Nvidia Tightens Compute Revenue-Sharing Agreements, Proactively Advancing "De-Circularity" Governance
nashnova research
Nvidia suspended the most controversial revenue-share arrangements within its AI compute partnership program, cutting back the loop where it finances customers to buy its own chips. This means → demand is not cooling, but the credit structure and antitrust boundaries around this business are being redrawn.
What exactly was suspended?
Not the entire compute partnership program — only the bundled deals where Nvidia simultaneously sold GPU clusters, backstopped unsold capacity, assisted financing, and took 50% of revenue above a threshold.
In plain terms = Nvidia was seller, guarantor, and revenue-share partner all at once. That triple role is what got paused.
Some preliminary agreements also required cloud providers to lease capacity only to Nvidia-approved customers, and Nvidia preferred spreading capacity across many small firms rather than concentrating it with one large buyer — a condition that drew strong pushback from major potential partners.
Nvidia employees internally flagged concerns that the arrangement could trigger antitrust scrutiny.
Is demand actually cooling?
No. Nvidia's FY2027 Q2 revenue rose 106% year-over-year to $96.221 billion; data-center revenue climbed 117% to $89 billion.
Q3 revenue guidance stands at $108 billion (±2%). Vera Rubin — Nvidia's next-generation GPU platform — is in full-scale production. Management projects ~70% revenue growth in FY2028, noting the figure is already constrained by memory-chip and TSMC advanced-node capacity.
Anthropic signed a six-year, $45 billion compute procurement deal with Nscale, locking in roughly 460 MW of power capacity for Vera Rubin deployments. This means → AI infrastructure capex is still accelerating, not slowing.
Raymond James analyst Simon Leopold raised Nvidia's price target from $352 to $550, implying a market cap of roughly $13 trillion.
How large are the off-balance-sheet commitments?
The $530.5 billion off-balance-sheet figure breaks into three parts: $366 billion in future commitments + $56 billion in additional AI-cloud and third-party data-center commitments + $108.5 billion maximum notional guarantee exposure.
Of that, $279 billion covers memory and manufacturing supply commitments — some of which can be cancelled, rescheduled, or adjusted. The $105 billion OpenAI-related guarantee phases in from FY2029 across nine construction stages.
In plain terms = the $530.5 billion is not a lump-sum liability due now; it is a staggered, conditional, partly revocable set of forward arrangements.
How should the cash-flow signal be read?
Accounts receivable rose from $38.466 billion to $63.059 billion; the top five customers account for roughly 70%.
Q2 operating cash flow was about $24.077 billion, well below $59.688 billion in GAAP net income. This means → there is a gap between paper profit and cash actually collected.
The gap stems mainly from equity-investment gains, inventory build-up, and investment-grade customers extending payment terms — it cannot be equated directly with credit deterioration.
Who faces real pressure?
TSMC-led capacity chains, high-bandwidth memory, advanced packaging, and hyperscaler-driven orders — these carry relatively high certainty still.
The real pressure falls on two groups: smaller, emerging cloud providers that rely on Nvidia's credit endorsement and lack long-term offtake contracts or stable cash flow; and the highly leveraged financing chains built around speculative data-center campuses.
This reflects the true nature of Nvidia's pullback: not a demand inflection, but a risk-boundary reset under triple pressure from circular-financing controversy, antitrust scrutiny, and customer pushback over operational autonomy.
Whether the revised program can strike a workable balance between antitrust boundaries and commercial control is the key verification point for the market going forward.
市场有风险，内容仅供研究参考，不构成投资建议。