Riding the AI Tiger

Why nobody can afford to slow down.

There is an old saying: if you ride a tiger, getting off is not an option. I think this is increasingly becoming the story of the global AI race.

The question is no longer "should we invest in AI?" For the major technology companies and nations, that decision has largely been made. The question is: how fast can we move without falling behind?

And that creates a dangerous feedback loop. Your competitor invests — you invest more. They build more compute — you need more compute. They hire more AI researchers — you increase compensation. They launch a better model — you accelerate yours. They enter a new market — you cannot afford not to.

The tiger gets faster. So everyone rides faster.

Look at the money

It is difficult to call this merely a marketing campaign when the world's largest technology companies are committing hundreds of billions of dollars to infrastructure.

Alphabet expects 2026 capital expenditure of approximately $175–185 billion, with the majority going toward technical infrastructure — a figure it has since revised to $180–190 billion, with 2027 spending expected to increase significantly further. Meta currently expects $130–145 billion of 2026 capital expenditure, driven substantially by infrastructure investment supporting its AI efforts. Amazon expects approximately $200 billion of 2026 capital expenditure across the company, explicitly citing AI, chips, robotics, and other infrastructure opportunities as major drivers. Microsoft spent $41 billion in capital expenditure in a single quarter of FY2026, with roughly two-thirds going toward short-lived assets — primarily CPUs and GPUs.

And these are not isolated bets. OpenAI's Stargate project was announced as a $500 billion, four-year AI infrastructure commitment, later reporting more than $450 billion of investment and more than 8 GW of planned capacity. Nvidia has announced an intention to invest up to $100 billion in OpenAI, linked to deployment of at least 10 GW of Nvidia systems. Anthropic raised $30 billion in its Series G in February 2026 and another $65 billion in Series H in May, reaching a reported $965 billion post-money valuation. Amazon separately committed more than $100 billion over ten years to AWS technologies for Anthropic, alongside up to 5 GW of new compute capacity. And xAI raised $20 billion in its Series E, ending 2025 with more than one million H100-equivalent GPUs across its Colossus infrastructure.

This isn't a startup hype cycle anymore. It is an industrial race.

And then something even more interesting happens

The money doesn't simply flow into AI companies. It flows through an entire ecosystem.

AI labs need enormous compute. Compute needs Nvidia, AMD, or custom silicon. That needs datacenters — Microsoft, Google, Amazon, Oracle, CoreWeave, and others. Datacenters need electricity — power generation, grids, nuclear, renewables. They need networking — Broadcom, fiber, subsea infrastructure. They need construction — developers, engineering, cooling, electrical equipment. And all of it needs capital — banks, private equity, sovereign funds, public markets.

More capital enables more capacity. More capacity enables better AI. Better AI creates more demand. More demand justifies more investment. Repeat.

AI isn't just becoming an industry. It is beginning to pull entire industries behind it.

The circular AI economy

This creates something that resembles a giant economic flywheel. The model company needs compute. The cloud provider supplies it. The chip company sells the hardware. The cloud provider builds more datacenters. The energy industry builds more generation. Investors fund the expansion. Customers buy AI services. Revenue justifies more infrastructure. More infrastructure enables larger models. Larger models create new applications. New applications create more demand. And demand justifies even more infrastructure.

The tiger feeds itself.

But here is the dangerous part

Every participant has a reason to continue.

Imagine Microsoft slowing down — Google doesn't necessarily slow down. Imagine Google slowing down — Meta doesn't necessarily slow down. Imagine the United States slowing down — China may not. Imagine China slowing down — another country may see an opportunity.

Imagine one AI company deciding, "perhaps we should spend less this year and focus on safety." Its competitor may interpret that not as responsibility, but as weakness.

This is the fundamental problem. What may be rational for humanity can be irrational for an individual competitor.

The prisoner's dilemma, at planetary scale

This is essentially a prisoner's dilemma. Everyone might benefit if everyone agrees to slow down. But nobody wants to be the one who slows down first, because the downside is asymmetric.

If everyone accelerates, we get enormous progress — and potentially enormous risk. If everyone slows down, we potentially gain safety. But if you slow down while your competitor accelerates, you may lose the future.

That is an extraordinarily powerful incentive. And it explains why calls to simply "stop AI development" sound much easier than they actually are.

Even the investors are riding the tiger

There is another layer. Investors aren't simply funding AI because it is fashionable — they are increasingly pricing AI into the future value of some of the world's largest companies.

SoftBank is perhaps the clearest example. It had already invested $34.6 billion in OpenAI by the end of fiscal 2025 and committed another $30 billion in 2026. By September 2026, SoftBank had drawn $30 billion under a financing facility associated primarily with that investment.

That creates its own pressure. Once enormous amounts of capital are committed, the question becomes: where is the return going to come from? The answer has to be some combination of more AI users, more enterprise adoption, more automation, more revenue, more productivity, more powerful models, new markets — eventually, perhaps, entirely new industries.

In other words: the AI economy has to work.

The ROI problem

And this is where the "marketing stunt" argument becomes interesting. Yes, there is hype. Yes, there is speculation. Yes, companies have incentives to make extraordinary claims. But that doesn't explain away the infrastructure being built.

The more difficult question is actually the opposite: what happens if these investments don't generate the returns investors expect? Hundreds of billions of dollars of infrastructure cannot remain economically irrelevant forever. At some point, the industry has to demonstrate extraordinary productivity gains or discover extraordinary new sources of revenue.

The pressure to prove AI's economic value therefore becomes another force pushing the tiger forward.

And governments are riding it too

This isn't purely a corporate race anymore. Governments increasingly view AI infrastructure as strategic infrastructure. Compute is becoming a national capability. Semiconductors are strategic. Energy is strategic. Data is strategic. AI talent is strategic. And increasingly, intelligence itself may become strategic.

The United States does not want China to dominate frontier AI. China does not want the United States to maintain an unassailable lead. Other countries don't want to become permanent consumers of someone else's intelligence infrastructure. India, the Gulf states, Europe, Japan, South Korea, and others are therefore trying to establish their own positions.

The race is becoming global.

China is not sitting still

China is building its own AI ecosystem. Alibaba, Tencent, Baidu, ByteDance, DeepSeek, and domestic semiconductor companies are all part of a much broader effort to reduce dependence on U.S. technology.

ByteDance, for example, secured a $29.6 billion bank loan in 2026, largely intended to support AI-related investment including chips and data-center infrastructure. Chinese AI companies increased AI-related capital expenditure by 105% year-on-year in Q2 2026. And China's domestic AI-chip ecosystem is attracting substantial capital — Tencent-backed Enflame raised approximately $912 million in its September 2026 IPO, as investors bet on China's push toward semiconductor self-sufficiency.

The strategic message is clear: nobody wants to depend entirely on another country's intelligence infrastructure.

So who is actually controlling the tiger?

This is the question that worries me.

We often talk about AI alignment, AI safety, AI regulation. But perhaps there is another problem: competitive alignment. Can the companies align with each other? Can nations align with each other? Can investors align with long-term societal interests? Can governments cooperate when national security is involved?

Because even if everyone agrees that uncontrolled AI could be dangerous, the incentive to get there first may remain stronger than the incentive to slow down.

And now comes the next stage

So far, much of AI is still something humans operate. We ask. AI responds. We instruct. AI executes.

But that is changing. The next leap is agentic AI — AI that doesn't simply answer. It plans. It uses tools. It executes. It coordinates. It monitors. It makes decisions. And potentially, it delegates work to other AI systems.

That changes the equation again. Because the tiger isn't merely becoming stronger. It is beginning to move on its own. And that is where this story gets considerably more interesting.

---

The question worth debating: if every major company and nation believes that slowing down could mean losing the AI race, who is actually capable of pressing the brakes? And perhaps the more uncomfortable question — what happens when the economic cost of slowing down becomes greater than the perceived risk of continuing?

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.