By Rajesh Beri · July 6, 2026
On July 6, 2026, NOTUS obtained a draft report from the U.S. Treasury Department that the administration never intended the public to see.
Career Treasury analysts — not political appointees, not AI skeptics, but the same analysts who monitor systemic financial risk for the Federal Reserve Chair and the Treasury Secretary — concluded that the AI market shares dangerous structural parallels with the dotcom bubble, that AI firms are "more deeply entrenched in the U.S. economy than their dotcom predecessors," and that a downturn would send "shockwaves throughout the entire economic ecosystem."
The report was prepared for Treasury Secretary Scott Bessent, Federal Reserve Board Chair Kevin Warsh, and various federal financial regulators. It has been completed for weeks but is awaiting formal approval.
The Treasury spokesperson's response was telling: "The official position of the Secretary and the U.S. Treasury is that Artificial intelligence will be a key driver of America's new Golden Age."
Translation: the people whose job is to find financial risk found financial risk. And the people whose job is to project confidence are projecting confidence.
For enterprise technology leaders, this creates an uncomfortable question: if the U.S. government's own financial risk analysts are comparing your AI infrastructure providers to dotcom companies, what does your vendor exposure actually look like?
The $1.2 Trillion Debt Pile Behind Your AI Stack
The numbers that alarmed Treasury analysts aren't speculative. They come from JPMorgan credit research, BIS annual reports, and SEC filings.
Here's the financial architecture underneath your cloud contracts and AI API calls:
The visible debt: AI-related corporate debt has risen to approximately $1.2 trillion, representing about 14% of JPMorgan's US Liquid Index — surpassing banking to become the single largest sector in the investment-grade credit market. Corporate bond issuance by hyperscalers hit approximately $121 billion in 2025 alone, 4.3 times the annual average of $28 billion from 2020 to 2024.
The hidden debt: This is what genuinely alarmed the BIS. Moody's estimates that the five major hyperscalers — Alphabet, Amazon, Meta, Microsoft, and Oracle — have signed but not yet recognized on-balance-sheet data center leases totaling approximately $662 billion, equivalent to 113% of their combined adjusted debt. These are structured through Special Purpose Vehicles (SPVs) and financed through private credit markets with limited regulatory oversight.
The capex-to-cash crunch: In 2023, capex for the five major hyperscalers consumed approximately 40% of operating cash flow. PIMCO projects that by 2026–2027, this ratio will climb to approximately 94%, essentially reinvesting nearly all operating cash flow into AI infrastructure.
Your cloud provider is spending faster than it earns. Your AI API vendor's parent company is borrowing at historically unprecedented rates. And the entities financing all of this operate in what the BIS calls "shadow borrowing" — debt that is economically real but largely invisible on corporate balance sheets.
Why This Is Worse Than Dotcom — and Why It's Not
The Treasury report draws explicit dotcom parallels, but it also identifies critical differences that make the current situation both more resilient and more dangerous than 2000.
Three Ways AI Is More Resilient Than Dotcom
1. Revenue is real. Unlike Pets.com and Webvan, today's AI companies generate substantial revenue. Hyperscaler cloud revenue exceeded $300 billion in 2025. OpenAI reportedly crossed $20 billion ARR. The question isn't whether revenue exists — it's whether it grows fast enough to justify the infrastructure investment.
2. Balance sheets are stronger. The dotcom era was defined by speculative excess and overreliance on debt financing from companies with no revenue. Today's hyperscalers are among the most profitable companies in history. Microsoft's operating cash flow exceeded $120 billion in 2025.
3. The technology works. The internet in 1999 was still proving its use case. AI is already generating measurable productivity gains — even if those gains are smaller and more complicated than the headlines suggest.
Three Ways AI Is More Dangerous Than Dotcom
1. Deeper systemic embedding. As Treasury analysts found, AI firms are more deeply entrenched in the economy than dotcom predecessors. Every major bank, insurer, manufacturer, and retailer now runs AI workloads on hyperscaler infrastructure. A disruption to cloud services wouldn't just crash stock prices — it would halt business operations.
