global geopolitics

Debt, Delusion, and the Coming Tech Depression

Tags: Artificial Intelligence, Venture Capital, Hyperscalers, Corporate Debt, Semiconductor Industry, US Defence Budget, Technology Policy, Financial Risk

For the better part of a decade, the technology industry has presented itself as the engine of modernity and economic growth. Its leading firms were celebrated as paragons of efficiency, generating immense wealth from comparatively little physical infrastructure. Software scaled, profits soared, and the promise of artificial intelligence seemed to be the next, inevitable chapter in this story of frictionless expansion. Yet a quiet but profound transformation has taken place within the balance sheets of the largest American technology companies. The five largest US hyperscalersAmazon, Microsoft, Alphabet, Meta, and Oracle, have, in the space of just two years, shifted from being asset-light, cash-rich enterprises to asset-heavy, cash-borrowing behemoths, their capital expenditure now consuming nearly all of their operating cash flow (50). This structural change, driven by an unquenchable thirst for graphics processing units and the memory chips that power them, signals the emergence of a financial bubble whose eventual deflation threatens not just a correction in tech stocks, but a depression of the industry itself.

The scale of the capital commitment is staggering. Consensus estimates for hyperscaler capital expenditure have climbed to nearly $690 billion for 2026 and $870 billion for 2027 (54). Goldman Sachs projects that these companies will allocate approximately 98 per cent of their cash flow from operations to capital expenditures in 2026-. This is a level of investment intensity that recalls the peak of the dot-com boom, but with a crucial difference: it is concentrated in the hands of a few dominant firms rather than distributed across a wide array of start-ups-. This spending is not merely large; it is increasingly debt-financed. US tech giants have already raised approximately $182 billion through bonds in 2026, far exceeding the $13 billion raised in the same period last year (10). The market for this debt, however, is showing signs of fatigue. Amazon’s $25 billion bond offering in July 2026 drew orders of just 1.6 times the issue size, a sharp decline from the heavy oversubscription the company enjoyed in March (11). To attract buyers, Amazon was forced to offer elevated new-issue concessions, a premium that signals a cooling appetite for AI-linked debt (11). As Bank of America noted, “investors are pushing back”(12). The sheer volume of issuance $335 billion globally in 2026 so far, has raised concerns about investor fatigue and portfolio concentration (11).

This debt-fuelled investment cycle is not merely a corporate finance story; it carries macroeconomic implications of a more troubling kind. Deutsche Bank’s Jim Reid has warned that this spending spree could paradoxically fuel inflation (40). The inflationary pressure is most visible in the semiconductor supply chain, where the hyperscalers’ demand for high-bandwidth memory has created a structural crunch. The three dominant memory manufacturer, Samsung, SK Hynix, and Micron, have diverted production capacity toward the lucrative HBM market, tightening supplies of conventional DRAM and driving prices higher (20). Samsung’s revenue per bit from traditional DRAM is forecast to rise 116 per cent year-on-year in 2026, while SK Hynix and Micron are expected to see increases of 78 per cent and 54 per cent respectively (20). This is not a temporary fluctuation; it represents a fundamental shift in pricing power. The memory cartel, as critics have labelled it, now wields unprecedented influence over the cost structure of the entire AI industry-. The hyperscalers, by creating this artificial scarcity, have locked themselves into a strategy where the cost of their core inputs is subject to the whims of three suppliers, and those costs are only expected to rise (24).

At the heart of this financial edifice lies a profound economic contradiction: the technology itself has never been profitable. Despite revenue growth, the leading AI firms continue to burn through cash at an alarming rate. OpenAI’s internal documents project a loss of $14 billion for 2026, roughly three times its 2025 loss-. Anthropic, while showing some operational improvement, is not expected to achieve full-year profitability until 2028-. The fundamental business model remains unproven. The market for large language models is characterised by intense competition, low barriers to entry for new models, and a lack of pricing power (59). Users are highly sensitive to price and performance, and switching costs are minimal, preventing any single firm from establishing a durable moat (59). The industry is, in essence, selling tokens at a loss, hoping that scale will eventually yield profitability. This is a strategy that has no historical precedent for success in a market of this size.

The comparison with the dot-com bubble is instructive but incomplete. The dot-com era saw a proliferation of companies with dubious business models, but the underlying infrastructure, fibre optic cables, networking equipment, retained value and enabled future innovation. The AI bubble, by contrast, is centralised within a handful of firms that have become so dominant that their failure would have systemic consequences. The market capitalisation of these companies is now so intertwined with the broader stock market that a correction could trigger a wider financial contagion (7). This is not a situation where a few start-ups fail and the industry moves on. As Ed Zitron has argued, the hyperscalers have become a “load-bearing” part of the financial system, and their exposure is enormous. The dependence on Nvidia, which accounts for approximately 65 per cent of all high-bandwidth memory consumption, means that any disruption to Nvidia’s supply chain or demand for its products would have cascading effects throughout the industry.

The question of a bailout is often raised, but it is a red herring that obscures the more fundamental problem. The financial crisis of 2008 necessitated intervention because the collapse of institutions like AIG threatened the entire mechanism of overnight funding and commercial paper, upon which the global financial system depended (54). The AI industry, for all its hype, is not systemically important in that sense. A bailout of OpenAI or Anthropic would not solve the underlying problem of an unprofitable business model-40. The US government has allowed large companies to fail before, and it would likely do so again, particularly if the political cost of bailing out data centres for billionaires proved too high. The more likely outcome is a period of creative destruction where overcapacity is wrung out of the system, and the surviving firms emerge leaner but chastened.

