Who Is Feeding the Tiger?

Follow the money. Follow the power. Follow the compute.

In Part 1, we asked whether the AI Tiger has already been released. In Part 2, we looked at why nobody seems able to slow it down. Now comes a different question: who is feeding it?

The answer is bigger than OpenAI. Bigger than Nvidia. Bigger than Silicon Valley.

The AI race has created an enormous ecosystem connecting chips, data centers, electricity, cloud providers, investors, governments, startups, and sovereign wealth. And something fascinating is happening: the more AI grows, the more infrastructure it requires. The more infrastructure gets built, the more capital flows into AI. The more capital flows in, the more AI capacity gets built. And the cycle starts again.

AI → GPUs → data centers → power → capital → more AI.

The Tiger isn't just being fed. The ecosystem around it is growing because everyone is feeding it.

Start with the GPUs

If AI is the brain, GPUs are becoming its metabolism — and the numbers are extraordinary.

Nvidia's fiscal 2026 revenue reached $215.9 billion, with Data Center revenue alone reaching $193.7 billion for the year. By Nvidia's fiscal Q2 2027, quarterly Data Center revenue had reached $89 billion, up 117% year over year. That is not a niche semiconductor market anymore. It is becoming one of the central infrastructure markets of the global economy.

And Nvidia is not alone. AMD is competing for accelerator demand. Google is developing TPUs. Amazon is developing Trainium. Microsoft is developing its own silicon. Startups are developing inference-specific chips.

The reason is simple: AI requires enormous amounts of compute, and the next generation may require even more — reasoning, video generation, robotics, autonomous agents, scientific simulation, AI-to-AI communication, continuous inference. The demand isn't simply "give me a better model." It is increasingly "give me more compute."

Then come the data centers

A GPU sitting in a warehouse isn't intelligence. It needs data centers, cooling, networking, storage, power, fiber, backup systems, physical security, land, construction, and skilled engineers. This is where AI starts looking less like software and more like industrial infrastructure.

OpenAI's Stargate initiative is a good example — described as a long-term effort to build the compute infrastructure required for the "Intelligence Age." In 2026, OpenAI and partners broke ground on a 1 GW data-center campus in Michigan.

And the scale keeps growing. PwC estimates global AI infrastructure investment could reach $31.6 trillion cumulatively through 2050, with annual data-center capital expenditure rising from roughly $800 billion in 2026 to $1.8 trillion by 2050.

Think about what that means. AI is creating demand not just for software engineers, but for construction, real estate, electrical equipment, transformers, cooling systems, power generation, fiber networks, semiconductors, cloud infrastructure, and debt financing. The Tiger is pulling entire industries behind it.

And then the tiger needs electricity

This may be the part of the story most people underestimate. There is no AI without electricity.

The International Energy Agency estimates global data-center electricity consumption was around 415 TWh in 2024 and could more than double to roughly 945 TWh by 2030, with AI expected to be the biggest driver of that growth. The IEA estimates data-center electricity consumption grew another 17% in 2025, while electricity consumption from AI-focused data centers grew by about 50%.

Suddenly the AI conversation becomes an energy conversation. More GPUs require more data centers. More data centers require more electricity. More electricity requires more generation and grid capacity — solar, wind, natural gas, nuclear, batteries, transmission, transformers, grid infrastructure. The AI race is beginning to influence the physical architecture of the energy system itself.

There's an interesting paradox here too. AI becomes more energy efficient at the individual task level, but as it becomes cheaper and more capable, people use it more — and new applications like reasoning, video, agents, and autonomous systems can consume vastly more compute. So efficiency doesn't necessarily mean lower total consumption. It can mean more intelligence becomes economically affordable, and therefore more intelligence gets consumed.

Now follow the capital

This is where the story gets particularly interesting. Someone has to pay for all of this — and increasingly, that someone isn't just the AI company. It's Big Tech, venture capital, private equity, banks, infrastructure funds, sovereign wealth funds, public markets, and governments.

The IEA makes an important observation: data-center investment has become too large to be funded entirely from corporate balance sheets. Capital markets are increasingly important to financing the buildout.

And the numbers are staggering. The OECD estimates global VC investment into AI companies reached approximately $258.7 billion in 2025 — around 61% of the total value of global venture capital investment. AI isn't merely attracting venture capital. It's becoming one of the dominant destinations for it. Carta reported that more than 60% of startup funding on its platform in Q1 2026 went to AI companies.

So now ask yourself: what happens when an enormous amount of capital becomes dependent on one technological thesis?

The startup economy joins the feeding chain

This is where the ecosystem becomes even bigger. AI doesn't consist of five companies building five models — there are thousands of companies building around them: AI infrastructure, AI security, data platforms, vector databases, model orchestration, inference optimization, AI agents, robotics, AI coding, synthetic data, AI cybersecurity, AI healthcare, AI finance, AI chips, AI networking, AI energy. Every new layer creates another investment opportunity.

A fascinating example appeared just this month: AI-chip startup Positron AI raised $875 million, taking its valuation to approximately $5 billion — more than four times its valuation only months earlier. This is the AI economy creating companies that create infrastructure for other AI companies, which then consume more infrastructure, which creates more demand, which attracts more investment, which creates more startups.

Then governments enter the game

This is where AI stops being merely an investment story. It becomes a sovereignty story.

Governments increasingly don't want to depend entirely on somebody else's compute, models, chips, data, cloud, or infrastructure. The concept is increasingly called Sovereign AI. A 2026 CNAS analysis tracking government-backed sovereign AI projects found 184 projects across 67 countries — up from just one project identified in its index in January 2023. That is an extraordinary acceleration.

