The GPU Boom’s Hidden Mechanics: Understanding Circular Financing in AI Infrastructure

The AI infrastructure boom has captured headlines with explosive growth and massive capital commitments. Microsoft and Meta have pledged over $120 billion to neocloud providers like CoreWeave and Nebius. Understanding the financing structures beneath the surface is essential for tracking the AI industry’s trajectory.

The Neocloud Phenomenon

Neoclouds are specialized cloud providers focusing on GPU-accelerated computing. They offer:

• Rapid deployment (months vs years)
• Latest hardware (first access to new Nvidia GPUs)
• Optimized utilization (superior GPU efficiency)
• Flexible capacity (scale without massive upfront capex)

The Scale of Commitments

Microsoft: ~$60 billion across CoreWeave, Nebius, and others
Meta: $35.2 billion to CoreWeave + $27 billion to Nebius = $62.2 billion
Combined: $122.2 billion (90% of AWS’s annual revenue)

For context:

• CoreWeave FY2026 revenue: $12.6 billion
• Nebius FY2026 revenue: $3.4 billion

Commitments are 10x larger than current revenues.

Why Hyperscalers Choose Neoclouds

1. Rapid Access to Latest Hardware

Neoclouds deploy new Nvidia GPUs faster than hyperscalers can build infrastructure. CoreWeave deploys new chips within 2 weeks of receipt.

2. Superior GPU Utilization

Model FLOPS Utilization (MFU) measures actual compute vs theoretical maximum:

• Industry average: 30-40%
• CoreWeave: 35-45% (20% better)
• Recent: >50% on Hopper GPUs

This efficiency gap is equivalent to getting 20% more compute from the same hardware.

3. Shifting Capex to Opex

Meta’s situation:

• Operating cash flow: $136 billion
• Capex guidance: $125-145 billion
• Without neoclouds: Negative free cash flow
• With neocloud contracts: Costs spread over 5-7 years as opex

The Circular Financing Structure

Nvidia’s multi-layered involvement:

1. Equity Investments

• $2 billion in CoreWeave
• $2 billion in Nebius
• Previous stakes in both companies

2. Financial Backstop
Nvidia commits to purchasing unsold GPU capacity:

• CoreWeave: $6.3 billion initial value
• Duration: Through April 2032
• Mechanism: If CoreWeave can’t find customers, Nvidia buys capacity

3. Supplier Relationship
Neoclouds use Nvidia’s investments to purchase GPUs from Nvidia, creating circular flow:

Nvidia invests $2B → CoreWeave buys GPUs from Nvidia → Nvidia records sales

The Funding Challenge

CoreWeave’s Situation

Q1 2026:

• Revenue: $2.08 billion (up 112% YoY)
• Operating cash flow: $2.98 billion
• Capex: $7.7 billion
• Free cash flow: -$4.71 billion
• Debt: $24.86 billion

Full-year 2026:

• Expected capex: $31-35 billion
• Expected OCF: $8.68 billion
• Funding gap: $17.33 billion
• Funding source: Debt (5x more than equity)

Nebius’s Situation

Stronger but still challenged:

• Cash: $9.37 billion
• Debt: $8.45 billion
• Net cash: $920 million
• Capex guidance: $22.5 billion
• Funding gap: $6.3 billion

GPU-Backed Debt

Both neoclouds rely on GPU-backed delayed draw term loans (DDTLs):

CoreWeave’s DDTL 4.0:

• Total: $8.5 billion
• Drawn: $1.26 billion
• First investment-grade GPU-backed debt
• Backed by: Customer contracts + GPU collateral
• Interest: U.S. Treasuries + 2% premium

Interest Rate Pressure:

• 3-year Treasury: <3.6% (early 2026) → 4.16% (june 2026)
• coreweave q1 interest: $536 million (25.8% of revenue)
• q2 guidance: $690 (27.3% revenue)

interest expense is becoming a significant profitability headwind.

the question

coreweave:

• capex />evenue ratio: 2.6x
• Must grow revenue faster than capex

Nebius:

• Capex/revenue ratio: 6.6x
• Even more dependent on revenue growth

Neoclouds must:

1. Grow revenue faster than capex
2. Achieve high utilization
3. Manage debt service
4. Maintain pricing

Risks

• Demand Risk: What if AI demand doesn’t grow as expected?
• Utilization Risk: Overbuilding capacity relative to demand
• Debt Risk: Rising rates increase debt service; refinancing risk
• Competitive Risk: Hyperscalers building internal capacity
• Nvidia Risk: If neoclouds fail, Nvidia’s $6.3B+ backstop activates

The Bull Case

• Genuine demand from hyperscalers
• Anthropic adding $15B ARR monthly demonstrates demand
• Neoclouds have demonstrated superior utilization
• Rapid deployment capabilities are real
• Market is large enough for multiple winners

Conclusion

The circular financing structure is not inherently problematic — it’s a rational response to massive capital requirements and genuine demand. However, it creates interdependencies and risks:

1. Scale is unprecedented ($120B+ commitments)
2. Profitability is uncertain (capex still far exceeds revenue)
3. Nvidia is deeply involved (investor, supplier, backstop)
4. Debt is rising (interest becoming significant burden)
5. Execution is critical (must achieve utilization and growth)

The AI infrastructure boom is real, but the financing structures are complex and carry risks. The next few years will be critical for determining whether this strategy proves sound or problematic.

Sign In