This financial reckoning is not confined to the private sector. The US military’s aggressive push to integrate artificial intelligence into its operations faces a parallel crisis of fiscal sustainability. The Pentagon has declared an “AI First” strategy, with plans for substantial modernisation spending-. However, the budgets meant to fund these advanced systems are being drained by the escalating costs of ongoing military commitments, particularly operations linked to the conflict with Iran. The $891 billion defence budget for fiscal year 2026 is under immense strain-. The result is a structural contradiction: a strategy built around advanced technology that is financed through accounts already under stress from conventional warfare. Companies that have invested heavily in defence AI, anticipating a steady stream of government contracts, now face the prospect of delayed payments or cancelled orders as the Pentagon prioritises immediate operational needs over long-term technological ambitions. The market signal is clear: the US government talks big about technological dominance but pays slowly, and in times of conflict, hardware will always crowd out software.

The long-term consequences of this investment cycle are difficult to overstate. If the AI bubble bursts, as many analysts now predict, the damage will extend far beyond the balance sheets of a few tech firms (40). The trust that has been the bedrock of the venture capital industry, the belief that technology investment is a reliable path to growth, will be severely damaged (54). The capital that has flowed into tech for decades may dry up, leaving a generation of entrepreneurs without funding (54). The “Magnificent Seven” firms, which have been the darlings of the stock market, will see their valuations reset-7. This is not merely a correction; it represents a fundamental reimagining of what the tech industry is and what it can be. The path we are on, one of infinite growth funded by ever-increasing debt, is broken. The industry has created a situation where the costs of its ambition are linearly increasing, while the returns remain stubbornly elusive. The punishment, when it comes, will be severe, but it may be the only thing that forces a necessary reckoning.

Authored By: Global GeoPolitics

If you prefer to make a one time donation in support of my work, you can do so by clicking any link below:

https://buymeacoffee.com/ggtv |

https://ko-fi.com/globalgeopolitics |

Bitcoin: 3NiK8BoRZnkwJSHZSekuXKFizGPopkE7ns

Reference List

1. AI capex boom strains hyperscalers’ cash flow as DRAM makers gain pricing power, says Jefferies (2026) Tribune India. Available at: https://www.tribuneindia.com/news/business/ai-capex-boom-strains-hyperscalers-cash-flow-as-dram-makers-gain-pricing-power-says-jefferies/[reference:35][reference:36]
2. AI Credit Expansion: Assessing the Micro and Macro Risks (2026) Advisor Perspectives. Available at: https://www.advisorperspectives.com/commentaries/2026/05/26/ai-credit-expansion-assessing-micro-macro-risks[reference:37]
3. Amazon’s new bonds sale draws muted demand as AI debt floods market (2026) Business Standard. Available at: https://www.business-standard.com/amp/world-news/amazon-s-new-bonds-sale-draws-muted-demand-as-ai-debt-floods-market-126070800198_1.html[reference:38]
4. Amazon’s $25 billion ‘surprise’ bond sale dangled extra yield to lure in buyers—and flashed a warning sign about the AI boom (2026) Yahoo Finance. Available at: https://au.finance.yahoo.com/news/amazon-25-billion-surprise-bond-192838890.html[reference:39]
5. U.S. tech giants raised $182 billion through bonds in 2026 (2026) KuCoin. Available at: https://www.kucoin.com/news/flash/us-tech-giants-raise-182b-via-bonds-in-2026-far-exceeding-last-year-s-13b[reference:40]
6. AI memory boom squeezes legacy DRAM supply, pushing prices higher (2026) S&P Global. Available at: https://www.spglobal.com/market-intelligence/en/news-insights/research/2026/01/ai-memory-boom-squeezes-legacy-dram-supply-pushing-prices-higher[reference:41]
7. HBM prices could double by 2027 on surging AI demand (2026) Yahoo Finance. Available at: https://sg.finance.yahoo.com/news/hbm-prices-could-double-2027-131159931.html[reference:42]
8. AI productivity gains are years away, Deutsche Bank analyst says (2026) Staffing Industry. Available at: https://www.staffingindustry.com/editorial/cws-30-contingent-workforce-strategies/ai-productivity-gains-are-years-away-deutsche-bank-analyst-says[reference:43]
9. 大模型API市場的獲利困局：營收三級跳 為何虧損仍達百億美元？ (2026) 鉅亨網. Available at: https://anuenews.cnyes.com/news/id/6529407[reference:44][reference:45]
10. Particle: Magnificent Seven Trade at Cheapest Relative Valuations in Over a Decade (2026) Particle. Available at: https://particle.news/story/magnificent-seven-trade-at-cheapest-relative-valuations-in-over-a-decade[reference:46]
11. AI Spending Surge Strains Cash Flows, Says Jefferies Report (2026) BW Businessworld. Available at: https://www.businessworld.in/article/AI-Spending-Surge-Strains-Cash-Flows-Says-Jefferies-Report/[reference:47]
12. Deutsche Bank warns 2026 will be ‘do or die’ for independent AI firms (2026) Wikifx. Available at: https://www.wikifx.com/en/newsdetail/202601218314663987.html[reference:48]