The A100 Contract Runs To 2029

CoreWeave Q2 2026 call, Aug 11: the company signed an A100 contract that "extends into 2029 at an attractive price. As a reminder, this SKU was introduced in 2020." Management added that prior-generation pricing sits "at or above where it was years ago" and that the older fleets are largely sold out. CEO Michael Intrator separately: "Pricing and margins for our Blackwell and Vera Rubin SKUs are setting new highs." Nebius Q2 2026, Aug 11: the company's first capacity auction cleared 15% above the highest price it has ever charged. Short-duration capacity is transacting at $40-50M per MW against $20-25M per MW on contract. Management: "We could sell our entire 2027 capacity on these terms today. We are deliberately not doing so because we see higher value in retaining some capacity for immediate customer needs." CoreWeave recently syndicated the first term loan to include shorter-duration customer contracts, priced during one of the worst credit weeks of the year, and it completed at tight spreads.

The depreciation argument has been the most cited structural bear case on the neoclouds, and it rests on a single assumption: that a GPU stops earning shortly after the generation that replaces it ships. A 2029 contract on 2020 silicon falsifies that assumption directly, and it does so with a signed counterparty rather than a model. The mechanism underneath it matters more than the headline. Asset-level leverage on these clusters amortizes inside the initial contract term. Recontracted capacity therefore sits on fully depreciated, unlevered hardware where power and some operating cost are essentially the only charges against the revenue. That is why residual value has moved from roughly zero in the bear model to materially positive, and it is why lenders financing shorter-duration deals are now, by construction, pricing future recontracting. It also explains something that happened on Monday. NVIDIA arranged a $500B financing package with Apollo, Blackstone, BlackRock's Global Infrastructure Partners, Brookfield, Goldman Sachs and KKR. Collateral with a demonstrable nine-year revenue life is underwritable in a way it was not a year ago. The financing and the contract disclosure are the same fact seen from two ends. Where this does not reach, and this is the part the bulls are skipping: extended asset life is an answer to the Layer 4 depreciation question. It is not an answer to the Layer 3 monetization question. Longer useful life makes the collateral good; it says nothing about whether the customers can pay what the pricing power is now extracting. CoreWeave's own CFO said on the same call that the company is passing component price increases down to customers, and those customers are OpenAI, Anthropic, Meta and Jane Street. Three of those four do not generate free cash flow. OpenAI carries roughly a billion users with about a tenth paying and has delayed its listing. Meta's free cash flow fell about 91% year over year to $784M while capex nearly doubled. So the correct update is not that the bear case is dead. It is that the risk has moved. The question is no longer whether the assets are worth less than the books claim. It is whether the buyers can afford what the assets have become worth.