Nvidia reported $107.6 billion in data center revenue Tuesday, up 110% from a year earlier, and disclosed up to $118 billion in financial guarantees tied to AI infrastructure projects. The guarantees put Nvidia on the hook for financing arrangements designed to help customers secure the land, power and facilities needed to deploy its systems.
The impressive quarter comes on the heels of a startling scandal that led to CEO Jensen Huang’s ouster and dominated headlines around the world. Huang resigned Nov. 12 after reports of multiple affairs, including one with a woman suspected by U.S. intelligence officials of having ties to Chinese intelligence. CFO Colette Kress is serving as interim CEO while the board seeks a permanent successor.
The campaign to bring Huang back attracted prominent supporters, including Elon Musk, Epic Games CEO Tim Sweeney and President Donald Trump. A public petition drew more than 180,000 signatures, including thousands of people identifying themselves as Nvidia employees. Huang said Tuesday he would be honored to return if the board decided it was in the company’s best interests.
The board rejected that possibility during Tuesday’s earnings call. Chairman Harvey Jones said directors unanimously accepted Huang’s resignation and did not intend to reinstate him. The board will consider internal and external candidates for a permanent CEO.
The leadership upheaval comes as Nvidia takes on a larger role in financing and securing the infrastructure needed to sustain its AI growth.
For data center operators, Nvidia’s guarantees address a constraint that comes before server deployment: financing and building facilities with enough power to support large AI workloads. Nvidia is helping customers bridge that gap while securing the components needed to manufacture its next generations of systems.
The largest guarantee supports SB Energy’s PORTS-Pike Technology Campus in Ohio. Nvidia is providing credit support for approximately 4.25 gigawatts across nine buildings. The campus will host Nvidia infrastructure under 20-year leases to OpenAI, subject to limited exceptions.
As buildings come online, Nvidia’s guarantee exposure will build. Each guarantee takes effect only after its associated building is completed and ready for service. The first is expected in fiscal 2029. Exposure will reach a maximum of $105 billion and then decline over the lease terms as OpenAI makes payments. Nvidia said it had made no guarantee payments as of Oct. 25.
The $118 billion maximum gross exposure also includes $9.5 billion in power and systems credit support guarantees and $3.5 billion in land, power and shell guarantees for neocloud providers.
Nvidia separately reported $32 billion in future commitments under neocloud agreements. Those commitments depend on providers securing financing, purchasing Nvidia infrastructure and deploying and operating the computing capacity.
Stephen Sopko, semiconductor and practice lead at HyperFRAME Research, cautioned against treating the full guarantee figure as exposure to neocloud financing problems.
“Of $118 billion in guarantees, $105 billion backs land, power and shell for one campus, PORTS-Pike, which is leased to OpenAI for 20 years. The neocloud commitments are $32 billion and only apply once the operator secures its own financing and the capacity is accepted,” he said.
Nvidia locks in supply years ahead
Nvidia’s supply commitments climbed to $390 billion at the end of the third quarter, up from $279 billion three months earlier. The company attributed most of the $111 billion increase to multiyear memory procurement.
The rest of the increase reflects deals across the manufacturing supply chain, including new substrate agreements with Kinsus and Samsung Electro-Mechanics and packaging expansions at Siliconware Precision Industries (SPIL) and Amkor Technology. Kinsus plans to increase advanced ABF substrate capacity dedicated to Nvidia by more than 3.5 times, with the additional capacity targeted for the third quarter of 2028. SPIL’s expansion runs through mid-2029.
Nvidia is securing components years before deployment, putting pressure on manufacturing capacity used by competing AI chip programs. The agreements cover memory, substrates and advanced packaging – components that require substantial capacity and long lead times.
“It’s no longer just leading-edge wafers. It’s memory, substrates and packaging, and that capacity takes years to build,” Sopko said.
Nvidia also has manufacturing relationships with suppliers including IBIDEN, King Yuan Electronics and Unimicron Technology. The company’s Vera Rubin platform will draw on the same TSMC 3-nanometer, 5-nanometer and CoWoS manufacturing base that the broader semiconductor industry depends on, Sopko said.
“Nvidia is locking in supply two to three generations ahead, which squeezes everyone else, including custom ASIC programs, for that same capacity,” he said.
Nvidia also reported $102 billion in combined cloud service agreements and data center lease commitments, up from $74 billion in the previous quarter. Those arrangements provide infrastructure for research and development, including chip engineering, product testing and AI model development.
Vera Rubin ramps as power limits persist
Nvidia began volume shipments of Vera Rubin NVL72 systems during the quarter, calling it the fastest product ramp in company history. Racks are running at CoreWeave, Google Cloud, Microsoft Azure, Oracle Cloud Infrastructure and Nebius, with systems expected soon at SpaceXSI.
Oracle Cloud Infrastructure announced general availability of Vera Rubin NVL72 systems and Vera CPUs. Google Cloud also announced general availability of its Vera Rubin-powered A5X instances. Nvidia expanded its collaboration with Meta, which plans to deploy Vera Rubin systems, Vera CPUs, Spectrum-X networking and BlueField-4 data processing units.
Nvidia reported that Vera Rubin delivered 10 times more tokens per watt than GB300 on the MLPerf Endpoints benchmark. The result measures performance under a specific test, not expected electricity savings across data center fleets.
“In power-capped sites, yes. Watts are the scarce input, so more tokens from the same grid allocation is real revenue capacity. But that 10x is an MLPerf result versus GB300, not a fleet average,” Sopko said.
He expects operators to use efficiency gains to run more workloads rather than reduce electricity consumption.
“They won’t bank the savings. They’ll spend them on bigger models and longer-running agents. Cost per token falls; total power demand doesn’t,” he said.
Nvidia forecasts $130B in fourth-quarter revenue
Nvidia reported $115.8 billion in total third-quarter revenue, up 103% year over year and 20% sequentially. Data center revenue reached $107.6 billion, up 110% year over year and 21% sequentially. Net income was $65.3 billion, compared with $31.9 billion a year earlier. GAAP diluted earnings per share were $2.70, up from $1.30.
Nvidia expects fourth-quarter fiscal 2027 revenue of $130 billion, plus or minus 3%. The forecast assumes no data center compute revenue from China. The company expects fourth-quarter GAAP and non-GAAP gross margins between 71% and 72%, reflecting the initial Vera Rubin ramp.
Sopko said the guarantees reflect a constraint beyond the availability of chips: getting powered facilities financed and built.
“The bigger shift is that powered shells are now the bottleneck, and Nvidia is acting as a credit enhancer to get them built,” he said. “It has also said it will be selective as other capital sources mature.”
Shane Snider is a senior news writer at TechTarget, covering AI infrastructure, hyperscale data centers, cloud platforms, and the power and energy systems driving modern compute expansion. You can reach Shane at [email protected] or on LinkedIn.