Microsoft, Meta and OpenAI: How Dependent Are They on Nvidia?
The AI industry relies heavily on Nvidia computing, but dependence looks different for cloud companies, model developers and technology platforms.
Harbour Creative Team2 min read
Harbour Creative Team2 min read
Image: Cover image supplied by Harbour Creative Blogs
Nvidia's importance to artificial intelligence is easy to state and difficult to measure precisely. Microsoft, Meta and OpenAI all need substantial computing capacity. But they do not buy, operate or use that capacity in the same way.
The first distinction: direct and indirect customers
In its 2026 annual filing, Nvidia explains that direct customers can include manufacturers, distributors, cloud providers and system integrators. Indirect customers may acquire systems through those businesses.
That makes claims about the exact revenue generated by an individual technology giant particularly difficult. Nvidia disclosed significant sales concentration among direct customers, but those anonymous figures cannot automatically be assigned to a named company.
Microsoft: infrastructure as a service
Microsoft sells cloud capacity to businesses and runs its own services on large computing systems. It therefore has two forms of exposure: the cost and availability of infrastructure, and the expectations of customers building software on top of that infrastructure.
An alternative chip may be commercially attractive, but switching takes more than ordering hardware. Software compatibility, networking, deployment reliability, engineering expertise and power requirements all affect the economics.
Meta: computing for its own products
Meta's spending supports advertising, recommendation systems, research and AI products. Unlike a pure infrastructure reseller, it can evaluate each system against internal workloads. That may allow a mix of hardware choices, with some tasks better suited to specialised accelerators than others.
Dependence on Nvidia today does not establish permanent exclusivity. Conversely, announcing an alternative processor does not mean an existing Nvidia-based deployment can be replaced cheaply or instantly.
OpenAI: a layered dependency
For a model developer, access to computing often runs through a hosting or infrastructure partner. GPU supply risk can therefore affect training schedules, inference capacity and costs even when the developer does not directly purchase every GPU.
The practical question is how much available capacity can be shifted to a substitute, on what timetable, and at what cost. Hardware alone does not determine what a model service can deliver.
The test for Nvidia's customers
Watch actual deployments, procurement disclosures and workload performance rather than headline chip announcements. A credible alternative has to combine adequate volume with a useful software environment and economic performance.
Nvidia's leadership transition adds a governance question, but the underlying commercial relationship is built on systems and delivery commitments. Customer dependence will be measured by the ability of those systems to keep operating and evolving.
About this article
This article provides analysis and interpretation of available information. Sources and supporting material, where applicable, are linked within the article.
Found an error? Contact the editorial desk. See our editorial standards and corrections policy.
From the Harbour Creative Blogs archive
Keep exploring this story
Related reporting, people and places from our published coverage. These are editorial connections, not a popularity ranking.
Company
Nvidia
The chipmaker at the centre of global AI investment: its revenue, products, leadership, customers and commercial relationships.
Company
Microsoft
Cloud computing, Windows, workplace software, AI investment and the consequences for Australian businesses.
Company
OpenAI
Artificial intelligence research, commercial AI services, compute infrastructure and strategic partnerships.


