No, the strange thing is that the money increasingly appears to move in a circle.
Nvidia invests in OpenAI. OpenAI buys Nvidia systems. AMD gives OpenAI warrants tied to future GPU deployment. Amazon invests in Anthropic, while Anthropic commits enormous future spending to AWS. CoreWeave buys and rents the raw material of the age, accelerated compute, while Nvidia takes a stake in CoreWeave. Private credit and infrastructure funds pour into data centres, power agreements, GPU-backed financing and lease structures. The whole thing begins to look less like a supply chain and more like a weather system.
Or perhaps a mine.
My great grandfather was a gold miner near Charters Towers in Queensland, part of the gold rush of the 1880s. He had travelled from his home in Wartemburg and worked as an amalgamator, the person involved in separating gold from crushed ore using mercury. It was dirty, dangerous, practical work. No glossy strategy deck. No “AI transformation roadmap”. Just rock, heat, labour and the promise that somewhere in the dust there might be something worth extracting.
That old gold rush feels uncomfortably modern.
Because the AI rush is also about extraction. Not just extracting answers from models, but extracting value from capital markets, power grids, chip supply, human attention, public trust and the future itself. The promise is astonishing: machines that can reason, code, tutor, diagnose, design, plan and perhaps one day run whole slices of the economy. But under that promise sits a harder question.
Who is actually paying for all this?
Let’s be real. AI is not “just another software wave”. Software used to scale beautifully because the marginal cost of serving one more customer was close to nothing. Write the code once, sell it forever, watch the gross margins purr like a well-fed cat. AI is different. It eats. It eats chips, power, cooling, land, water, engineering time, data, debt and managerial attention. It needs data centres that look less like offices and more like industrial plant.
That changes the economics. It also changes the risk.
A normal boom has buyers and sellers. A suspicious boom starts to blur the distinction. In today’s AI economy, the supplier may also be the investor. The customer may also hold equity upside in the supplier. The cloud provider may also be the strategic backer. The data-centre operator may be financed by future contracts with customers whose own business models depend on yet more capital arriving. Everyone is validating everyone else.
This does not mean the boom is fake. That would be too easy, and probably wrong. The gold in Charters Towers was real. Railways were real. The internet was real. Housing was real. The problem with bubbles is not usually that the underlying thing is imaginary. The problem is that the financial story gets ahead of the productive reality.
That is where AI feels dangerous.
Nvidia’s proposed partnership with OpenAI is the cleanest example of the new circularity. Nvidia intends to invest up to $100 billion in OpenAI as OpenAI deploys vast amounts of Nvidia infrastructure. On one level, this makes sense. OpenAI needs compute. Nvidia makes the most coveted compute. The two companies are joined at the hip anyway, so why not make it formal?
But viewed another way, it has the shape of vendor-financed demand. Nvidia backs the customer. The customer buys Nvidia systems. The purchases support Nvidia’s growth story. Nvidia’s growth story supports the value of the whole ecosystem. Round and round it goes.
AMD’s OpenAI deal has a similar flavour, though with its own twist. OpenAI receives warrants tied to deployment milestones for AMD GPUs. If AMD’s position in AI strengthens, OpenAI can benefit from AMD’s rising share price. Again, not improper by itself. Strategic alignment is not a crime. But it does mean that what looks like independent market demand is partly wrapped in financial engineering.
Amazon and Anthropic tell a related story. Amazon invests heavily in Anthropic. Anthropic commits heavily to AWS. Amazon’s chips, cloud and infrastructure become part of Anthropic’s future. Anthropic’s growth story becomes part of Amazon’s AI story. Everyone can point to everyone else and say: look, the demand is real.
Perhaps it is.
But if you need a hall of mirrors to prove there’s a window, you should at least check the glass.
CoreWeave sits right in the middle of this. It is one of the purest expressions of the AI infrastructure boom: a company built to provide the GPU-rich cloud capacity that AI labs crave. Its contracts are enormous. Its financing is enormous. Its ambition is enormous. It is not hard to see why investors like the story. In a gold rush, selling picks and shovels is usually better than panning in the river.
Yet even picks and shovels can be overproduced.
The more capital floods into AI infrastructure, the more the industry needs future demand to arrive exactly on schedule. Data centres must be filled. GPUs must be utilised. Power must be contracted. Debt must be serviced. Customers must keep spending. Models must keep improving. Users must pay enough for the services to justify the compute beneath them.
