Synopsis: Major players such as Nvidia and AMD have invested heavily in AI firms — and in turn, those firms committed to buying massive volumes of their hardware (millions of GPUs), effectively channeling the investment back into the investors themselves. This “circular deal” structure has enabled staggering valuations — often disconnected from actual profitability. The inherent risk is systemic: if demand for AI infrastructure slows or doesn’t yield expected returns, valuations could collapse — potentially wiping out paper gains and leaving stranded assets.
The world’s biggest tech companies are stuck in a risky web of mutual investments and chip orders. What happens when the bills come due? The past year has witnessed an unprecedented convergence of capital, ambition, and circular reasoning in the artificial intelligence sector. Starting with Nvidia’s stunning $100 billion commitment to OpenAI in September 2025, a sprawling ecosystem of interlinked deals has emerged, with each company betting billions on its competitors while simultaneously profiting from their dependence.
It’s a dizzying financial dance that would have raised red flags in any previous market boom, yet today it’s being hailed as the foundation of a generational wealth creation opportunity. But beneath the rhetoric of “virtuous cycles” and “AI factories,” a more troubling reality is taking shape: a system of circular financing that bears unsettling similarities to the excesses that preceded the dotcom collapse.
How the Circular Ecosystem Took Shape
The modern AI infrastructure arms race didn’t begin with complex philosophical debates about artificial general intelligence. It began with a deceptively simple problem: OpenAI and other leading AI labs needed compute power, and Nvidia had it. That straightforward supplier-customer relationship has morphed into something far more intricate and risky.
In mid-September 2025, Nvidia announced it would invest up to $100 billion in OpenAI, with the ChatGPT maker committing to purchase millions of Nvidia chips, potentially 4 to 5 million GPUs, nearly double Nvidia’s annual output. Two weeks later, unwilling to miss out on the opportunity, OpenAI struck a nearly identical arrangement with AMD, securing an option to take a 10% stake in the chipmaker in exchange for purchasing up to six gigawatts of AMD’s Instinct MI450 chips, beginning in 2026. These weren’t ordinary commercial transactions. They were investments that would directly fund purchases of the investors’ own products, a pattern that defines what’s now known as “circular deals.”
The architecture of these arrangements reveals the tension at their core. In the Nvidia deal, the chipmaker would invest non-voting shares worth billions into OpenAI, which would then use that capital to purchase Nvidia hardware. With AMD, OpenAI received warrants to buy 160 million shares at just $0.01 each, prices that will only vest as AMD’s stock price hits increasingly ambitious targets, with the final tranche requiring the stock to reach $600 per share. As The New York Times observed at the time, the structure resembled a form of insider stock manipulation cloaked in venture capital language; companies were essentially using their own capital commitments to predictably pump the equity that would compensate them for those very commitments.
The web expanded further. Oracle joined Nvidia and SoftBank in what some analysts called a “Stargate” venture, a $500 billion infrastructure initiative to build AI data centres globally. Microsoft committed $34.9 billion in capital expenditures in a single quarter, while Amazon announced $125 billion for the full year, with both hinting that 2026 spending would be substantially higher. Google and Meta followed suit, each pledging massive increases in infrastructure spending. By late 2025, the major hyperscalers were collectively committing over $370 billion annually to AI infrastructure, with projections suggesting $470 billion by 2026.
Yet here’s the critical detail that most financial commentary glossed over: much of this capital wasn’t flowing from genuine economic returns. It was flowing in circles. As one financial analyst put it, “OpenAI will likely buy gear from Nvidia, which will reinvest those profits in OpenAI, which will likely use those funds to buy even more Nvidia gear.” CoreWeave, a rising AI infrastructure company, offers another example; it raised funding at a $19 billion valuation with capital that included commitments from the very hyperscalers that would later become its primary customers.
The Bubble Signals Are Unmistakable
The similarities to previous market collapses are difficult to ignore. In the run-up to the dot-com crash, venture capital would flow into startups that would then spend that capital on the venture firm’s preferred vendors, inflating both valuations and revenue figures without creating genuine economic value. What made that era so dangerous was that it created an illusion of demand where none existed. Companies would buy products from each other simply to justify their valuations, not because customers actually needed those products.
Today’s AI sector shows several warning signs. First, valuations have become detached from demonstrated profitability. OpenAI’s latest valuation has reached $500 billion, supported by $1 to $1.5 trillion in committed investments, yet the company burns through billions quarterly without clear paths to profitability. Mira Murati, OpenAI’s former CTO, founded Thinking Machines Lab in 2024 and is seeking funding at a $50 billion valuation, less than a year after launch, a valuation with no revenue track record.