2. Institutional concentration. Fewer retail investors are backing AI compared to dotcom, according to the Treasury report. That means a sustained AI downturn would have a greater impact on institutional investors — pension funds, insurance companies, and banks — that are fundamental to economic stability.
3. Opaque financing. The dotcom bubble was financed primarily through public equity markets where everyone could see the prices. AI infrastructure is increasingly financed through private credit vehicles, hedge funds, and non-bank intermediaries that operate with less oversight. As BIS Asia-Pacific representative Zhang Tao warned: "The interconnectedness of the financial system and interplay of vulnerabilities could mean the speed of a correction could be much faster than previous banking crisis episodes."
The Depreciation Time Bomb: 2026–2028
There's a specific mechanism through which this trillion-dollar infrastructure bet translates into financial stress — and it's already ticking.
When a hyperscaler builds a data center, the cash goes out immediately but the expense hits the income statement gradually through depreciation. The five major hyperscalers collectively deployed over $1 trillion in AI-related capex in 2025–2026. Those assets are sitting on balance sheets, not yet depreciating. They will concentrate their transfer to fixed assets in 2026–2028.
The Wall Street Journal, citing analyst forecasts, projects that Alphabet alone will see depreciation jump from $21.1 billion in 2025 to approximately $78 billion in 2029. Morgan Stanley forecasts that four major companies will accumulate over $520 billion in depreciation over three years.
What this means for enterprise buyers: the companies providing your AI infrastructure are about to experience significant margin compression. That creates three immediate risks:
- Price increases. When margins compress, vendors raise prices. If your AI API contracts don't have price caps, expect repricing.
- Service cuts. Underperforming regions, lower-tier support, and less profitable product lines get rationalized. Your workloads may be affected.
- Consolidation. Weaker players get acquired or exit. Your vendor landscape shrinks, your negotiating power diminishes, and your switching costs rise.
The ROI Gap That Makes It All Fragile
The entire AI financial edifice rests on a single assumption: that enterprises will adopt AI at scale and generate enough revenue to justify the infrastructure investment. The Treasury report explicitly flags this risk: "AI investors are taking risks so significant that much of the financial system now rests upon AI meeting expectations for productivity gains and profitability."
The current data is mixed at best:
- Only 25% of AI initiatives deliver expected ROI, according to an IBM CEO study. Just 16% have scaled enterprise-wide.
- Only 28% of AI projects fully achieve their ROI expectations, according to Gartner's April 2026 survey of 782 infrastructure and operations leaders, with 20% failing entirely.
- Companies that replace workers with AI see no ROI improvement versus those that don't, per Gartner. Layoffs create budget room but don't deliver returns.
- 40% of Americans believe AI will be a negative societal force over the next two decades, versus 16% who believe it will be positive, per Pew Research.
- Employees spend 6.4 hours per week "botsitting" — feeding context, debugging mistakes, and cleaning up AI outputs.
Ed Yardeni of Yardeni Research ran what he calls a "capex payback test" — checking whether OpenAI and Anthropic are adding users fast enough to cover their spending commitments with the hyperscalers. His conclusion: "The AI ecosystem is not fully end-user revenue-backed yet, but it is not entirely speculative either." The math works by 2030 — if current growth forecasts hold and compute efficiency improves.
That's a lot of "ifs" supporting $1.2 trillion in debt.
Framework #1: Enterprise AI Vendor Exposure Audit
The Treasury report isn't actionable for enterprises as-is — it's written for financial regulators. But the risks it identifies translate directly into vendor management decisions. Use this framework to assess your organization's exposure.
