Weekly NVIDIA Circular Financing Risk Monitor 2026-06-08

Key Changes This Week

No sudden systemic risk events this week, but structural risks continue to accumulate. NVIDIA’s Q1 FY2027 earnings (disclosed May 20) posted record $81.6B revenue, up 85% YoY, with surface-level metrics looking strong. However:

1. CoreWeave long-term debt surged to $17.53B (as of March 2026), up 254% YoY, with Q2 interest expense projected at $650-730M. Stock dropped on lowered growth guidance.
2. OpenAI Q1 loss of $6.95B (non-GAAP), operating margin of -122%, ChatGPT weekly active user growth stalling at 905M.
3. NVIDIA’s investment-purchase loops with CoreWeave and OpenAI continue to deepen, further entrenching circular financing structures.

Overall risk rating: ⚠️ Medium-High (↑ slight increase)

1. NVIDIA Related-Party Transactions & Vendor Financing

This Week’s Developments

Key Structural Analysis:

• NVIDIA → invests in OpenAI → OpenAI buys NVIDIA chips: The most classic circular path. NVIDIA invests $10B in non-voting OpenAI shares; OpenAI then uses those funds to procure NVIDIA GPUs (via Azure, AWS, etc.). Fortune identified this as vendor financing as early as September 2025.
• NVIDIA → invests in CoreWeave → CoreWeave buys NVIDIA GPUs → NVIDIA backstops unused capacity: Double loop — both investment and capacity commitment. A significant portion of CoreWeave’s $17.5B debt was used to purchase NVIDIA chips.
• NVIDIA issued a memo in November 2025 denying circular financing, but two prominent short sellers publicly disagreed. The May 2026 annual report mentioned reviews of related-party transactions for potential conflicts of interest.

Risk Assessment

2. OpenAI Financial Health

Core Metrics

Key Findings

1. Losses accelerating: $695M Q1 loss (non-GAAP, excluding stock-based comp); GAAP likely higher. HSBC projects OpenAI won’t be profitable before 2030, with a remaining $207B funding shortfall.

2. Growth ceiling approaching: ChatGPT weekly active users at 905M, below the expected 1B target. Paid conversion rate only 6%, partly driven by cheap Go subscriptions ($5-8/month). Free annual Go subscriptions given to all Indian subscribers in October 2025 — growth quality questionable.

3. Funding “inflated”: Of the $122B headline raise, only ~$37B actually arrived at close. SoftBank, Amazon, NVIDIA commitments are tranches with conditions.

4. Compute commitments astronomical: $25B Azure + $38B AWS + $90B AMD + $35B Broadcom custom accelerators. Stargate project at $500B/4 years. OpenAI’s own revised estimate: $600B in compute spending by 2030.

Risk Assessment

3. GPU Cloud Provider Health

CoreWeave (CRWV)

Core Risks:

• Most of the $17.5B debt was used to purchase NVIDIA GPUs, which depreciate rapidly with each technology generation
• Quarterly interest of $650-730M ($26-29B annualized) requires extremely high utilization rates to cover
• Component cost inflation may further raise 2026 capex (WSJ, May 7)
• Circular path: NVIDIA invests CoreWeave → CoreWeave buys NVIDIA GPUs → NVIDIA backstops unused capacity → CoreWeave uses GPU revenue to service NVIDIA-linked obligations

Lambda Labs

Core Risks:

• Still loss-making, requires continuous pre-IPO funding
• Heavily reliant on NVIDIA GPU supply chain
• Faces same industry cycle risks as CoreWeave

4. Circular Financing Risk Signals

Loop Structure

NVIDIA ──Invests $11B+──→ OpenAI / CoreWeave / Nscale / Nebius
  ↑                          │
  │                          │ Uses investment to purchase
  │                          ↓
  │                     GPU chips / cloud capacity
  │                          │
  │                          │ Generates revenue
  │                          ↓
  │                    Repurchases NVIDIA chips (cycle)
  │                          │
  └──── Revenue growth / profit ←──────┘

New Signals This Week

Historical Comparison: Telecom Bubble Parallel

Tomasz Tunguz (Oct 2025) compared NVIDIA’s $11B vendor financing strategy to Lucent/Nortel’s playbook during the telecom bubble:

• Similarities: Vendor provides financing to customers to promote purchases, revenue recognition is accelerated, true demand is masked
• Differences: NVIDIA’s financial position is far stronger than Lucent’s was, and AI demand has real-world applications
• Key distinction: NVIDIA doesn’t bear customer default risk (investments are equity), but investment targets’ value is highly dependent on NVIDIA chips

5. Competitor Custom Chip Progress

Microsoft Maia 200 (Announced Jan 2026)

Impact: If Maia 200 delivers as claimed and Azure deploys at scale, it could reduce Microsoft’s reliance on NVIDIA GPUs. Microsoft is one of NVIDIA’s largest customers — this signal warrants close monitoring.

Google TPU v6 (Trillium)

Impact: Google uses TPUs extensively for Gemini training/inference internally and is expanding external sales. TPU v6’s inference cost advantage may divert some NVIDIA demand.

Amazon Trainium2

Impact: Amazon uses Trainium to provide compute to Anthropic while reducing its own NVIDIA procurement dependency.

Summary Assessment

Barclays projection: By 2026, inference will account for over 70% of general AI compute needs — roughly 4.5x more than training. Inference chip competition will directly impact NVIDIA’s market share.

Overall Risk Assessment

Risk Matrix

Overall Rating: ⚠️ Medium-High (↑ slight increase)

Core Conclusions

1. NVIDIA’s fundamentals remain strong (Q1 revenue $81.6B, +85% YoY), but its growth is increasingly dependent on “invest and sell back to yourself” circular structures. Of the $11B+ in client investments, OpenAI and CoreWeave are the two largest exposures — and both are deep in the red.

2. OpenAI is the single largest risk point. Q1 non-GAAP loss of $695M, operating margin of -122%, ChatGPT growth stalling — any deterioration could trigger a chain reaction. The $122B funding round seems like ample ammunition, but HSBC’s projected $207B gap means OpenAI needs more capital — which ultimately flows back to NVIDIA chip purchases.

3. CoreWeave’s $17.5B debt is a ticking time bomb. Quarterly interest of $650-730M, debt/equity ratio of 738%, GPU collateral facing technology depreciation. If GPU utilization drops or compute demand slows, CoreWeave’s debt servicing capacity will deteriorate rapidly.

4. Custom chip progress demands close attention. Microsoft Maia 200, Google TPU v6, Amazon Trainium2 are all advancing. With inference accounting for 70%+ of compute demand, inference chip competition will be fiercer than training chip competition. If major customers successfully develop alternatives, NVIDIA’s vendor financing loop loses a critical pillar.

5. Key monitoring metrics for Q2-Q3 2026:

   • CoreWeave Q2 financials (interest coverage, client utilization)
   • OpenAI next-quarter loss trajectory
   • Any new NVIDIA client investments/guarantees
   • Microsoft Maia 200 deployment announcements