AGI or bust: Why trillions of dollars are leaning on an uncertain finish line
AI is being financed like a sure thing. We're talking $2.9tn projected for datacentres, a chip supplier trading above $4tn, and nine-figure hiring packages. The pitch is simple: AGI delivers laborless profits across white-collar work, and the capital stack gets paid.
If that promise slips, the downside hits portfolios, credit markets, and growth. If it lands, the winners look obvious. Either way, positioning matters more than predictions.
The thesis driving the capital
Several top investors say the current spend only makes sense if AGI arrives. Others warn we could hit a wall: progress stalls, solutions don't show up on schedule, and the financing built on smooth capability curves unwinds. The base case is still forward progress-but not necessarily on investor timelines.
The scaling bet-and its limits
Most money today is placed on scaling transformers. More chips, more data, more datacentres. If AGI needs a different approach, that spend becomes a very expensive ladder to the moon-tall, but not enough to get there. That timing risk is the core financing risk.
Where the leverage is building
- Datacentres: About $2.9tn through 2028, with roughly half funded by hyperscaler cash flow. The rest is private credit, investment-grade bonds, high-yield, and asset-backed structures tied to DC rents. Borrowing has already surfaced at scale, with names tapping private credit and securitization.
- Credit signals: Rising credit default swaps on select issuers, high-yield exposure from DC operators, and broad participation across shadow banking. This creates multiple channels for contagion if cash flows miss.
- Equities concentration: The "Magnificent 7" represent over a third of the S&P 500's value. Warnings from central banks and global institutions highlight valuation risk if AI delivery trails expectations. See recent notes from the Bank of England and the IMF's Global Financial Stability Report.
- Circular flows: Vendor-client loops (e.g., chip providers investing in model developers that buy their chips) add reflexivity. Great on the way up, disorderly on the way down.
If AGI underdelivers
- Debt contagion: Stress could hit IG bonds, HY, private credit, and ABS simultaneously. Refinancing risk rises into 2027-2029 if utilization or pricing disappoints.
- Equity drawdowns: Multiple compression for firms priced on AI growth, not revenue realized. Concentration risk amplifies the move.
- Macro spillovers: A pullback in AI-led capex and productivity narratives would dent growth assumptions and ripple through correlated assets.
The bull case that doesn't need "true AGI"
Generative AI can still rewrite ad markets, search, software, and enterprise workflows. Capex looks big, but other sectors spend comparable amounts each year. You don't need human-level AI for strong unit economics-just products that people and businesses actually pay for at scale.
Portfolio playbook for finance leaders
Public markets
- Scrutinize AI revenue attribution vs. narrative. Separate GPU-scarcity trades from durable cash flows.
- Track inference cost curves, pricing power, and model-release cadence. Capability plateaus are a red flag for multiples.
- Consider hedges where earnings are most levered to AI cycles; pairs where suppliers outrun customers' realized revenue.
- Monitor concentration: sizing rules for mega-cap exposure and scenario tests for a 20-30% multiple reset.
Credit
- Map the capital stack of datacentre operators and AI infra plays. Are contracts take-or-pay? What's the termination language?
- Check covenants, amortization profiles, and refi windows. Model 60-70% utilization scenarios and slower price declines.
- For ABS tied to DC rents, look at tenant concentration, renewal risk, and pass-through clauses for power and upgrades.
- Watch CDS for names linked to DC buildouts and vendors with circular financing exposure.
Private markets
- Demand evidence that gross margins hold after inference costs, revenue sharing, and support. Aim for clear payback periods, not vanity metrics.
- Favor customers with budget authority and clear displacement of existing spend (ads, search, software seats, support).
- Stress-test stranded asset risk if architectures shift or model efficiency leaps reduce compute needs.
Risk and governance
- Run a 12-24 month "AI delay" scenario across equities, credit, and real assets. Add a utilization shock and a 200-300 bps spread widening.
- Cap position sizes where vendor-client reflexivity is high. Limit counterparty loops within the same AI stack.
- Link incentive comp to realized AI revenue and margin, not just announcements or partnerships.
Signals worth tracking in 2026
- Datacentre disclosures: utilization, pre-lease rates, and cancellation data.
- Inference pricing and cost-per-token trends vs. demand growth.
- Credit spread moves on AI-linked issuers; changes in private credit terms and covenant packages.
- Management language: from "build it and they will come" to concrete contribution margins and churn.
Bottom line
The upside case is real, but the financing is fragile. Treat AGI as optionality, not a base case. Pay for cash flows, test the capital structure, and size positions like timelines can slip.
Need practical ways to capture value now?
For teams building near-term productivity gains and ROI without betting on AGI timelines, see curated AI tools for finance.
Your membership also unlocks:
