ME News

wrote a column ·

Huang's Bet

Article by: Xiang Xianzhi

When people think of the current AI boom, they typically picture either top-tier vendors racing to release new models or Taiwan Semiconductor’s foundries running around the clock alongside NVIDIA’s skyrocketing revenue and stock price.

Everyone is watching chip manufacturing capacity closely. Many assume that as long as chips can be produced, AI development will accelerate without restraint.

The reality is far more complex. Today, the bottleneck for this massive AI infrastructure buildout is shifting. Although Taiwan Semiconductor and memory manufacturers still face supply shortages, sustaining this imbalance is becoming increasingly unsustainable.

It’s common knowledge that building large-scale AI computing clusters burns cash at an extraordinary rate. What’s less commonly understood is just how enormous this financial burden truly is.

According to industry research firm SemiAnalysis, by 2029, global outstanding debt incurred from purchasing AI hardware and constructing supporting data centers will exceed $7 trillion.

Apple posted a record-breaking profit in fiscal year 2025, netting approximately $112 billion for the year. At that pace, even if Apple devoted all its net income solely to repaying this debt, it would take over 60 years to clear the balance.

Faced with such a terrifying financial black hole, even traditional deep-pocketed players are feeling overwhelmed.

Historically, funding for AI compute infrastructure has primarily come from tech giants like Amazon, Google, Meta, and Microsoft footing the bill themselves.

Right now, the entire industry urgently needs to find new sources of funding—otherwise, the engine driving AI compute expansion will stall due to a lack of capital.

Since tech supergiants alone can’t shoulder this burden, a group of nimble 'prospectors' has naturally emerged in the market. Many new AI cloud providers—dubbed Neoclouds—are stepping forward to act as 'computing capacity contractors.'

Their plan is to borrow money from financial institutions to buy NVIDIA GPUs, build out computing clusters, and then flexibly lease them to various AI ventures.

If this approach works, it could ease the bottleneck in computing infrastructure development. But while the vision is compelling, reality is harsh. These newcomers have barely entered the arena before slamming headfirst into a wall, leaving them stuck in a no-win dilemma.

If no one can afford to buy or build new computing clusters, who stands to lose the most? Undoubtedly, it’s NVIDIA—the company riding high on GPU sales.

To protect this cash-cow business and prevent computing capacity channels from being fully monopolized by a few established giants developing their own chips, NVIDIA has made a rare and ambitious cross-industry move.

It has decided to stop being just a well-behaved hardware supplier and instead reach directly into Wall Street’s financial playbook.

These emerging AI cloud providers—known as Neoclouds—face a deadly triangle. To successfully build a computing cluster, they must simultaneously secure three things:

Funding (bank loans), off-take agreements (customer leases), and data center space (facility real estate). These three elements form an interlocking cycle.

Banks are taking a highly pragmatic stance. In the eyes of financial institutions, AI ventures desperately seeking funding could collapse at any moment if they fail to secure their next round of financing. Leasing expensive GPU computing power to these high-risk, short-term tenants simply cannot safeguard hundreds of millions of dollars in loans.

To eliminate risk entirely, Wall Street banks have imposed an exceptionally stringent rule: emerging cloud providers must first submit a 'letter of intent'—essentially a pledge—to qualify for loans.

Specifically, they must secure a five-year take-or-pay computing capacity agreement with a technology giant that holds an 'investment-grade' credit rating—such as Microsoft, Meta, or Oracle.

When evaluating loan applications, banks completely disregard the business potential of these emerging cloud providers themselves; what truly matters to them is the massive balance sheet of the tech giant standing behind as guarantor.

This raises a puzzling question: Microsoft and Meta are themselves hyperscale cloud providers with vast resources at their disposal—so why would they lease hardware from these startup 'computing capacity contractors'?

The reason is that during the current AI boom, demand for computing power is growing at an explosive pace—one that outstrips even these giants’ ability to build data centers, secure power approvals, and scale their teams quickly enough.

To seize first-mover advantage, these tech giants have opted to lock in capacity upfront, effectively leasing entire clusters already built by emerging cloud providers.

This has created an absurd and ironic feedback loop: emerging cloud providers originally aimed to serve the broader startup community and act as partial alternatives to established giants.

Yet intense financial pressures have forced them into becoming nothing more than 'sublandlords' or even 'low-tier subcontractors' for those very same giants.

As a result, the very AI ventures and inference service providers who need flexible, short-term access to GPUs continue to face severe shortages—because the majority of GPU capacity on the market has already been locked up by the tech giants.

When a large number of AI ventures—those truly in need of flexible short-term rentals—come knocking, emerging cloud providers simply don’t have spare GPUs to offer.

If emerging cloud providers attempt to bypass the big players and sign one-year short-term contracts directly with ventures, then seek bank loans, banks will impose even more unreasonable terms.

For example, they might require ventures—without any credit rating—to prepay the entire year’s substantial rental fee upfront as collateral.

Securing funding and clients is merely the beginning of this nightmare. Even if emerging cloud providers reluctantly accept the tech giants’ ‘co-optation,’ they still face stringent scrutiny from data center operators.

These landlords, who control physical data centers, are equally risk-averse. In their view, leasing valuable data hall space and power capacity to emerging cloud providers carries enormous risk.

