SIAgentNews.com
← All news

Nvidia provides $118 billion in guarantees to build AI infrastructure

The chipmaker is acting as a credit enhancer to help customers secure the land, power, and facilities needed for massive AI workloads.

· First reported by datacenterknowledge.com

$118 billionmaximum gross exposure in infrastructure guarantees

The short version

  • Nvidia reported $107.6 billion in data center revenue, a 110% increase from a year ago.
  • The company is committing $390 billion to suppliers to secure memory, substrates, and packaging.
  • A major $105 billion guarantee supports a campus in Ohio leased to OpenAI for 20 years.
  • Nvidia's new Vera Rubin systems reportedly delivered 10 times more tokens per watt than the GB300 on specific benchmarks.

Why it matters

Nvidia is moving beyond just selling chips to becoming a primary financier of the physical infrastructure required for AI. This shift suggests that the biggest hurdle for AI growth is no longer just hardware production, but the ability to build and power the data centers that house them.

Financing the AI backbone

Nvidia is taking on a significant role in financing the physical infrastructure required to sustain its growth. The company revealed that it has tied up to $118 billion in financial assurances for projects involving AI infrastructure. These guarantees help customers bridge the gap of securing land, power, and facilities needed for large AI workloads.

The largest of these arrangements supports a technology site in Ohio. Nvidia is offering credit backing for roughly 4.25 GW across nine buildings. This campus will host Nvidia infrastructure under 20-year leases to OpenAI, with the first buildings expected to come online in fiscal 2029.

As datacenterknowledge.com reported, these guarantees act as a credit enhancement. This means Nvidia is helping customers secure the funding necessary to build "powered shells," which are now considered a primary bottleneck for the industry. Nvidia has stated it will be selective as other capital sources mature.

Securing the supply chain

To ensure it can meet future demand, Nvidia is locking in its supply chain years in advance. The company's supply commitments rose to $390 billion at the end of the third quarter, an increase from $279 billion just three months earlier. Most of this increase is attributed to multiyear memory procurement.

The company is also expanding deals across the manufacturing chain for components like substrates and advanced packaging. For example, Kinsus plans to expand its advanced ABF substrate capacity for Nvidia by more than 3.5 times by 2028. Other agreements include expansions at SPIL and Amkor Technology to ensure long-term availability.

These components require substantial capacity and long lead times to manufacture. By securing these materials years before they are deployed, Nvidia is effectively squeezing the available capacity for competitors. This includes other custom ASIC programs that rely on the same manufacturing base used by Nvidia's Vera Rubin platform.

Performance and power limits

Nvidia recently began volume shipments of its Vera Rubin NVL72 systems, calling it the fastest product ramp in the company's history. These systems are already being deployed by major cloud providers including Google Cloud, Microsoft Azure, and Oracle Cloud Infrastructure. The company reported that Vera Rubin delivered 10 times more tokens per watt than the GB300 on the MLPerf Endpoints benchmark.

While the efficiency gains are significant, experts suggest that operators will likely use this extra power to run larger models rather than reducing their total electricity use. Because watts are a scarce resource in many locations, more tokens from the same grid allocation represents real revenue capacity for the data center operators.

The goal for these efficiency gains is to lower the cost per token for users. However, this does not necessarily mean a reduction in total power demand across the fleet. Instead, it allows companies to run longer-running agents and bigger models within their existing power limits.

Leadership and financial growth

The company's financial results show a massive surge in data center activity. Nvidia reported $107.6 billion in data center revenue, which was up 110% from a year earlier. This contributed to a total third-quarter revenue of $115.8 billion and a net income of $65.3 billion.

This growth comes during a period of significant leadership upheaval at the company. CEO Jensen Huang resigned in November following a scandal involving multiple personal affairs. The board of directors has unanimously accepted his resignation and is currently seeking a permanent successor.

While Huang expressed a willingness to return if it served the company's interests, the board has rejected that possibility. CFO Colette Kress is currently serving as the interim CEO while the board evaluates both internal and external candidates for the permanent role.

What to watch

  • Whether Nvidia's $118 billion in guarantees leads to any defaults or payment issues.
  • The selection of a permanent CEO to replace Jensen Huang.
  • How many other 'neocloud' providers will sign the $32 billion in future commitments.

Try it

Look up the MLPerf Endpoints benchmark to see how different AI chips are measured for performance per watt.

Sources and links

First reported by datacenterknowledge.com. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.

Liked it? Say why

Add a line for other readers.

No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.

More AI news