The logic is understandable: no country wants to depend on another for the infrastructure that runs its government AI, defense AI, healthcare AI, financial systems, national data, and critical infrastructure. AI capability increasingly becomes intertwined with national capability. So governments invest in compute, compute requires data centers, data centers require power, power requires infrastructure, infrastructure requires capital, capital requires expected returns, and expected returns require AI demand. The loop gets tighter.

The circular AI economy

This brings us to the central idea of this article. AI companies need more compute. GPU and accelerator companies need more customers. Cloud providers and data centers need more electricity. Energy companies and utilities need more infrastructure. Banks, investors, and sovereign funds provide capital. More AI infrastructure means more compute becomes available. AI becomes more capable. More applications become possible. More businesses adopt AI. More revenue expectations follow. More capital arrives. And the cycle begins again.

AI → compute → infrastructure → energy → capital → AI.

This is what I call the Circular AI Economy. And I don't mean the economy is fake — quite the opposite. There is genuine demand, genuine products, genuine revenues, genuine productivity gains. But something important is happening alongside all of that: capital is accelerating the physical expansion of AI, while that physical expansion is creating the capacity for even more AI. The system is becoming self-reinforcing.

And this is where the risk appears

Every investment cycle eventually faces the same question: what happens if expected returns don't arrive?

Imagine hundreds of billions — or eventually trillions — committed to data centers, GPUs, power plants, networks, AI startups, cloud infrastructure, debt, private equity, public markets, and sovereign projects, all based partly on the expectation that AI demand will continue growing.

What happens if it does? The infrastructure becomes incredibly valuable. What happens if it grows even faster than expected? The Tiger becomes enormous. But what happens if AI adoption disappoints? If model improvements slow? If inference becomes dramatically more efficient? If customers refuse to pay enough? If thousands of AI startups fail? If data centers are built faster than demand?

Suddenly the question changes from "how powerful will AI become?" to "who is carrying the financial risk of making AI so powerful?" That question deserves much more attention.

The financial system is now watching the tiger

This isn't just a theoretical concern. Recent financial-market commentary has already highlighted growing AI-related debt and concerns about whether infrastructure investment can generate sufficient returns.

Reuters reported that AI-related debt issuance had approached $500 billion by early August 2026, with lenders becoming more cautious as projects encounter electricity, permitting, and construction constraints. Meanwhile, the Bank for International Settlements has warned that the AI boom could create financial-stability risks, particularly if enormous investment commitments are not matched by future profits.

This doesn't mean AI is a bubble. It doesn't mean the investments are irrational. It certainly doesn't mean AI will fail. It means something more interesting: the financial system is increasingly becoming part of the AI race. And once that happens, slowing down becomes harder — because it is no longer just a technology decision. It becomes an economic decision.

The question nobody wants to ask

Suppose you are a CEO. Your competitor is investing $50 billion in AI infrastructure. You believe perhaps only $20 billion is economically justified. What do you do — wait, or invest anyway because your competitor might gain a strategic advantage?

Now imagine every major company makes the same calculation. Then governments. Then investors. Nobody necessarily believes every dollar is rational. But nobody wants to be the company, country, or investor that underestimates the technology. That is how competitive races work — the fear of missing the future can become almost as powerful as the expectation of profiting from it.

Who benefits if AGI arrives?

And now we reach the uncomfortable question. If AGI — or something approaching it — actually arrives, who benefits most? The companies building the models? The companies supplying the GPUs? The cloud providers? The energy companies? The investors? The governments? The countries controlling the compute? The startups building on top of the models? Or the billions of people who ultimately use the intelligence?

Perhaps the answer is: whoever controls the bottleneck. And today, there are several potential bottlenecks — compute, energy, capital, data, talent, distribution, models. Whoever controls enough of these layers may hold extraordinary economic power.

But there is another possibility

What if the opposite happens? What if AI becomes so cheap and abundant that intelligence itself becomes almost a commodity? Imagine a world where every individual and every small business can access intelligence comparable to today's largest corporations. Then the economic value may shift away from intelligence itself, and toward what people do with it.

That could create the greatest democratization of capability in human history. Or it could create one of the greatest concentrations of economic power in human history. Both outcomes are possible, and we don't yet know which direction we're heading.

My view

I don't think the important question is "is AI overhyped?" Of course it is — every transformational technology is surrounded by hype. The more important question is: what happens when hype, genuine technological progress, strategic competition, and enormous capital investment reinforce each other?

That is what makes the AI story different. The Tiger isn't being fed by one company. It is being fed by an ecosystem. Investors need AI growth. AI companies need compute. Compute companies need infrastructure. Infrastructure needs energy. Energy needs capital. Governments want sovereignty. Startups want the next giant opportunity. And everyone is looking at everyone else, waiting to see who slows down first.

The three questions I can't stop thinking about

What happens when trillions of dollars depend on AI succeeding? Does capital accelerate innovation, or does it create pressure to keep expanding even when the economics become uncertain?

Can markets objectively assess AI risks when so much capital is invested? Can investors genuinely say "maybe this technology is too risky" when their portfolios, competitors, and future returns increasingly depend on it?

Who benefits most if AGI arrives? The company that creates it? The country that controls it? The investors who finance it? Or humanity itself?

---

We've spent three parts asking why we can't slow the Tiger, why everyone is riding it, and who is feeding it. But the Tiger is no longer just consuming GPUs, electricity, and capital. It is beginning to hunt.

AI agents are moving from answering questions to executing tasks. From generating information to taking actions. From being tools to becoming participants in economic systems. And when that happens, the economics of AI could change again — because the Tiger may no longer need humans to feed it. It may begin to feed itself.

About the author: Mohamed Ismail is an AI thought leader with 26+ years in enterprise transformation and solution architecture on AI, Cloud & Data. He writes on the future race on AI and its consequences.