What this page shows
NVIDIA's quarterly filing reports $36 billion of AI-cloud agreements as of July 26, 2026. Partners buy its infrastructure; NVIDIA commits to cloud services. Partners can instead sell that capacity to other customers at better rates, reducing NVIDIA's commitments as capacity is used. NVIDIA says the arrangements help providers meet demand constrained by infrastructure availability.
A separate, earlier disclosure makes the overlapping roles concrete. CoreWeave's September 2025 filing describes NVIDIA as its GPU supplier and shareholder, alongside a $6.3 billion initial order covering residual unsold capacity through April 2032, subject to service requirements and termination provisions.
The analytical implication: distinguish independent customer demand from supplier-supported capacity commitments. These disclosures alone do not establish fabricated demand, improper accounting or an eventual loss.
Recorded evidence
Limitations & review notes
- Research draft. Conclusions may change; review the sources and limitations before relying on them.
- Restored archival research draft: not independently reverified for this republication and not actively monitored. Check the linked originals and dated scope before relying on it.
- The $36 billion is a dated commitment balance, not money already paid or a forecast loss. Do not add the earlier $6.3 billion to it: the reviewed disclosures do not reconcile their overlap.
- CoreWeave's agreement has conditions; it is not an unconditional guarantee of every dollar of future revenue. Neither disclosure is a newly uncovered secret.
SuperSmart← Back to DiscoverDiscover