That is a lot of “must”.
And this is where the black hole metaphor matters. A bubble floats. A black hole pulls. AI is doing both.
It has bubble characteristics because some valuations and investment assumptions appear to rely on future profits that have not yet been proven at anything like the necessary scale. There is real revenue, yes. Real adoption, yes. Real productivity in some areas, yes. But there is also a very large gap between “people use this” and “this can support trillions of dollars of infrastructure investment”.
Plenty of things are useful without being profitable enough to justify their own mythology.
But AI also behaves like a black hole because it pulls everything towards it. Capital. Talent. Government attention. Boardroom anxiety. Electricity infrastructure. Procurement language. Consultancy decks. Change programmes. Every organisation is being asked to look again at its processes, its workforce, its data, its risk appetite and its suppliers. Even deciding not to use AI now requires an AI position.
That is not hype. That is gravity.
For those of us working in organisational change, the more interesting danger is not that AI disappears. It won’t. The danger is that AI becomes the universal explanation for everything. Slow service? Add AI. Backlog? Add AI. Poor data? Add AI. Fragile workflow? Add AI. Weak governance? Add AI and call it innovation.
But AI does not abolish the boring work. It exposes whether the boring work was ever done properly.
If your data is messy, AI will not make it clean. It may make the mess more persuasive. If your process is unclear, AI will not make it accountable. It may make the ambiguity move faster. If your governance is weak, AI will not make your decisions safer. It may simply produce a better-formatted mistake.
This connects to the quieter fear behind AI hallucinations. People often talk about hallucination as if it were a chatbot problem. A funny little quirk. The model invented a source, misread a question or confidently made up a fact. Fine, we laugh, we correct it, we move on.
But once AI is placed inside a workflow, hallucination becomes institutional. It stops being a weird answer in a text box and becomes a recommendation, a case note, a triage decision, a project assumption, a procurement summary or a risk assessment. The problem is not that the machine sounds stupid. The problem is that it sounds competent.
That is the hallucination of competence.
Markets can hallucinate competence too. A giant contract looks like demand. A strategic investment looks like validation. A data-centre buildout looks like destiny. A rising share price looks like proof. Everyone agrees the ore body is rich because everyone has already bought equipment to dig it.
The 1880s gold rush had its own version of this. A few people struck gold. Many more followed. Towns grew. Claims were sold. Equipment was financed. Fortunes were imagined before they were extracted. Some made money. Others inhaled dust, handled mercury and discovered that being close to gold was not the same as owning it.
That may be the best way to understand the AI economy now. There is gold in the ground. But there is also mercury in the process.
The uncomfortable question is whether enough real economic value can be extracted to justify the machinery already being built around it. Not someday. Not in a pitch deck. Not in a demo where the future is always six months away. In actual customer revenue, actual productivity, actual cost reduction, actual public value and actual cash flow.
For public services, this should produce neither panic nor paralysis. We should not treat AI as a toy, but nor should we treat it as magic. The sensible position is disciplined curiosity. Use it where the task is bounded. Test it where the risk is low. Demand evidence where claims are made. Keep humans accountable where rights, money, law or vulnerability are involved. Build audit trails before building mythology.
Most of all, separate the tool from the theatre.
The current AI boom may not collapse. It may become the next layer of industrial infrastructure. The data centres may fill, the models may improve, the applications may mature and the capital may eventually find its return. That is possible. It would be foolish to pretend otherwise.
But it would be equally foolish to ignore the circularity. When suppliers fund customers, customers buy from suppliers, cloud providers fund model companies and private credit finances the data centres needed to keep the loop spinning, the burden of proof rises. The more circular the capital becomes, the more external value the system must eventually demonstrate.
Otherwise, we are not watching an economy being transformed.
We are watching a mine being dug with borrowed money, under floodlights, while everyone argues about the price of gold.
AI is probably not just a bubble. It is too useful, too widely adopted and too deeply embedded for that. But it is also not simply the next inevitable step in progress. It is a black hole with a bubble at the edge: real gravity, real danger, real extraction and a shining rim of speculation.
The question is not whether there is gold in the rock.
The question is who gets rich, who breathes the mercury and who is left paying for the machinery when the rush moves on.
Comments
Post a Comment