Second, the spending patterns themselves contain hidden risks. Goldman Sachs researchers documented that megacap tech companies cut share buybacks, a historically sacred shareholder return mechanism, precisely as AI capital expenditures surged 24% year-over-year in the second quarter of 2025. This wasn’t a sign of confidence in future returns; it was a sign that management viewed share buybacks as less attractive than speculative AI infrastructure investments. By the second half of 2025, S&P 500 buybacks had stalled entirely, redirecting capital into ventures with uncertain timelines for returns.
Third, the circular structures are creating dangerous interdependencies. If demand for AI services slows, perhaps because customers discover the productivity gains don’t justify the infrastructure costs, the entire ecosystem faces a simultaneous shock. Nvidia would see chip demand collapse while its equity stakes in OpenAI and other AI labs became worthless. AMD would watch its warranty obligations to OpenAI become millstones. Microsoft, Amazon, and Google would be left with massive data centres consuming enormous power while generating insufficient revenue.
The financial community began sounding alarms in November 2025. Peter Thiel’s hedge fund quietly sold its entire $100 million Nvidia stake in the three months ending September, according to regulatory filings disclosed just ahead of Nvidia’s earnings season. SoftBank sold $5.8 billion of Nvidia shares. Michael Burry, the investor famous for forecasting the 2008 housing collapse, purchased over $9.2 million in put options against Nvidia and Palantir, positioning himself to profit if those stocks cratered.
On November 17, 2025, the market finally acknowledged the risks that had been building for months. Global equities fell sharply as the tech-heavy Nasdaq closed below a key technical level for the first time since April. Bitcoin lost all of its 2025 gains. Japan’s Nikkei dropped over 3%, while Hong Kong, Europe, and other markets followed. The sell-off was triggered by a simple realization: the spectacular returns promised by AI infrastructure investments might never materialize.
The Sector Outlook
The AI semiconductor and chip market appears structurally sound, for now. The global semiconductor industry recorded $208.4 billion in sales during Q3 2025, marking a robust 15.8% increase year-over-year. DRAM and NAND flash prices are surging by 25-30% as companies rush to build data centres. Intel CPUs are seeing price increases of up to 20%, while specialized AI components face supply constraints. The High Bandwidth Memory (HBM) market is projected to exceed $100 billion by 2030, with next-generation HBM4 components priced 60% above current products.
But this apparent strength masks structural vulnerabilities. The entire supply chain is premised on infinite demand for AI compute, which remains unproven. As companies desperately seek cost advantages, they’re stretching depreciation schedules, a subtle accounting trick that suggests future earnings quality may be weaker than headline figures suggest. CoreWeave, which went public at a $19 billion valuation, ended Q2 2025 with a $30.1 billion contracted backlog. On paper, this looks like evidence of demand. In reality, it’s a bet that hyperscalers will continue spending regardless of returns.
Data centre power consumption presents a more concrete threat. Goldman Sachs projects that data centre electricity demand will increase 160% by 2030, driven almost entirely by AI infrastructure. The International Energy Agency estimates global data centre power consumption will more than double from 415 TWh in 2024 to 945 TWh by 2030, with AI accounting for the vast majority of growth. Approximately 60% of this new capacity will require new power generation, placing enormous strain on already-tight electrical grids across developed economies.
This energy constraint isn’t merely theoretical. Power transmission remains the binding constraint in most developed regions; new generating capacity cannot be connected to grids fast enough to support data centre expansion. In the United States, the preferred mix for new capacity will consist of 30% natural gas combined cycle turbines, 30% natural gas peakers, and only 27.5% solar. This mix will create enormous and rising energy costs, reducing the profit margins on which the entire AI infrastructure narrative depends. If energy costs surge faster than AI service revenues grow, the entire investment thesis collapses.
If the Bubble Bursts
The scenario most feared by sophisticated investors resembles a controlled demolition gone wrong. If demand for AI services moderates, even mildly, the circular ecosystem creates immediate contagion vectors.
Nvidia would simultaneously experience collapsing chip demand and massive equity dilution. Its stakes in OpenAI and other portfolio companies could lose 50-75% of their value as funding dries up. Microsoft, which has invested tens of billions in OpenAI and owns its own equity stake, would face similar dynamics. Amazon’s $11 billion Indiana data centre and $5 billion South Korea commitment would become stranded assets, generating insufficient cash flow. Meta, which has struggled to demonstrate any return on its $70+ billion annual infrastructure spending, would face intense shareholder pressure.
The international capital flows that have fueled this boom could reverse with shocking velocity. As sophisticated investors recognize the circular nature of these deals, confidence will evaporate. Capital that flowed obsessively to U.S. tech stocks could redirect to markets with more fundamental value, potentially including emerging economies like India, whose equity valuations have remained relatively contained by comparison.
