AI Is Shifting Risk From Balance Sheets to Capital Markets

“"Follow the capital stack. Follow the electricity. Follow the long bond."”

Nvidia is reducing the size of its direct OpenAI financing guarantee while considering a $3 billion equity investment in the power-and-data-centre platform behind the same project. Anthropic's prospective IPO is being valued against 2028 revenues. Hyperscalers have already issued almost $220 billion of bonds this year. The AI Supercycle is not slowing. Its financial architecture is being rebuilt.

Data cut: 13:00 CEST | Sunday, August 16, 2026. Sunday edition: U.S. and European cash markets are closed. Today's Pulse combines Friday's closing prices with new developments reported through Sunday morning.

The most important AI development this weekend is not another processor. It is a change in who carries the financial risk of the processor economy.

Nvidia is reportedly discussing an investment of up to $3 billion in SB Energy, the SoftBank subsidiary developing OpenAI's enormous Ohio data-centre campus. Roughly half could be invested when the project agreement is signed and half in a planned SB Energy IPO that could raise at least $5 billion. The discussions sit alongside approximately $100 billion of credit support being considered for the project.

But this follows Friday's report that Nvidia expects to provide an initial guarantee of less than $120 billion, compared with as much as $250 billion previously discussed, after investors raised concerns about the chipmaker's exposure. The proposed Ohio campus is a 10-gigawatt project. This is not retreat. It is risk redistribution.

1. 01 AI Financing Is Moving From Concentrated Guarantees to Distributed Risk

   Nvidia has reportedly reduced the initial Ohio guarantee to less than $120 billion from as much as $250 billion, while separately considering a $3 billion investment in SB Energy. SB Energy could seek at least $5 billion through an IPO, and the parties are discussing around $100 billion of credit support. Nvidia has also created financing platforms with six major institutional investors targeting more than $500 billion of third-party AI infrastructure capital.

   The capital requirement has become too large to remain concentrated on technology-company balance sheets. Risk is being unbundled along a chain: equity sponsor → project company → vendor support → customer lease → private credit → infrastructure debt → public equity → institutional refinancing. AI infrastructure is becoming a capital market, not merely a capex programme. The question shifts from "Does AI demand exist?" to "Who owns the residual risk if utilization, pricing or technological value falls short?" signal: AI infrastructure is becoming a capital market — the residual risk, not the demand, is now the decisive variable.

2. 02 AI Valuation Is Moving Further Into the Future

   Anthropic is preparing investors for a potential IPO using a projected $190–200 billion of 2028 revenue, compared with a run rate above $47 billion reported in May. Bankers are reportedly considering multiples based on forward forecasts rather than current earnings. Anthropic has projected at least $10.9 billion of Q2 revenue and its first quarterly operating profit.

   This is a textbook duration trade. What is unusual is the combination of extraordinary future growth assumptions, extraordinary present capital intensity, and historically expensive long-duration money. The valuation chain — 2028 revenue → future margin → infrastructure efficiency → capital requirement → discount rate → present value — compounds at every step. AI revenue expectations are accelerating; so are real yields. One increases terminal value, the other discounts it. signal: The Great Split is now inside the valuation formula itself.

3. 03 AI Capital Formation Is Starting to Raise Its Own Hurdle Rate

   Alphabet, Amazon and Meta have issued almost $220 billion of bonds in 2026, already more than double their 2025 total. U.S. 30-year real yields are near 3%, with German and UK real yields near decade highs. Reuters reports investors increasingly identify heavy AI borrowing alongside government issuance as a driver of competition for long-duration capital.

   AI is now large enough to affect demand for capital itself. The feedback loop: AI investment → bond issuance → competition for savings → higher real yields → higher WACC → higher required AI ROIC. This is the first endogenous financial constraint of the AI Supercycle. Capex can eventually become self-limiting even if technological demand remains strong — not because investors stop believing in AI, but because AI requires too much capital simultaneously with governments. signal: The more successful capital formation becomes, the higher the return threshold on the next dollar.

4. 04 Capital Moves Toward Europe as Industrial Capital Turns Geopolitically Selective

   European equity funds attracted $13.52 billion in the latest week, far above U.S. inflows of $2.58 billion. Yet German companies invested only €4.3 billion directly in the United States in H1 2026 — nearly two-thirds less than a year earlier and almost 80% below 2024. The STOXX 600 remains within roughly 1% of its record, with Q2 earnings expected to increase 23.4%, the strongest in almost four years.

   Financial capital is moving into Europe; European industrial capital is becoming more cautious about moving into America. That is The Great Split applied to cross-border capital. Europe has an unusual opening — to attract global financial capital while retaining more of its industrial capital at home — but only if it converts that into accelerated computing, energy infrastructure, advanced manufacturing, defence, mobility and Industrial AI. signal: Europe's question is whether it can turn a capital-flow advantage into a capital-formation advantage.

5. 05 Energy Security Is Becoming Explicit Sovereign Industrial Policy

   India ordered domestic refiners to target a maximum 63,810 tonnes of LPG production per day while maintaining storage and transport buffers; before the Iran war disrupted the Strait of Hormuz it sourced roughly 90% of LPG imports from the Middle East. The U.S. is preparing additional pressure on Iran, potentially targeting oil shippers, Chinese refiners and financial intermediaries; China bought more than 80% of Iran's shipped oil per 2025 Kpler data. Brent settled Friday at $88.52, up 6% for the week, with Hormuz traffic severely constrained.

   Energy security is now managed as national resilience architecture — domestic production targets, inventories, transport capacity, diversified sourcing. Every country building sovereign accelerated-computing capacity must answer the same question: can intelligence keep operating when energy supply is politically disrupted? The sovereign AI stack becomes chips + data + grids + generation + storage + logistics + capital. signal: Without the energy layer, technological sovereignty is incomplete.

Key takeaways

the second-order AI trade

• Nvidia is shrinking its direct guarantee (below $120B, from up to $250B) while considering a $3B SB Energy stake and an IPO of at least $5B — risk redistribution, not retreat.
• Anthropic's valuation is being anchored to $190–200B of projected 2028 revenue, a duration trade colliding with near-record real yields.
• Hyperscalers' ~$220B of 2026 bond issuance is now large enough to raise AI's own hurdle rate — the first endogenous financial constraint of the Supercycle.
• Financial capital is flowing into Europe even as German industrial investment in the U.S. collapses — The Great Split in cross-border capital.
• Energy security (India's LPG targets, Hormuz, Brent at $88.52) and record central-bank gold buying show sovereigns pricing resilience alongside intelligence.