Dimension 1: Vendor Financial Health (Score 1–5)
| Factor | Low Risk (5) | Medium Risk (3) | High Risk (1) |
|---|---|---|---|
| Revenue growth | >30% YoY, diversified | 15–30% YoY, AI-concentrated | <15% or negative |
| Capex-to-cash ratio | <50% of operating cash flow | 50–80% of operating cash flow | >80% of operating cash flow |
| Debt trajectory | Stable or declining leverage | Moderate new issuance | Aggressive debt growth, private credit reliance |
| Customer concentration | Top 10 customers <20% of revenue | Top 10 customers 20–40% | Top 10 customers >40% |
| Profitability | Positive free cash flow | Free cash flow breakeven | Negative free cash flow, burning reserves |
Dimension 2: Contract Vulnerability (Score 1–5)
| Factor | Low Risk (5) | Medium Risk (3) | High Risk (1) |
|---|---|---|---|
| Price protection | Multi-year price caps in contract | Annual price adjustment with caps | No price protection, usage-based |
| Exit provisions | Data portability guaranteed, <90-day migration | Partial portability, 90–180 days | Lock-in, proprietary formats, >180-day migration |
| SLA enforceability | Financial penalties for downtime | Credits only | Best-effort SLAs |
| Multi-vendor capability | Workloads run on 2+ providers today | Tested on alternative, not in production | Single-vendor dependency |
| Contract duration vs. vendor runway | Contract term < vendor's cash runway | Approximately equal | Contract extends beyond visible financial runway |
Dimension 3: Concentration Risk (Score 1–5)
| Factor | Low Risk (5) | Medium Risk (3) | High Risk (1) |
|---|---|---|---|
| AI spend concentration | <30% with any single vendor | 30–60% with one vendor | >60% with one vendor |
| Critical workload dependency | No critical workloads on a single AI vendor | Some critical workloads on one vendor | Mission-critical workloads, single vendor |
| Geographic concentration | Multi-region, multi-provider | Multi-region, single provider | Single region, single provider |
Scoring: Total your scores across all 13 factors (max 65). 50–65: Low exposure — maintain monitoring cadence. 35–49: Moderate exposure — begin diversification planning. 20–34: High exposure — initiate immediate vendor risk mitigation. Below 20: Critical exposure — escalate to CFO and board risk committee.
Framework #2: The AI Vendor Stress Test — 5 Scenarios Every Enterprise Should Model
The Treasury report warns about what happens if "financial conditions change, productivity goals are missed, or various choke points stymie growth." Here's how to translate those macro risks into concrete enterprise planning scenarios.
Scenario 1: The Price Shock (Probability: High — 12–18 months)
Trigger: Hyperscalers pass depreciation costs through to customers. AI API pricing increases 30–60%.
Questions to answer now:
- What is your total AI spend as a percentage of IT budget? (If you don't know, that's the first problem.)
- Which AI workloads are price-elastic — meaning you'd scale them down or eliminate them at 2x cost?
- Do you have contractual price caps? When do they expire?
Action: Run a sensitivity analysis on your AI workloads at 1.5x and 2x current pricing. Identify which use cases survive the price test and which were only viable at subsidized rates.
Scenario 2: The Vendor Consolidation (Probability: Medium — 18–36 months)
Trigger: Smaller AI vendors lose funding access as VC concentrates further. Acquisitions and shutdowns reduce your vendor options.
Questions to answer now:
- Which AI vendors in your stack have less than 18 months of runway?
- Are you using open-source models or proprietary-only? Open-source provides a floor.
- What's your model portability story? Can you swap LLM providers in days or months?
Action: Map every AI vendor to their last funding round, revenue trajectory, and burn rate. Create a migration playbook for your top 3 most at-risk vendor dependencies.
Scenario 3: The Credit Event (Probability: Low-Medium — 24–48 months)
Trigger: A major AI infrastructure financing deal fails. Private credit markets tighten for data center projects. Cloud capacity expansion slows or halts.
Questions to answer now:
- Can your workloads run on existing cloud capacity if expansion pauses?
- Do you have committed capacity reservations or are you on-demand?
- What's your plan if GPU availability tightens again?
