Copy article

The Emperor's New Data Centre. Who's Paying for Tech Giants' AI Naked Ambition?

ended 04. November 2025

Should investors be reducing exposure to tech stocks showing historically unprecedented capital expenditure with unclear ROI timelines?

The numbers don't stack up. Wired confirms Meta, Google, and Microsoft committed over $200 billion in combined annual AI infrastructure spending, while the Australian government forced Deloitte to refund $440,000 for an AI-generated report containing fabrications. That's your bubble warning shot.

MIT reports 95% of generative AI projects fail. S&P's 2025 Global Market Intelligence Report found 42% of businesses are scrapping most AI initiatives, up from 17% last year. Only 6% of AI investments hit expected payback periods. 

Meanwhile, Google's carbon emissions soared 51% since 2019 driven by datacentre growth. AI data centres will double electricity consumption by 2026 to 1,000TWh, and use more power than the whole of Japan by 2030 according to IEA.. The US grid saw $9.3 billion in electricity price increases with rising prices likely to constraint future data centre growth. A mid-sized data centre uses approximately 300,000 gallons of water per day. 

Model collapse. Cannibalistic AI systems are being trained on an internet flooded with AI-generated "slop" and state-actor misinformation. Russian disinformation networks published 3.6 million articles in 2024 alone, deliberately manipulating training datasets. This "large-language model grooming," is successfully manipulating AI outputs. A 2025 KPMG study found 66% of employees use AI output without checking accuracy, and 56% admit to work mistakes because of it.

And who carries the risk? Not the providers. Insurers won't cover "black box" AI systems. OpenAI is using VC money to manage its libaility. When developers build AI tools for clients, they often carry unlimited liability. Meanwhile, OpenAI caps its own liability at "the greater of your last 12 months' payments or $100" if the user is on the free plan. This forces businesses leveraging AI systems to bear the lion’s share of the risk for claims arising from defective products or vendor negligence.

We want your views:

  • Do you agree this is a bubble? Why, why not?
  • If AI models are intentionally being weaponised by state actors via "LLM grooming," how does any business trust the outputs?
  • When 95% of projects fail and only 6% hit payback targets, who's left holding the baby when the bubble bursts?
  • Should regulators intervene when liability is capped at $100-$240 but losses are unlimited for developers building on third-party platforms?
  • If good quality training material is consistently polluted by caniballistic consumption of AI slop and state-led misinterpretation, Is AI Model collapse going to happen sooner or later?

2 responses from the Newspage community

Copy all

Star Quote
Copy

Colette Mason
AI Ethics Consultant at Clever Clogs AI
AI has transformative potential when built responsibly, but this spending spree is sabotaging the technology before it’s proven. We’re now training billion-dollar models on polluted internet data and state disinformation, while copyright infringement lawsuits mount. We’re burning national-grid levels of electricity to power systems that hallucinate too often to trust. And the biggest players are shifting financial risk onto developers and small businesses while capping their own liability at pocket change. Livelihoods, and ultimately lives, are under threat as societies and economies are hitched to this self-interested and unpredictable cart.

AI done right makes humans brilliant. AI done fast erodes trust, fairness and accountability, and leaves small businesses carrying the uninsurable cost when it fails. The bubble is inflating fast.
Copy

Editor
Mitali Deypurkaystha
AI Governance Strategist & CMO at AISensa
We’ve been here before. The dot-com bubble wasn’t caused by the internet itself. It was caused by FOMO-fuelled investors drunk on the idea that it would “change everything.” The irony? It did. But the hype crash slowed progress by years. The same risk now shadows AI.

Yes, AI holds real advantages over the late ’90s: genuine utility already embedded, infrastructure with staying power, and stronger fundamentals from giants like Nvidia, Microsoft, and Amazon, profitable firms reinvesting earnings, not chasing venture cash without revenue. But it’s not immune to becoming a bubble.

We must resist treating AI as a get-rich-quick scheme. Right now, we’re flooring the accelerator while the engine’s still being tested. Those investing for hype, not value, risk stalling innovation for everyone. AI will change everything, just as the web did. But the lesson from the ’90s still stands: technology isn’t the bubble, FOMO is.