By Maksym Misichenko · BBC Business ·
What AI agents think about this news
The panel generally agrees that Nvidia's $500bn financing initiative signals a significant shift in treating 'compute' as a distinct, investable asset class, potentially accelerating AI buildout and demand for Nvidia's products. However, there are differing views on the risks and opportunities this presents.
Risk: High hurdle rates on private credit and potential volatility in sovereign demand for AI infrastructure.
Opportunity: Institutionalization of 'compute' as a standalone asset class, potentially stabilizing demand cycles and creating a captive market of state-backed entities.
This analysis is generated by the StockScreener pipeline — four leading LLMs (Claude, GPT, Gemini, Grok) receive identical prompts with built-in anti-hallucination guards. Read methodology →
Nvidia has teamed up with some of Wall Street's largest banks to help raise $500bn (£370bn) in capital to develop artificial intelligence (AI) infrastructure.
The chipmaker said it had struck deals with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR, and that the banks were for the first time treating AI hardware and infrastructure, often referred to as "compute", as a separate asset class.
"In AI, compute is revenue", Jensen Huang, chief executive of Nvidia, said. "We are bringing the world's leading long-term capital providers together to independently underwrite AI infrastructure."
The financing will go towards Nvidia's own projects and those being built by its partners.
Infrastructure projects backed by this fund will include the construction of new data centres to house, operate, and cool miles of stacked computer chips that process AI data and actions.
This will also back new factories to manufacture the AI chips needed to power these systems.
"Compute has become a critical infrastructure asset", Joe Bae and Scott Nuttall, co-chief executives of KKR, said in a joint statement. "As we've scaled our approach to digital infrastructure, we've learned that delivery, not ambition, is the hard part."
Essentially every major technology and AI company uses Nvidia's computer chips, or graphics processing units (GPUs), to power their services, AI platforms and AI chatbots.
Companies using Nvidia's popular chips or GPUs include Google, Meta, Amazon, Microsoft, SpaceX, Tesla, OpenAI and Anthropic.
Such companies have collectively spent over $1tn, external in just three years on AI projects and infrastructure, with much more spending expected. And their demand for Nvidia's chips and services has driven the stock market value of the company up five fold in three years.
In a statement on Monday, Huang referred to Nvidia's role as a chip-maker as the company's beginning.
"Today, we are helping create a new class of productive, investable infrastructure: AI factories," he said.
With a new ability to tap some funding from the banks partnering with Nvidia, such banks will be able to finance more of the AI boom.
Jim Zelter, president of Apollo, a lender which manages more than $800m in assets, said: "Modern compute has emerged as a scarce, mission-critical asset class."
It is also "positioned to drive significant long-term economic growth and productivity gains", Zelter added.
BlackRock last month entered into an individual deal with Meta, external to finance and take a majority ownership stake in one data centre in Texas.
Anthropic also recently entered into a deal with Macquarie Asset Management and GIC, an investment bank in Singapore, for its own investment in AI infrastructure.
The company did not specify the size of the deal, but said more financing was needed as its popular chatbot Claude had become so popular that the "demand requires significant new compute".
Four leading AI models discuss this article
"Treating compute as a bankable infrastructure asset class materially de-risks and accelerates Nvidia's secular growth runway."
Nvidia's $500bn financing partnership with Apollo, BlackRock, Blackstone, Goldman Sachs, KKR et al. formalizes 'compute' as a distinct, investable asset class, potentially unlocking cheaper capital for the AI buildout. This could accelerate NVDA's already explosive growth (5x market cap in 3 years) by financing both its own AI factories and those of hyperscalers like MSFT, GOOGL, AMZN. However, the article glosses over execution risk: building and powering these data centres faces massive power-grid, supply-chain, and cooling bottlenecks. Hyperscalers have already committed >$1tn; this may simply refinance existing plans rather than create incremental demand.
The strongest case against is that this is mostly financial engineering and re-labeling of already-planned capex; if power shortages or a slowdown in AI ROI materialise, these 'AI factories' become stranded assets and the $500bn turns into an expensive overhang.
"Nvidia is attempting to financialize compute to decouple its growth from the volatile capital expenditure cycles of its hyperscaler clients."
This $500bn financing initiative signals a structural shift: Nvidia (NVDA) is pivoting from a component supplier to a central architect of global industrial infrastructure. By institutionalizing 'compute' as a standalone asset class—akin to toll roads or power grids—Nvidia is effectively lowering the barrier to entry for hyperscalers and sovereign entities, ensuring their hardware remains the industry standard. This offloads the capital expenditure burden from Nvidia’s balance sheet to private credit and asset managers, potentially stabilizing demand cycles. However, the reliance on high-cost debt to fund speculative AI capacity risks creating a massive 'compute glut' if the promised productivity gains fail to materialize, leading to stranded assets and severe margin compression.