What Governments Must Do Now
The path forward requires governments to make difficult choices about intervention and infrastructure planning, choices that current policy frameworks are ill-equipped to handle.
First, regulatory bodies must immediately audit the circular deal structures. The Securities and Exchange Commission should investigate whether these arrangements constitute violations of insider trading rules, round-tripping statutes, or other securities laws. The warrants granted to OpenAI by AMD, particularly the $0.01 pricing, warrant particular scrutiny. Regulators in Europe, where the EU AI Act is already shaping policy frameworks, should use this incident as a case study for how financial engineering can enable circular behaviour that violates the letter and spirit of thoughtful regulation.
Second, governments must accelerate energy infrastructure planning. The 160% projected increase in data centre power demand by 2030 cannot be met with current infrastructure investment rates. Policymakers should establish dedicated permitting processes for renewable energy projects targeting data centre clusters, similar to fast-track mechanisms for defence or critical infrastructure projects. They must also establish realistic pricing for grid access that reflects true capacity constraints. If power becomes expensive enough, it will mechanically slow data centre expansion and reduce the pressure on the system.
Third, regional economic development strategies must account for AI infrastructure’s regional disparities. South Korea and Taiwan, which supply a disproportionate share of advanced semiconductors, face existential economic risk if the AI trade unwinds. Governments in these regions should diversify away from semiconductor concentration. Simultaneously, countries like India should view the potential unwinding of AI allocations as an opportunity to attract capital to more fundamental infrastructure investments that create genuine economic value independent of technology cycles.
Fourth, policymakers should mandate transparency in AI infrastructure deals. OpenAI, Microsoft, and other key players should be required to publish the economic assumptions underlying their infrastructure commitments, specifically, projections for AI service revenue per unit of compute, utilization rates, and timeline to profitability. This information should be subject to the same audit standards applied to government contractors. The current system, where private companies can commit billions to speculative infrastructure with minimal disclosure, is incompatible with responsible capital allocation.
Finally, governments should consider temporary circuit breakers for concentrated AI sector allocations. Some policymakers have proposed taxes on speculative AI investments or restrictions on AI infrastructure funding flows that lack demonstrated customer demand. While such measures face legitimate free-market objections, the concentration of capital in this sector, over 88% of foreign fund flows into U.S. equities, is historically extreme and unsustainable. Circuit breakers won’t prevent a downturn, but they can prevent the downturn from cascading into broader financial system instability.
The Road Ahead: Separating Signal from Hype
Nvidia CEO Jensen Huang has spent the past year making a compelling case for the long-term opportunity in AI infrastructure. His argument, that global GDP incorporates roughly $50 trillion in cognitive labour that could see 2-3x productivity gains through AI deployment, justifying $5 trillion in infrastructure investment by 2030, has a surface logic that appeals to investors desperate for narrative certainty. The parallel between AI infrastructure and previous computing revolutions (mainframes, PCs, the internet) is superficially persuasive.
But there’s a fatal flaw in this logic: previous computing revolutions were driven by proof of performance, not by circular capital flows between suppliers and customers. The internet grew because companies and consumers actually wanted to use it. PCs succeeded because individuals preferred them to mainframes. The AI infrastructure boom, by contrast, is being driven by a small number of hyperscalers locked in competitive dynamics that require them to build regardless of profitability, and a corresponding ecosystem of vendors dependent on those players’ capital allocation decisions.
The question investors should be asking isn’t whether AI will eventually be transformative; it probably will be. The question is whether today’s valuation multiples, capital allocation patterns, and circular deal structures can survive even moderate delays in achieving the promised returns. History suggests they cannot.
The AI sector’s current trajectory is neither sustainable nor inevitable. What began as a genuine technological revolution has morphed into a self-reinforcing bubble of speculation, circular deals, and increasingly desperate capital deployments. If the past eighteen months have taught us anything, it’s that bubbles can persist longer than rationality would suggest, but not forever. When the circular flows finally pause, the consequences for technology stocks, semiconductor demand, and broader equity allocations could prove as severe as the euphoria was extreme. The countdown has begun. The only question now is how much damage will be done before reality catches up with the narrative.
Written by – Rajat Baddi
Disclaimer
The views and investment tips expressed by investment experts/broking houses/rating agencies on tradebrains.in are their own, and not that of the website or its management. Investing in equities poses a risk of financial losses. Investors must therefore exercise due caution while investing or trading in stocks. Trade Brains Technologies Private Limited or the author are not liable for any losses caused as a result of the decision based on this article. Please consult your investment advisor before investing.
