One River CIO Calls AI Debt Cycle a Late-Stage Ponzi in the Making
One River CIO Eric Peters is drawing an explicit CDO parallel to the AI infrastructure debt cycle, pointing to a collapsed $30B Australian neocloud IPO and a 150 bps rise in Anthropic's cost of capital as early crack signals.
Eric Peters draws the 2007 CDO parallel as a $30B neocloud IPO collapses and Anthropic's borrowing cost climbs 150 basis points in four months.
Key takeaways
- Firmus Technologies, an Australian neocloud backed by Nvidia and Blackstone, pulled a planned ~US$30 billion IPO in early October 2026 after failing to attract sufficient buyer demand, a data point Peters explicitly compares to the Bear Stearns hedge fund failures of summer 2007.
- Anthropic's cost of capital on a comparable chip-financing deal rose from 8.5% in June 2026 to over 10% in October 2026, a 150-plus basis point move in under six months, as CoreWeave's project-finance bonds trade below 80 cents on the dollar.
- Eric Peters, CIO of One River Asset Management, whose Weekend Notes letter is read by professionals deploying over $1 trillion in alternatives, is now explicitly calling this a "classic late cycle" with a self-reinforcing reversal underway.
Eric Peters, CIO of One River Asset Management, is calling the AI infrastructure build-out a late-cycle Ponzi dynamic in his latest Weekend Notes letter, drawing a direct structural parallel to CDO issuance in 2006 and 2007. The signal he's pointing at is specific: the marginal deal is already failing, and cost of capital is already rising.
The letter, distributed privately and published by ZeroHedge, cites three data points in sequence. Firmus Technologies, an Australian neocloud backed by Nvidia and Blackstone, pulled a planned A$44 billion (~US$30 billion) IPO in early October 2026 after failing to attract sufficient investor demand. It had only 46 MW of capacity running against a 912 MW pipeline and a $37.7 billion build plan. Peters writes that the collapse is "the type of sign we saw when the Bear Stearns hedge funds failed" in summer 2007.
Concurrently, Anthropic offered Blackstone a yield above 10% on a $9 billion junior tranche of a $60 billion debt package (Project North Star), first reported by the Financial Times and picked up by Reuters via Yahoo Finance. Four months earlier, Anthropic closed Project Big Sky at 8.5%. That is a 150-plus basis point rise in AI private credit cost in under six months. CoreWeave's 20 project-finance data center bonds, totaling roughly $45.6 billion across approximately 4 GW of capacity, are now trading below 80 cents on the dollar, per HyperDensity.
The CDO Parallel Peters Is Making
Peters's framing is precise. In every large credit cycle, the last paper issued is the most complex and the hardest to price. "Who's going to buy the last CDO? Who's going to buy the last CDO-squared?" he writes. "The cost of capital starts going up. This is how it always ends."
AI infrastructure lease-backed debt fits that description cleanly. The collateral is a five-year compute lease with a company that has never turned a profit. The marginal buyer of that paper at 10% yield is making the same bet subprime ABS buyers made at 7% in 2006.
When that buyer blinks, the flywheel reverses. "The greater fool theory starts to reverse and self-perpetuates," Peters writes. "When the process starts in earnest, it becomes self-fulfilling, irreversible."
The 150 bps move in Anthropic's cost of capital is the number the market is not pricing clearly. That spread widening, against the backdrop of CoreWeave bonds already in distress, means the private credit machine backing AI compute is repricing risk in real time. Peters's point is that once that repricing reaches the weaker players, it pulls the hidden support from the broader structure. "Once the cost of capital rises for the weaker players, you crack the hidden pillars supporting the edifice."
This dynamic connects directly to the broader concern about casino-level risk assumptions embedded underneath institutional-grade paper in the AI chip financing edifice.
What This Means for Sound Money
The AI infrastructure cycle is debt-financed, rate-sensitive, and backstopped by the same private credit expansion, Blackstone, Apollo, Broadcom, that is indirectly plugging gaps the Treasury market can no longer fill cleanly. When the weakest links in that chain start failing and cost of capital rises 150 bps in four months for a company as prominent as Anthropic, it signals strain in the broader private credit machine. That is the same machine operating in the shadow of the sovereign debt spiral.
Peters draws the portfolio conclusion directly. He frames the moment as one where investors no longer need to fear the FOMO move that rips higher and makes them look foolish, and instead can position for what comes after. Peters writes: "You can build a portfolio for the sacking and looting ahead." That is a ceiling call on risk assets from a CIO whose letter reaches professionals deploying over $1 trillion in alternatives. When that cohort starts publishing "sacking and looting" portfolio positioning, the rotation away from levered risk is already underway.
Bitcoin, with a fixed supply and no counterparty risk, becomes more relevant as that rotation accelerates. Fiat credit systems are structurally compelled to run these cycles to their terminal conclusion because the incentive to issue one more instrument, one more CDO-squared, one more lease-backed compute bond, never stops until the marginal buyer disappears. The Firmus IPO failure is evidence the marginal buyer is already stepping back.
What to Watch
The falsifiable version of Peters's thesis: if Anthropic closes Project North Star at or below the offered 10% yield, CoreWeave's bonds recover above par, and the Firmus pipeline gets refinanced and relaunched within 12 months, the marginal buyer is still present and the cycle has more runway. A meaningful Fed pivot that reopens the private credit window would also reset the clock. Watch those three data points. Until one of them flips, the structure Peters is describing continues to lose its fasteners.
Sources
- One River Asset Management, Weekend Notes archive
- Reuters / Yahoo Finance: Anthropic offers Blackstone double-digit yield on Project North Star
- HyperDensity: CoreWeave and neocloud bond distress data
- Newcomer: Investors turn on data center builders, Firmus IPO collapse
- NewsBytesApp: Anthropic Project Big Sky at 8.5%, June 2026
- First reported by the Financial Times (Anthropic / Project North Star, October 16-17, 2026)
- First reported by the Financial Times (Firmus IPO collapse, October 2026)
Frequently Asked Questions
What is a neocloud, and why are their IPOs failing now?
A neocloud is a capital-intensive data center operator that leases AI compute capacity, typically GPU or TPU clusters, to AI companies and enterprises rather than owning the workloads itself. They are pure infrastructure plays, dependent on long-term lease commitments from tenants whose own economics are unproven. Firmus had 46 MW running against a 912 MW build plan and could not attract buyers for a $30 billion IPO because investors are no longer willing to fund the gap between current revenue and promised scale at the valuations being asked.
How does the AI infrastructure debt cycle compare to the 2007 CDO collapse?
The structural parallel is in the collateral and the complexity. CDOs in 2006-2007 were backed by mortgages whose ability to service the debt depended on home prices continuing to rise. AI infrastructure bonds are backed by compute leases whose ability to service the debt depends on AI companies continuing to expand capacity, raise capital at declining rates, and eventually turn profitable. In both cases, the paper is rated on assumptions that are already cracking at the margin, and the last securities issued carry the most risk while appearing the most structured.
What does rising AI infrastructure cost of capital mean for Bitcoin?
Rising cost of capital in AI private credit is a signal that the private credit machine is under strain. That machine is not separate from the sovereign borrowing apparatus; it is an extension of it. As levered risk assets reprice and the greater-fool bid reverses, non-sovereign hard assets with fixed supply and no counterparty risk become more attractive as a destination for capital rotating defensive. Peters is not making a Bitcoin call, but the dynamics he describes are precisely the environment where Bitcoin's fixed supply serves a practical purpose, not merely a philosophical one.