Landlords much prefer signing stable, ten- to fifteen-year leases directly with established giants.

To compensate for this perceived high risk, landlords demand a higher premium from emerging cloud providers, resulting in rental costs (or required yields) that are 3% to 5% higher than those of the big players.

The increasing concentration of computing power in the hands of a few oligopolies poses an existential threat that NVIDIA dreads most. These tech giants, who control critical infrastructure chokepoints, are all secretly investing heavily in developing their own custom AI chips. If computing infrastructure becomes monopolized by these giants, NVIDIA’s market dominance will be significantly weakened.

Faced with this interlocking dilemma, traditional hardware sales strategies have completely failed. Jensen Huang must now step in personally and deploy an unprecedented form of financial force to shatter this deadly triangle that has trapped countless players.

NVIDIA’s solution is 'debt backstop,' which can be categorized as a form of financial innovation. In a sense, NVIDIA is assuming a role akin to that of a central bank in the traditional financial system.

Many people may be unfamiliar with the concept of the 'lender of last resort.' In traditional financial crises, when commercial banks face bank runs and all financial institutions refuse to lend to one another due to extreme panic, the entire financial system’s liquidity chain can collapse instantly.

At such moments, central banks, leveraging their authority to issue fiat currency, act as the 'lender of last resort' by injecting liquidity into the market.

This absolute credit backing can significantly alleviate market panic and restore the flow of capital.

What NVIDIA is doing now is effectively serving as the 'central bank backstop' in the world of computing power.

Faced with Wall Street banks’ risk aversion toward the AI compute leasing market, NVIDIA has decided to step in directly, acting as the 'buyer of last resort' and credit guarantor for the entire AI compute credit ecosystem.

Specifically, the backstop agreement NVIDIA has signed with emerging cloud providers establishes a sophisticated mechanism that tightly aligns interests and risks—far more intricate than a simple 'guarantee.'

First, there’s a six-year 'minimum revenue guarantee.' NVIDIA offers emerging cloud providers a minimum income guarantee typically spanning six years—a duration that aligns precisely with the lifecycle and depreciation schedule of data center hardware assets.

Second, a comprehensive 'take-or-pay' mechanism with no blind spots.

What if emerging cloud providers build out their compute clusters, only to find insufficient demand from third-party AI ventures for GPU rentals due to market volatility?

NVIDIA has committed that, in the worst-case scenario, it will personally pay to lease back the idle GPU capacity at a pre-agreed price curve (or directly make up any revenue shortfall).

This means that even if the compute market cools down, emerging cloud providers will still receive a highly stable minimum cash flow—sufficient to cover principal and interest payments on their bank loans.

Buffett once said the first rule of investing is to preserve capital. Banks apply the same principle when lending: what matters most isn’t whether you’ll earn profits in the future—because no matter how much you earn, your loan repayments are fixed by contract.

They only care whether you can still repay the loan under the worst-case scenario.

With NVIDIA acting as the ultimate guarantor, Wall Street gained confidence and was willing to bypass traditional tech giants and readily approve hundreds of millions of dollars in loans directly to emerging cloud providers.

Of course, NVIDIA isn’t doing this out of charity—it achieves a 'double win' through this model.

This leads to the third key element of the agreement: tiered profit-sharing on excess returns. Since NVIDIA assumes the downside risk, it is entitled to claim a larger share of the upside.

According to the agreement terms, 100% of rental revenue up to the guaranteed baseline accrues entirely to the emerging cloud provider.

However, if computing power is in short supply and they flexibly lease it to various customers at a significant market premium, NVIDIA would take a large share—say, 40%—of the excess profits above the guaranteed floor through revenue sharing.

Through this mechanism, NVIDIA has successfully built a perfect 'computing-power circular financial ecosystem.'

On the front end, it continues to receive massive hardware payments from emerging cloud providers purchasing GPUs, ensuring robust cash flow for its core business.

On the back end, it secures a steady stream of long-term cloud service revenue through its share of cloud rental income.

The deeper strategic implication of this arrangement is that it frees emerging cloud providers from long-term contractual lock-ins with traditional giants.

They no longer need to be forced to bundle and 'wholesale' their computing capacity to just a few major players; instead, they can flexibly divide it into smaller units and rent it out monthly or annually to AI ventures that genuinely need computing power.

This not only fosters a thriving ecosystem for foundational AI innovation but also tightly binds numerous entrepreneurs to NVIDIA’s ecosystem, shielding it from market erosion by in-house chips developed by big tech firms like Google's TPU.

However, this model isn’t without flaws. NVIDIA is essentially engaging in a form of indirect supplier financing.

It leverages its massive balance sheet to stimulate and sustain market demand for its own chips.

This is walking a tightrope. If global demand for AI large-model inference and training over the next few years falls short of expectations, leading to excess capacity in the computing power market, NVIDIA will have to foot the bill itself to cover the resulting massive revenue gap.

NVIDIA is willingly taking on market volatility and credit risk head-on, stepping beyond the passive role typical of traditional hardware manufacturers—essentially leveraging its industry dominance and strong financial position to secure long-term market leadership.

This cross-sector financial move ultimately represents a precise balancing of risks and rewards—a strategic long-term play.

61K Views