Inside the Circular AI Economy: Growth, Risk, and the Coming Correction

A complex web of partnerships now defines the global artificial intelligence economy. At the center of it all stands Nvidia — valued at over $4.5 trillion — surrounded by OpenAI, Microsoft, Oracle, AMD, and a constellation of startups. Together, they form what experts increasingly describe as a “circular AI economy”, a self-reinforcing system where money, chips, and cloud power continuously circulate among the same few players.

But behind the dazzling valuations and trillion-dollar promises lies a question haunting Silicon Valley and Wall Street alike: Is this the next big tech bubble?

The Core Loop: Money in, Compute Out

At the heart of this ecosystem is Nvidia, the undisputed supplier of AI’s computing backbone. Its GPUs power the vast data centers running OpenAI’s models, Microsoft’s Copilot tools, and Oracle’s cloud services. In return, these firms pour billions back into Nvidia — either as direct customers or as investors. Nvidia itself has pledged to invest up to $100 billion in OpenAI, the maker of ChatGPT, tightening the loop even further.

OpenAI, now valued at roughly $500 billion, operates as both Nvidia’s biggest client and a cornerstone of Microsoft’s AI strategy. Microsoft integrates OpenAI’s models into its own software ecosystem — from Word to Azure — while maintaining one of the largest compute partnerships in tech history. Oracle, meanwhile, has signed a $300 billion cloud deal with OpenAI and spends tens of billions annually on Nvidia chips to support its expanding AI infrastructure.

AMD is the wild card. In a bold move, it offered OpenAI up to 6 gigawatts of GPU capacity and the option to buy 160 million AMD shares, signaling an effort to break Nvidia’s dominance. Yet the ecosystem remains tightly knit — and heavily dependent on one another’s success.

The Super Network of AI Power

This circular system of mutual investment and supply creates both strength and fragility. It ensures steady demand for GPUs and AI services, but also binds the industry into a single point of failure. If Nvidia’s production stalls, or if OpenAI’s valuation dips sharply, the entire AI sector could wobble.

Economists compare it to the financial loops before the 2000 dot-com crash — where companies invested in each other’s growth stories, inflating valuations far ahead of real revenue. The key difference: this time, there’s tangible infrastructure behind the hype. AI is not just an idea — it’s data centers, chips, and billions of dollars of hardware already online.

Still, this interdependence blurs the line between investment and speculation. “We’re witnessing a digital ouroboros,” one analyst quipped — “a system feeding on itself.”

Short-Term Outlook: Exuberance and Expansion

In the next 12 to 18 months, expect an unprecedented wave of spending across the AI supply chain. Microsoft, Oracle, Amazon, and Meta are projected to collectively invest over $500 billion in new data centers. Nvidia’s GPU backlog will remain fully booked, while OpenAI and xAI will race to release newer, more capable models.

This phase mirrors the dot-com buildout — vast infrastructure expansion, some of it unnecessary, justified by expectations of future demand. Private valuations will likely climb further, and pressure for high-profile AI IPOs will grow. If OpenAI or Anthropic go public, it could mark the peak of the current AI cycle.

Medium-Term (2027–2028): Correction and Consolidation

By 2027, reality will begin to test the narrative. Many AI startups — especially those orbiting Nvidia and OpenAI — will struggle to survive without continued investment. The first signs of correction will appear as markets demand profitability over hype.

Analysts forecast a 20–40% drop in AI valuations, echoing the dot-com correction of the early 2000s. The strongest companies — Nvidia, Microsoft, OpenAI, Oracle — will endure and consolidate power. Smaller firms will be absorbed or shut down.

The technical focus will also shift. Instead of chasing ever-larger models, the industry will move toward smaller, specialized systems orchestrated through frameworks like the Model Context Protocol (MCP). The race will be for efficiency — less energy, fewer GPUs, smarter computation.

Long-Term (2028–2030): AI Becomes Infrastructure

By the end of the decade, AI will fade from buzzword to backbone. It will run quietly behind operating systems, devices, and global logistics. Training frontier models will be the domain of a few megacorporations, while smaller firms focus on applied, domain-specific AI.

But the challenges will also grow. Global AI energy consumption could exceed 500 terawatt-hours annually by 2030 — equivalent to powering 80 million homes. This will push governments to enforce green AI mandates, compute quotas, and new carbon accounting standards. The AI boom could soon become the largest new source of global electricity demand.

Meanwhile, the concentration of compute power within a handful of U.S. tech firms will raise antitrust alarms. Expect a wave of AI sovereignty initiatives, with the EU, Japan, Saudi Arabia, and the UAE investing in national AI clouds to reduce dependence on American hardware and infrastructure.

Winners, Losers, and Lessons

Winners:

• Nvidia, which will evolve from a chipmaker into a full AI infrastructure company.
• Microsoft, which already embeds AI across its entire enterprise stack.
• Oracle, which has successfully repositioned itself as an AI cloud powerhouse.
• AMD, which will benefit from cost-conscious customers seeking alternatives to Nvidia.

Losers:

• The dozens of AI startups without proprietary data or models — many will burn through their funding.
• Late-stage investors who bought into inflated valuations expecting rapid IPOs.
• Any firm overly dependent on a single AI supplier or cloud provider.

The Coming Normalization

The AI economy is not a mirage — it’s real, valuable, and transformative. But it is also overheated and overleveraged. The next few years will bring a natural correction: inflated valuations will shrink, weak players will vanish, and the survivors will build a leaner, more efficient AI infrastructure.

This isn’t the end of AI. It’s simply the end of its first speculative chapter.

Just as the Internet bubble paved the way for the modern web, this “AI bubble” — if it bursts — will clear the field for a more sustainable, energy-conscious, and globally distributed future for artificial intelligence.

Ahmed Banafa's books

Covering: AI, IoT, Blockchain and Quantum Computing