Are we in an AI bubble�or just the most capital-intensive tech buildout in a generation? A recent discussion highlighted how analysts historically spot bubbles using multi-factor checklists? It�s a timely frame: the AI story now spans trillion-dollar market caps, a $7 trillion data center race, and early signs that unit economics aren�t yet matching the narrative?
The credit tell: AI�s infrastructure is being built on long-duration debt
The most concrete signal is in the bond market? Big Tech is leaning on investment-grade debt to finance an unprecedented wave of data center construction? One estimate pegs global AI-related data center investment at up to $7 trillion by 2030? Companies have issued bonds maturing in the 2050s and 2060s�effectively asking long-term savers to underwrite the AI era?
Investors are starting to flinch? Risk premiums on some issuers�particularly those with heavier balance sheets�have climbed to multi-month highs? Oracle, for example, holds roughly $110 billion of debt�about triple its annual EBITDA�and is rated BBB, near the bottom of investment grade? That doesn�t spell crisis today, but it does mean the cost of capital can rise quickly if sentiment turns?
Valuations vs? reality: the Damodaran caution
Market veteran Aswath Damodaran warns that AI equity prices may be running ahead of plausible cash flows, spotlighting Nvidia as the poster child? His blunt read: �The AI story is wildly overbought� The valuation of Nvidia, in particular, is mad: it implies trillion dollar revenues and 8 percent gross margins out to the far horizon?� He also argues that if the �Magnificent Seven� stumble, correlations could spike and pull down the broader market, reducing the diversification refuge investors expect?
Yet the countercase remains strong: U?S? growth has held up, several large-cap tech names trade at more grounded multiples, and AI demand is spilling into the real economy? One example: industrial suppliers are posting gains tied to data center orders, underscoring that this cycle isn�t just software-on-software�it�s steel, turbines, and power?
Follow the cash: unit economics under stress
If debt and valuations are the market-level tells, unit economics is the operator-level truth? Leaked documents suggest OpenAI shared roughly $493?8 million with Microsoft in 2024 and $865?8 million in the first three quarters of 2025 under revenue-share terms? Based on a 20% split, that implies at least $2?5 billion in 2024 revenue and $4?33 billion through Q3 2025? But inference compute costs clocked in around $3?8 billion in 2024 and $8?65 billion in the first nine months of 2025�indicating some frontier AI businesses may still be spending more to serve queries than they collect in cash revenue?
OpenAI declined to comment; Microsoft didn�t respond to requests? CEO Sam Altman has said revenue is �well more� than $13 billion annually and could top a $20 billion run rate by year-end, aiming for $100 billion by 2027? Ambition is not in short supply? Cash conversion might be?
Venture rules rewritten: froth or discipline?
At the early-stage edge, venture capitalists say the playbook has changed? Some AI startups have jumped from zero to $100 million in a year�fuel for bubble claims? At the same time, Series A investors are pushing late-stage rigor onto seed rounds: demanding real go-to-market muscle, proprietary data advantages, and technical depth? As DVx Ventures� Jon McNeill put it, �breakout companies� don�t have the best tech? They have the best go-to market?� Kindred Ventures� Steve Jang countered that strong technology and distribution are both necessary to endure?
So, bubble�or boom with risk?
Here�s the balanced read? The �bubble checklist� lights up on multiple fronts: leverage-fueled spend, grand narratives, and pockets of weak unit economics? But unlike dotcom, today�s AI demand is capital-heavy and bleeding into heavy industry and utilities? Investment-grade balance sheets are carrying much of the load, and some Big Tech fundamentals remain robust?
What business leaders should watch next
- Debt spreads and issuance: Rising risk premiums�especially for long-duration bonds�will signal how much more AI infra lenders can stomach?
- Cloud capex guidance: Watch hyperscalers� quarterly capex and power procurement? Delays or cuts would ripple through chip, power, and industrial supply chains?
- Inference gross margins: Track reported AI service margin disclosures from cloud providers and model vendors�this is where hype meets math?
- Order books across the stack: From power equipment to networking gear, broad-based backlog softening would flag demand normalization?
The bubble debate isn�t academic? If you run an enterprise AI program, your cost of compute�and the durability of your vendors�hinges on these trends? If you manage a balance sheet, your pension or insurer may already be financing the data centers that make generative AI possible? The question isn�t whether AI matters? It�s whether the capital stack behind it can earn its keep before the bill comes due?