The Bear Case Gains Weight, but the Financing Story Differs
A growing chorus of sophisticated risk analysts is sounding alarms over the artificial intelligence trade, and for forex and macro traders, the equity-market overhang they describe has direct implications for risk appetite, dollar flows, and volatility regimes. The core argument, articulated most clearly by industry veterans, is that concentration, circular financing, and a widening gap between capital expenditure and actual revenue mirror the conditions that preceded the dot-com collapse.
Fred Hickey, who has managed The High-Tech Strategist since 1987, has described the current datacenter buildout as a "more dire situation than the great fiber-optic capacity overbuilds" of 2000. Michael Burry has been tracking Nvidia's credit-default swap spreads as the chipmaker increasingly assumes the roles of banker, landlord, and equity partner to its own customers. The Bank for International Settlements identified approximately $1.65 trillion in off-balance-sheet obligations held by the largest hyperscalers — a figure exceeding their on-balance-sheet liabilities. David Cahn at Sequoia estimated the annual shortfall between AI infrastructure outlays and ecosystem revenue at close to $600 billion. Allianz quantified the capex-to-revenue divergence at roughly 46%, well above the 32% threshold that characterized the 2001 telecom bust. An MIT study published in August found that the majority of corporate AI pilots had yet to generate any measurable top-line contribution.
A Bank of America survey revealed that 54% of professional fund managers now label AI a "bubble." The ten largest equities collectively represent approximately 43% of the S&P 500, an all-time record that dwarfs the 27% peak recorded at the height of the dot-com era. The equal-weight index has already begun to decouple from the cap-weighted benchmark — a structural divergence that historically precedes megacap weakness. For FX desks, this concentration means that a sharp AI-sector correction would transmit quickly into equity-beta currencies, yen carry unwinds, and dollar risk-premium repricing.
The Circular Financing Problem and the Depreciation Cliff
The circular-financing critique deserves serious attention. When a chip manufacturer takes an equity position in a client that then commits to purchasing that manufacturer's products, a portion of what appears as organic demand is actually the seller financing its own sales channel. This distorts the true demand signal and warrants the scrutiny it is receiving.
Compounding the issue, trailing capital expenditure of roughly $434 billion now substantially exceeds the $149 billion in depreciation flowing through income statements. That gap will materialize as a real cost burden between 2027 and 2029, regardless of whether the associated revenue materializes on schedule. The AI bears did not fabricate these figures; they extracted them directly from corporate filings.
Why the 2000 Analogy Breaks Down: Who Is Writing the Checks
The critical distinction, and the one that separates a manageable drawdown from a systemic collapse, is the source of financing. In the run-up to the 2000 fiber-optic overbuild, the money came from leveraged entities with no organic capacity to service their obligations. WorldCom, Global Crossing, and a wave of upstart long-haul carriers strung cable on debt, while vendor financing from Lucent and Nortel extended credit to customers who could not repay. When revenue failed to materialize, those balance sheets had no cushion, and the structure cascaded into bankruptcy proceedings.
The current buildout operates on a fundamentally different axis. Approximately two-thirds of projected 2026 capital expenditure is funded directly from the operating cash flow and equity reserves of Microsoft, Alphabet, Amazon, and Meta — four of the most cash-generative enterprises in corporate history. The existing debt component is investment-grade and represents a minority of total spending. The balance sheets absorbing this cycle are not those of WorldCom, and that single difference alters the failure mode entirely.
Revenue is also arriving, albeit on a lag. AI inference workloads now clear roughly 70% gross margins. Microsoft's AI segment has surpassed a $37 billion run rate. Amazon's AI-related revenue is expanding at triple-digit pace. Anthropic reportedly scaled from approximately $9 billion to a $47 billion run rate within a single twelve-month period. Even Nvidia, the bears' primary target, has seen forward earnings grow rapidly enough that its valuation multiple has actually compressed as fundamentals caught up with prior expectations — the inverse of Cisco in 2000, which peaked near 30 times sales before earnings evaporated.
The Binding Constraint Is Power, Not Demand
The bears predict a capacity glut. The data suggests the opposite. Microsoft is reportedly holding approximately $80 billion in Azure orders it cannot fulfill due to electricity shortages, with GPUs sitting idle in inventory awaiting grid connections. More than 60% of the datacenter capacity slated for 2027 has not yet broken ground. If demand is rationed by the power grid rather than by customers retreating, the market faces a shortage, not a surplus.
That said, the strongest bear objection holds: constructing two to three years' worth of generation and transmission infrastructure today could convert this year's scarcity into next year's oversupply. That is a legitimate timeline risk, but it is a 2028 question, not a 2026 one.
Positioning Discipline for Traders Who Cannot Afford to Call the Top
The strategist's central message for investors — and by extension for FX traders whose portfolios are correlated to equity beta — is not to abandon the AI trade wholesale, but to manage it with discipline. Nvidia has experienced drawdowns of 55% or more on five separate occasions since 2000 and recovered to new highs after each. Investors sized to endure those corrections benefited enormously. The prescription is to hold AI exposure at a weight where a 50% drawdown is painful but not portfolio-destroying.
Not all AI exposure carries identical risk. The self-funding hyperscalers and the power-infrastructure bottleneck resemble 2000 the least. The leveraged periphery — companies dependent on vendor financing or speculative revenue projections — resembles 2000 the most, and those are the first positions to trim when the narrative wobbles. Selling the profitable compounders simply because a bear called a peak is how investors forfeit years of compounding while awaiting a crash that arrives late or not at all.
The index itself has quietly become the bet. With the ten largest names accounting for nearly 43% of the S&P 500, passive index ownership is a concentrated wager on the same handful of companies made without anyone actively deciding it was wise. For FX traders, this means that broad "buy the market" strategies carry embedded AI-sector risk that must be priced into currency positioning.
Consensus risk is also elevated. "Long the Magnificent 7" has ranked as the most crowded trade on Wall Street for nearly two years, and with more than half of managers now calling AI a bubble, the consensus has already tilted. As Bob Farrell's Rule #9 reminds us: "When all the experts and forecasts agree, something else usually happens." That does not invalidate the bear case, but it suggests the obvious crash may refuse to arrive on the obvious schedule.
Howard Marks offered a sobering counterpoint: "Being too far ahead of your time is indistinguishable from being wrong." Notably, Hickey himself holds his AI-bear thesis at roughly 1% of his portfolio in put options — a hedge, not a conviction short. The posture worth adopting for any multi-asset book is to own the compounders, hedge the tail, and let revenue prove or disprove the thesis on the tape.
The question that actually matters for the next 12-to-24 months is not whether excess exists — it plainly does — but what happens when the richest companies on earth overbuild with their own money rather than borrowed money. That is the variable the 2000 analogy cannot answer, and it is the one that will determine whether this episode produces a managed drawdown or a systemic event with spillover into global currency markets.