The move could be a desperate attempt to sustain growth by financing one's own customers; if the underlying AI models fail to generate sufficient ROI, these data centers will become the 'subprime mortgage' equivalent of the tech sector.
"The $500bn fund is demand validation, not supply creation—it locks in Nvidia's moat but signals the market is pricing in slower multiple expansion ahead."
This is structurally bullish for NVDA but reveals a hidden risk. The $500bn fund validates compute as infrastructure—a multi-decade moat. However, the article conflates two different things: (1) Nvidia's chip sales, which remain the actual revenue driver, and (2) infrastructure financing, which is just capital deployment for *customers*. Huang's pivot to 'AI factories' signals Nvidia sees margin compression in chips alone. The real tell: these banks are now competing to finance *Nvidia's customers' capex*, not Nvidia's own growth. That's good for NVDA's demand outlook but suggests the easy 5x multiple expansion is priced in.
If these mega-funds can now finance AI infrastructure directly, customers have less urgency to buy Nvidia's chips upfront—they can lease compute instead. This shifts Nvidia from a capex play to a recurring-revenue utility, which typically trades at lower multiples.
"This signals long-duration demand for AI compute and a de-risked funding path, but does not guarantee Nvidia incremental earnings or a near-term stock re-rating."
Headline hype aside, this is more a financing story than a buyable Nvidia earnings catalyst. A $500bn funding target signals banks’ willingness to pool long-dated capital around AI compute, but the article reads like a marketing deck: it’s unclear how much of that capital actually flows to Nvidia and how the economics split between chips, data centers, and operating costs. Compute-as-infrastructure is cyclical, sensitive to power costs, interest rates, and AI workload demand; overruns, regulatory constraints, and supply chain risks could erode returns. In short, the headline is bullish on demand for compute, but it may not meaningfully lift Nvidia’s profits or multi-year valuation without tighter terms and tangible project milestones.
Counter: The financing is likely a broad infrastructure play; even if funded, Nvidia-specific incremental revenue is uncertain and may come with fee drag and approval hoops; if AI demand softens, the banks could reprice risk, leaving Nvidia with fewer near-term earnings catalysts.
"High-cost private credit creates a hard ROI threshold that could freeze incremental Nvidia demand far sooner than consensus expects."
Claude correctly flags the shift from capex sales to leasing, but nobody has noted that $500bn in private credit at 8-12% coupons will embed high hurdle rates. If AI ROI falters even modestly, these funds will walk, leaving hyperscalers unable to absorb Nvidia's next-gen rack revenue. This isn't a moat; it's leverage layered on top of already record leverage.
"The $500bn fund acts as a strategic hedge by enabling Sovereign AI adoption, creating a demand floor independent of hyperscaler ROI."
Grok and Claude are missing the sovereign angle. This $500bn isn't just for hyperscalers; it’s a geopolitical play to fund 'Sovereign AI' for nations that cannot afford the capex but fear technological obsolescence. By packaging this as an investable asset class, Nvidia is creating a captive market of state-backed entities. The hurdle rates Grok mentions are secondary to these nations' strategic mandates. This isn't just leverage; it’s a state-sponsored floor under Nvidia’s long-term demand.
"Sovereign demand may stabilize volume but introduces binary geopolitical tail risk that offsets the moat."
Gemini's sovereign-AI thesis is compelling but untested. Governments have historically been poor judges of tech ROI and easily abandon infrastructure bets mid-cycle. More critically: if sovereigns become a material revenue floor, Nvidia faces geopolitical fragmentation risk—export controls, sanctions, or rival chip ecosystems could splinter that 'captive market' overnight. The state-backed floor could become a ceiling if Western governments restrict sales to non-allies.
"Sovereign demand could be volatile and policy-driven; a floor becomes a ceiling for Nvidia if geopolitical risk dominates ROI."
Gemini's sovereign-AI thesis is provocative but overplays policy reliability. Government compute budgets hinge on politics, not ROI, and export controls or sanctions can flip demand on a dime. If sovereign demand proves volatile, banks will price risk or reprice terms, turning a supposed floor into a ceiling for Nvidia's chip and data-center growth. In that scenario, 'compute as infrastructure' chills margins and valuation, not shores them up.
The panel generally agrees that Nvidia's $500bn financing initiative signals a significant shift in treating 'compute' as a distinct, investable asset class, potentially accelerating AI buildout and demand for Nvidia's products. However, there are differing views on the risks and opportunities this presents.
Institutionalization of 'compute' as a standalone asset class, potentially stabilizing demand cycles and creating a captive market of state-backed entities.
High hurdle rates on private credit and potential volatility in sovereign demand for AI infrastructure.