Nvidia’s $500 Billion Bet Puts CUDA at the Center of Global AI Compute

CUDA sits at the center of Nvidia’s AI compute business, supporting training, inference and other accelerated workloads on the same hardware base.

Nvidia’s latest quarter showed the scale of that demand: Data Center revenue reached $75.2 billion, including $60.4 billion from compute and $14.8 billion from networking.

With that revenue already visible, Wall Street is preparing more than $500 billion for Nvidia-linked AI infrastructure.

The bet is shifting from access to processors toward the earning power of installed compute, with CUDA helping keep that capacity useful across a wider range of workloads and customers.

CUDA Is What Gives Nvidia Compute Its Economic Life

CUDA gives Nvidia’s GPUs a common software base for training models, running inference, powering agents, scientific computing and industrial workloads.

That makes the same hardware useful across different stages of the AI cycle instead of tying it to a single application or customer.

Nvidia has extended that base with CUDA-X, Dynamo, Nemotron and AI Enterprise, widening the range of work its systems can handle as AI demand shifts.

The result is a compute platform that can stay productive across more workloads and over a longer period, which helps explain why Wall Street is willing to commit another $500 billion to the infrastructure built around it.

AMD Is Trying to Break CUDA’s Hold on AI Compute

AMD is attacking the part of Nvidia’s advantage that sits above the chip.

ROCm gives developers an alternative software stack for running AI and high-performance workloads on AMD accelerators, while migration tools are making it easier to move code away from CUDA.

Google and Amazon are applying pressure from another direction with TPUs and Trainium built into their own cloud infrastructure.

The contest is shifting from accelerator performance to the software ecosystems that keep expensive compute in use.

Easier workload portability would weaken some of the stickiness behind CUDA-based infrastructure; continued dependence on CUDA would keep demand anchored to Nvidia systems.

The software layer is becoming part of the contest over which AI infrastructure attracts the next round of capital.

Wall Street Is Moving AI Compute Into the Global Infrastructure Market

Wall Street is moving deeper into AI compute as the spending cycle starts to rival traditional infrastructure for long-term capital.

Apollo estimates hyperscaler capex could run at roughly 3% of U.S. GDP from 2027 through 2029, more than double its 2025 share, putting AI infrastructure among the largest capital-buildout themes in the market.

Nvidia is offering investors exposure to that expansion through compute used by AI labs, enterprises and cloud providers, with revenue tied to how much capacity stays in use.

CUDA widens the workloads that can run on those systems, while the $500 billion financing platforms give more customers a route into infrastructure that had largely been funded by Big Tech balance sheets.

Nvidia’s Next Earnings Will Test the $500 Billion Compute Thesis

Nvidia’s last earnings season gave Wall Street evidence that AI spending is extending beyond accelerators into larger computing systems.

Data Center growth remained 92% year over year while networking revenue nearly tripled.

The next earnings need to show whether that demand is broadening beyond hyperscalers into AI clouds, enterprises and industrial customers, and whether projects tied to the new financing platforms are beginning to convert capital commitments into deployed infrastructure.

Nvidia is now testing whether the economics behind AI compute can support a much larger capital market.

CUDA keeps a wide range of workloads anchored to Nvidia systems, while outside financing gives more customers access to those systems without relying on Big Tech-sized balance sheets.

The next phase of AI investment can therefore draw more heavily on private credit, infrastructure funds and institutional portfolios.

AI computing is moving into the global infrastructure market, with capital increasingly financing the capacity itself rather than only the companies buying the chips.