Meta, Microsoft, Tesla – do have the saving grace of profitable products aside from AI. Indeed, modern life is so interwoven with the internet that, even if we see an AI retreat, the toothpaste is not going back into the tube, and their original sources of enrichment may forgive their future-focused folly.
Yes, corporations will continue to bake enterprise AI solutions into their operations because – well – FOMO. And sure, people will subscribe to chatbots they have vaguely befriended, perhaps even paying up to $20 or $30 per month to feed the addiction. Other users will be bombarded with ads and see once-free features shuffled behind paywalls. But will it all add up? Will enough revenue be dumped into the cavernous hole of expenditure? Will there be enough corporate or consumer appetite to sustain ChatGPT and Claude and Gemini and Copilot and Perplexity and Grok and DeepSeek? Probably not. Subscription fatigue is real, and the average Joe will probably only engage with a couple of AI products worth the outlay.
Right now, Wall Street is seeing some AI pullbacks and associated volatility, but nothing major, en masse, in terms of compounding downward trends. Again, though, economics is not my forte, and the stock market is a dangerous game I do not play, so my musings should not be taken as advice. I’m simply just sounding an alarm regarding incredible numbers that do not add up – even to an admitted dilettante.
Overall, the AI bubble feels like a fantasy of reckless speculation built on hopeful yet unsubstantiated vibes without justification. The economics are insatiable, and our shared tolerance of them is born of mass delusion. We would all do well to pull back from the ledge and recalibrate with reality. Otherwise, we will drive ourselves – and everyone else – to insanity.
Sources
1. Bort, Julie. TechCrunch. [Online] November 6, 2025. https://techcrunch.com/2025/11/06/sam-altman-says-openai-has-20b-arr-and-about-1-4-trillion-in-data-center-commitments/.
2. Jesse Noffsinger, Mark Patel, Pankaj Sachdeva, Arjita Bhan, Haley Chang, Maria Goodpaster. McKinsey. [Online] April 28, 2025. https://www.mckinsey.com/industries/technology-media-and-telecommunications/our-insights/the-cost-of-compute-a-7-trillion-dollar-race-to-scale-data-centers.
3. Wright, Ian. Business Energy UK. [Online] September 4, 2025. https://www.businessenergyuk.com/knowledge-hub/chatgpt-energy-consumption-visualized/.
4. Silberling, Amanda. TechCrunch. [Online] July 21, 2025. https://techcrunch.com/2025/07/21/chatgpt-users-send-2-5-billion-prompts-a-day/.
5. Sigalos, MacKenzie. CNBC. [Online] September 24, 2025. https://www.cnbc.com/2025/09/23/sam-altman-openais-850-billion-in-planned-buildouts-bubble-concern.html.
6. Akwuaka, Blaise. MSN. [Online] October 8, 2025. https://www.msn.com/en-us/money/markets/openai-is-about-to-use-more-power-than-switzerland-and-portugal-combined/ar-AA1O39DQ?ocid=finance-verthp-feeds.
7. Thompson, Derek. [Online] October 2, 2025. https://www.derekthompson.org/p/this-is-how-the-ai-bubble-will-pop.
8. Lockett, Will. [Online] October 23, 2025. https://wlockett.medium.com/you-have-no-idea-how-screwed-openai-actually-is-8358dccfca1c.
9. Smith, Dave. Fortune. [Online] November 11, 2025. https://fortune.com/2025/11/12/openai-cash-burn-rate-annual-losses-2028-profitable-2030-financial-documents/.
10. Liu, Phoebe. Forbes. [Online] November 10, 2025. https://www.forbes.com/sites/phoebeliu/2025/11/10/openai-spending-ai-generated-sora-videos/.
11. Zitron, Ed. Where's Your Ed At? [Online] September 29, 2025. https://www.wheresyoured.at/the-case-against-generative-ai/.
12. Smith, Brad. Microsoft. [Online] January 3, 2025. https://blogs.microsoft.com/on-the-issues/2025/01/03/the-golden-opportunity-for-american-ai/#:~:text=In%20FY%202025%2C%20Microsoft%20is,based%20applications%20around%20the%20world..
13. McCallum, Shiona. BBC. [Online] August 29, 2025. https://www.bbc.co.uk/news/articles/ckgyk2p55g8o.
14. Gerken, Tom. BBC. [Online] June 18, 2024. https://www.bbc.co.uk/news/articles/c722gne7qngo.
15. State of AI in Business 2025. Aditya Challapally, Chris Prease, Ramesh Raskar, Pradyumna Chari. s.l. : MIT Nanda, 2025.
