Everyone's talking about AI. But who's actually pocketing the money?
OpenAI? Still not profitable. Anthropic? Same. Google and Microsoft earn money, but mainly from other things. So where does the money that companies and users are currently pumping into AI actually go?
Layer 1 — The shovels: Nvidia
When gold was discovered in California in 1848, hundreds of thousands rushed to the mountains. Most came back empty-handed. Who made money? The merchants selling shovels, tents, and supplies.
AI in 2026 works exactly the same way. Most AI companies are digging. One company sells the shovels: Nvidia.
Nvidia makes GPUs — the specialized chips without which neither ChatGPT nor Claude would run. The result: over $215 billion in annual revenue, $120 billion in net profit. Three years ago that profit was $4.4 billion. Today Nvidia earns the same amount in under ten days.
Layer 2 — The cloud: AWS, Azure, Google Cloud
Right behind Nvidia: the major cloud providers. AI models don't just need hardware to train on — they need servers running continuously. Every request to ChatGPT flows through Amazon's, Microsoft's, or Google's data centers.
Microsoft invested $13 billion in OpenAI early on. The return comes via Azure, which attracts more AI workloads every month. Amazon has funded Anthropic with billions — and gets a valuable anchor customer in return. The pattern repeats everywhere: Big Tech subsidizes AI startups and collects on the backend as cloud provider.
Layer 3 — The early investors
Anyone who invested in Nvidia in 2019 has fifty times their money today. Anyone who got into OpenAI early — including Microsoft — is sitting on unrealized gains of astronomical proportions.
That doesn't apply to all investors. Entering late into overvalued AI startups can mean losses. But the early venture capitalists in companies like Anthropic, Perplexity, or Mistral have already won enormously — on paper, at least, and as long as the valuations hold.
Layer 4 — AI tools that actually earn money
Not all AI companies are burning cash. Some tools are already profitable.
Midjourney, the AI image generator, deliberately took no venture capital — and claims to be profitable. Small team, high margins, clear utility.
Perplexity is growing rapidly and has demonstrable user willingness to pay. Currently probably still unprofitable, but the path there is clearer than for OpenAI.
Then there are hundreds of SaaS tools that have embedded AI as a feature — writing tools, code assistants, image editing, document analysis. Most pay for the OpenAI or Anthropic API and resell the result at a premium. High-margin, as long as the base AI stays cheap.
Layer 5 — Freelancers and independents
An often overlooked group: people who use AI to deliver significantly more work in the same time.
A copywriter who used to produce three articles per week now manages eight — with AI assistance. A solo developer delivers projects that used to require a team. A graphic designer accomplishes in an hour what previously took a day.
This isn't a promise — it's the reality for many who've learned to work with AI. This group doesn't earn from AI — but through AI they earn considerably more. No glamorous IPO, but a very real economic effect.
What happens when the bubble bursts?
Everyone's waiting for it. Many call it Dot-Com Bubble 2.0.
It holds — at least partially. Back then too, billions were pumped into companies that never became profitable. Many disappeared. Stock prices collapsed.
But the internet didn't disappear. Amazon, Google, eBay survived and became trillion-dollar companies. The technology stayed — the hype disappeared.
Most likely the same with AI. When the bubble bursts — and the if has become a when for many observers — countless AI startups will fail. Valuations will be corrected. Capital will become more cautious.
But language models, image generators, code assistants won't disappear. The underlying costs will keep falling through better hardware and more efficient models — that happens independently of the hype cycle. Whether users see those savings is a different question: without investor subsidies, survivors need to finally turn a profit. Short-term, prices could actually rise. Long-term, hardware efficiency pushes costs down. And without the gold rush atmosphere, the genuinely useful tools will become clearer — that much is certain.
What remains
Nvidia? Stays — not just because hardware is always needed, but because they've proven they can deliver what the market demands, when it demands it, at a pace no competitor has matched. Add the CUDA platform that half the AI world depends on. That's structural indispensability, not a generic hardware argument. Cloud providers? Stay, because infrastructure stays. Early investors? Already won or got caught in the burst. LLM providers? Will consolidate — not all survive. AI tools with real utility? Stay. Speculative hype? Goes.
And freelancers who've learned to work with AI? They have an enormous advantage: experience — and that's not something you can buy.