Fiat Elpis · AI infrastructure credit note 05

CoreWeave’s credit can work even when the equity disappoints

The likely failure channel for a top-tier neocloud is no longer an imminent debt default. It is equity dilution, poor return on capital or a long wait for free cash flow.

Saâd Filali Khattabi 21 August 2026 9 min read

From 300 recent @FiatElpis posts · expanded with primary-source research

Credit and equity do not need the same outcome. Contracted revenue can protect lenders and sustain the AI infrastructure build while common shareholders absorb funding costs, dilution and uncertain terminal returns.

01 · The information delta

Default risk fell; capital intensity did not

CoreWeave’s 2025 filing reported $60.7 billion of remaining performance obligations with a weighted-average contract duration of roughly five years. Its March 2026 10-Q showed $25.1 billion of debt, $8.8 billion of undrawn availability and an $8.5 billion delayed-draw facility tied to a customer contract.

The structure matters. Much of the infrastructure is financed against take-or-pay commitments, assets and customer demand rather than unsecured hope. In April, CoreWeave added a multi-year Anthropic agreement; management has also said older A100 capacity is contracted into 2029, evidence that prior-generation GPUs may retain revenue value longer than the fastest depreciation bear case assumes.

None of that makes the stock cheap. It makes the cash-flow waterfall easier to distinguish: customers pay for capacity, lenders sit ahead of equity, and shareholders own whatever return remains after interest, depreciation, capex and future financing.

“The credit looks much safer than the equity.”
The original Fiat Elpis market note on X

02 · Signals to track

Why lenders and shareholders see different companies

Backlog

Contracts support debt service

Multi-year take-or-pay revenue gives asset-level lenders a clearer repayment path than a merchant cloud with short, cancellable demand.

Residuals

Older GPUs are still monetizable

An A100 contract extending into 2029 reduces the risk that useful collateral instantly becomes stranded when a new NVIDIA architecture ships.

Interest

Coverage can improve before free cash flow

Adjusted EBITDA can cover cash interest even while enormous investment spending keeps reported free cash flow negative.

Equity

Growth can be financed at shareholders’ expense

Convertible debt, stock issuance and repeated capital raises can preserve the enterprise while reducing the per-share value of the eventual payoff.

The failure channel

A company can avoid default, build the capacity and still produce a disappointing equity return. Solvency asks whether promised cash reaches creditors. Equity asks whether the assets earn more than their full cost after dilution and reinvestment.

03 · What may be mispriced

Neocloud risk should be tiered, not generalized

CoreWeave has strong counterparties, long contracts, substantial financing access and scale. Those features do not automatically extend to smaller merchant neoclouds with weaker customers, less prepayment and speculative power commitments.

Nebius offers a useful contrast. Its second-quarter 2026 materials reported a 50% adjusted EBITDA margin in the AI cloud business, more than $9 billion of expected 2026 customer prepayments and a first secured financing at SOFR plus 2.5%. That strengthens the top-tier credit case while still leaving equity holders exposed to $5.7 billion of quarterly capex and capital-market choices.

Match each debt facility to the customer contract and assets that support it.
Separate adjusted EBITDA from cash after interest, leases and growth capex.
Track backlog concentration, cancellation rights and customer prepayments.
Model dilution per share, not only enterprise revenue growth.

04 · What would change my mind

What would put the credit thesis at risk

The safer-credit view depends on contracts converting into timely cash:

A major customer delays deployment, disputes delivery or exits a commitment.
Data-center construction slips while debt service and lease obligations continue.
Older GPU renewal pricing falls much faster than depreciation schedules assume.
Financing markets close before contracted capacity becomes billable.
Interest coverage deteriorates despite revenue growth.

Bottom line

The AI build can survive an equity disappointment

The top-tier neoclouds increasingly look like financeable infrastructure platforms rather than imminent credit events. That is supportive for the broader AI trade because it reduces the chance of a sudden debt-market stop.

For the common stock, however, the hurdle is higher: contracts must convert on time, assets must retain value, and the return after every financing layer must accrue per share. Credit safety is a floor for the ecosystem, not proof of attractive equity economics.

Sources & method

Primary sources, thesis separated from fact

This note expands themes from the author’s recent X posts. Reported facts are linked to their sources; market interpretation is explicitly the author’s view. Market levels may change after publication.