Nvidia's Revenue Test Moves From Order Books to Real Workloads

Nvidia's growth story is shifting from the size of its order book to whether anyone actually uses what it ships. Two developments this week frame that transition: OpenAI is putting the company's Vera-CPUs into production for agent workloads with sandboxed code execution, and CEO Jensen Huang has floated a software licensing model that would tie recurring revenue directly to installed hardware.

Neither constitutes booked income. But together they sketch a company trying to convert technical availability into paying demand — a gap investors have been watching closely.

Oracle Deal Puts Vera-CPUs to Work

The expanded Oracle collaboration reported today moves the Vera-CPUs into the foreground. Availability of Vera Rubin NVL72 and Vera-CPUs on Oracle Cloud Infrastructure gives Nvidia a concrete deployment beyond an expansion announcement; the multi-gigawatt superclusters remain a future ambition. OpenAI's production use of the processors for agent applications adds a real-world application to the cloud capacity on offer.

For shareholders, the question is no longer whether another large buildout gets announced. It is whether the available technology performs well enough economically to pull in additional demand. The Oracle partnership offers a way to watch that handoff — though availability alone does not establish any revenue impact.

Should investors sell immediately? Or is it worth buying Nvidia?

Early tests presented at Oracle AI World offer a positive data point. Results attributed to Tae Kim show 51% more Python throughput than a leading x86 CPU and 2.2x the memory bandwidth per thread. Those figures point to possible performance advantages on the tasks tested. They say nothing general about all workloads or the economics of entire data centers, which is why real-world deployment matters more than a single benchmark.

A $2,000-per-GPU Software Idea

Huang used his October 22 appearance at GTC Berlin to raise a different lever: licensing the agent software suite at roughly $2,000 per GPU per year, according to media reports. The figure describes a potential annual revenue stream, not a signed customer contract or reported software sales. What would matter economically is how widely customers actually adopt such an offering.

It dovetails with the OpenShell Reference Architecture unveiled on October 21. According to the company, it is designed to deploy and manage AI agents across enterprise infrastructure and edge systems. Nvidia called the architecture immediately available and named partners spanning cloud, software, cybersecurity and systems.

The two steps carry different weight. The architecture is an available product; Huang's licensing comment describes a possible monetization approach. Commercial success cannot be inferred from availability alone.

Demand Outruns the Production Line

Supply capability remains the other test. Reuters reported on October 21 that ByteDance, Alibaba and Tencent placed orders for more than 3 million RTX PRO 5500 graphics cards, with a combined list value above $35 billion. Nvidia could not guarantee it would fill the orders in full, according to the report, because demand exceeded planned production. Those orders are therefore no proof of realized revenue — the distance between order interest and shipped hardware sits at the center of the valuation case.

Nvidia announced on October 21, alongside several suppliers, an expansion of manufacturing capacity. Substrate agreements are meant to deliver a 2.5x increase within three years. That addresses a different side of the growth question than software licensing: it aims to create the conditions for additional shipments, though announced capacity is not yet available production. For new U.S. test facilities, Nvidia is also making a $1.1 billion prepayment to KYEC to reserve capacity.

Meta, Google and the Wider Ecosystem

Under a favorable scenario, Nvidia establishes itself at Oracle not just with accelerators but as a broader infrastructure platform, with Vera-CPUs and Vera Rubin handling different tasks inside the same AI environment. The planned supercluster buildout would carry that combination to larger capacity.

Nvidia at a turning point? This analysis reveals what investors need to know now.

In parallel, Nvidia is deepening ties with Meta. Yesterday it announced a planned $3 billion investment in the company's share offering, with both firms aiming to optimize the Muse models for training and inference. Meta also plans heavier use of Vera Rubin, Vera-CPUs and Nvidia networking gear. Oracle and Meta thus pursue different routes to broader adoption of the same product families — cloud offerings on one side, in-house AI infrastructure and model optimization on the other. Software optimization reinforcing hardware use would benefit Nvidia, since more capable applications could justify additional compute. That remains a growth scenario, not a secured consequence of the partnerships.

The counterargument is substantial. Available systems and large buildout plans guarantee neither high utilization nor economically attractive AI applications. If the utility lags expectations, customers could postpone expansions, and the gap between announced infrastructure and actually needed capacity would become the problem. The planned Meta stake also shows how tightly capital provision and technology partnership now intertwine: Nvidia intends to invest in a customer that wants to use more Nvidia technology. That can strengthen collaboration, but it sharpens the question of whether the infrastructure pays for itself through application utility.

Oracle is not betting solely on Nvidia either. From 2027, Google TPUs are to be offered on Oracle Cloud Infrastructure, with capacity growing over several years. That does not automatically mean a loss for Nvidia, but it gives Oracle customers additional choices, and Nvidia must keep justifying its position through performance and fit for specific workloads. Presence on a major cloud platform alone creates no exclusivity.

Tonight's Earnings Are the Next Signal

The next concrete date arrives after the close tonight, when Microsoft and Alphabet report quarterly results. Investors will focus on cloud growth, infrastructure spending and the financing of AI outlays. Those reports will not prove the Oracle offerings a success, but they can show whether investment appetite in AI holds up. For Nvidia holders, that link is the one that counts: the technical deployment is becoming more concrete, and economically viable demand now has to follow.

Ad

Nvidia Stock: New Analysis - 28 October

Fresh Nvidia information released. What's the impact for investors? Our latest independent report examines recent figures and market trends.

Read our updated Nvidia analysis...