Action: Secure reserved capacity for critical workloads. Evaluate on-premises or hybrid options for workloads where cloud dependency creates unacceptable risk.
Scenario 4: The Regulatory Shock (Probability: Medium — 6–18 months)
Trigger: The AI Bubble Transparency Act or similar legislation passes, requiring financial institutions to disclose AI exposure. Disclosure reveals concentrated positions. Markets reprice AI equities.
Questions to answer now:
- How would a 30% decline in your primary AI vendor's stock price affect their strategic direction?
- Do your contracts survive a change-of-control event?
- Is your procurement team monitoring vendor financial health quarterly?
Action: Add vendor financial health monitoring to your quarterly risk review. Include change-of-control provisions in all new AI contracts.
Scenario 5: The Slow Fade (Probability: Medium-High — 12–36 months)
Trigger: AI fails to deliver transformative productivity gains. Enterprise adoption plateaus. Vendors shift from growth mode to efficiency mode — cutting support, sunsetting products, raising prices on remaining customers.
Questions to answer now:
- What's your measured ROI on AI investments? (Not projected — measured.)
- Which AI use cases have delivered quantifiable value versus which are still "promising"?
- If AI spending budgets were cut 40%, which investments would survive?
Action: Conduct a ruthless AI portfolio review. Kill zombie pilots. Double down on proven use cases. Build your narrative for the CFO conversation that's coming — because if only 25% of AI initiatives deliver expected ROI, CFOs will eventually notice.
What the Smart Money Is Already Doing
The legislative response is accelerating. Senator Elizabeth Warren's AI Bubble Transparency Act would compel financial firms to report their debt and equity exposure to chip makers, data centers, cloud providers, and hyperscalers to the Office of Financial Research. Whether it passes or not, the direction is clear: disclosure requirements are coming.
Some enterprises are already acting:
Multi-cloud by default. Organizations that learned from the single-vendor cloud lock-in mistakes of the 2010s are applying the same discipline to AI. Running workloads across multiple providers isn't just a technical decision — it's a financial risk hedge.
Open-source as insurance. If your AI stack runs exclusively on proprietary APIs from companies carrying $1.2 trillion in debt, having open-source model alternatives isn't philosophical — it's prudent. Meta's LLaMA, Mistral, and others provide a floor that survives vendor disruption.
AI FinOps maturity. The enterprises that track their AI spend with the same rigor they apply to cloud costs are the ones that will survive budget scrutiny. If you can't show cost-per-outcome for your AI workloads, you're one earnings miss away from cuts.
Contract hardening. Price caps, data portability clauses, SLA teeth, change-of-control protections. The enterprises negotiating these terms today will be grateful in 18 months. The ones signing standard vendor paper will be trapped.
The Bottom Line
The Treasury report's most important insight isn't that AI is a bubble. It's that the financial infrastructure supporting AI is more fragile, more interconnected, and more opaque than the public narrative suggests — and that fragility flows directly into enterprise risk.
Treasury Secretary Bessent praised the $750 billion in AI investment this year and compared it favorably to the dotcom boom. His own analysts, in the same building, wrote a report explaining why that comparison should terrify everyone.
You don't need to decide who's right. You need to decide what happens to your AI strategy if the analysts are right — even partially.
The enterprises that survive the next phase of the AI cycle won't be the ones that bet the most on AI. They'll be the ones that stress-tested their bets before the stress arrived.
Run the audit. Model the scenarios. Harden the contracts. Do it now — while you still have leverage.
Continue Reading
- 43% of All VC Went to 2 Companies. Your AI Vendor Plan Is Broken.
- Your AI Vendor's New Boss: Washington's $42.6B OpenAI Stake
- 2 AI Labs Hit $1 Trillion. Your Vendor Strategy Isn't Ready.
- 100% of CIOs Budgeting for AI. Half Already Blew Their Budgets.
- AI Saves 11 Hours a Week. Workers Waste 6.4 Babysitting It.