16. Stokel-Walker, Chris. The History of the Internet in Byte-Sized Chunks. 2023.
17. Corporate Finance Institute. [Online] https://corporatefinanceinstitute.com/resources/career-map/sell-side/capital-markets/dotcom-bubble/#:~:text=Understanding%20the%20Dotcom%20Bubble,estimated%20at%20around%20$5%20trillion..
18. McCullough, Brian. How the Internet Happened: From Netscape to the iPhone. 2018.
19. Financial Times. [Online] October 5, 2025.
20. Dellinger, AJ. Gizmodo. [Online] November 13, 2025. https://gizmodo.com/the-big-short-guy-shuts-down-hedge-fund-amid-ai-bubble-fears-2000685539.
21. Reuters. [Online] November 17, 2025. https://www.reuters.com/business/media-telecom/peter-thiels-fund-offloaded-nvidia-stake-third-quarter-filing-shows-2025-11-17/.
22. Kaye, Danielle. BBC. [Online] October 29, 2025. https://www.bbc.co.uk/news/articles/cp8e970vn5vo#:~:text=Nvidia%20hits%20new%20milestone%20as%20world's%20first%20%245tn%20company&text=Nvidia%20has%20hit%20a%20new,tn%20(%C2%A33.8tn)..
23. Faisal Islam, Rachel Clun. BBC. [Online] November 18, 2025. https://www.bbc.co.uk/news/articles/cwy7vrd8k4eo.
24. Rogelberg, Sasha. MSN. [Online] November 18, 2025. https://www.msn.com/en-us/news/technology/i-m-deeply-uncomfortable-anthropic-ceo-warns-that-a-cadre-of-ai-leaders-including-himself-should-not-be-in-charge-of-the-technology-s-future/ar-AA1QC6Kh?ocid=msedgntp&pc=DCTE&cvid=3f2d9650cf5241f0f10a670932578877&.
25. Kharpal, Arjun. CNBC. [Online] October 3, 2025. https://www.cnbc.com/2025/10/03/jeff-bezos-ai-in-an-industrial-bubble-but-society-to-benefit.html.
26. Butts, Dylan. CNBC. [Online] August 18, 2025. https://www.cnbc.com/2025/08/18/openai-sam-altman-warns-ai-market-is-in-a-bubble.html.
27. Emily Forgash, Agnee Ghosh. Bloomberg. [Online] https://www.bloomberg.com/news/features/2025-10-07/openai-s-nvidia-amd-deals-boost-1-trillion-ai-boom-with-circular-deals.
28. Blakeley, Grace. Grace Blakeley. [Online] October 21, 2025. https://graceblakeley.substack.com/p/the-ai-circular-economy.
29. Peter Hoskins, Charlotte Edwards. BBC. [Online] January 28, 2025. https://www.bbc.co.uk/news/articles/c4gpq01rvd4o.
30. Cosgrove, Emma. BusinessInsider. [Online] January 27, 2025. https://www.businessinsider.com/explaining-deepseek-chinese-models-efficiency-scaring-markets-2025-1#:~:text=The%20price%20of%20DeepSeek's%20open,from%20OpenAI%2C%20Bernstein%20analysts%20said.&text=But%20the%20potentially%20more%20nerve,months%2C%20the%2.
31. Economist. [Online] October 15, 2025. https://www.economist.com/by-invitation/2025/10/15/gita-gopinath-on-the-crash-that-could-torch-35trn-of-wealth?utm_medium=cpc.adword.pd&utm_source=google&ppccampaignID=18156330227&ppcadID=&utm_campaign=a.22brand_pmax&utm_content=conversion.direct-response.
32. International Monetary Fund. [Online] https://data.imf.org/en/Data-Explorer?datasetUrn=IMF.RES:WEO(9.0.0).
33. Rob Wile, Steve Kopack. NBC News. [Online] November 6, 2025. https://www.nbcnews.com/business/business-news/openais-sam-altman-backtracks-cfos-government-backstop-talk-rcna242447.
34. Cowen, Tyler. Conversations with Tyler podcast. [Online] November 5, 2025. https://conversationswithtyler.com/episodes/sam-altman-2/.