NVIDIA used to be a company most people associated with graphics cards.
Gaming. GPUs. Better graphics. Faster PCs.
That story has changed dramatically.
AI has turned GPUs into infrastructure.
And now things are getting even more interesting.
Reuters recently reported that financial markets are moving toward futures contracts linked to NVIDIA GPU rental prices.
Think about that for a second.
Companies may soon be able to hedge the future cost of AI computing capacity in a similar way industries manage exposure to commodities.
That tells a much bigger story than another NVIDIA earnings headline.
AI Needs Somewhere to Run
ChatGPT, AI agents, image generators and enterprise AI platforms may feel like software.
But underneath all that intelligence is physical infrastructure.
GPUs.
Data centres.
Electricity.
Cooling.
Networking.
Storage.
Lots of it.
Reuters estimates NVIDIA chips can represent around 60% of the cost of an AI focused data centre, while NVIDIA maintains more than 80% share of the AI chip market.
That puts NVIDIA in a very unusual position.
It isn't simply selling chips anymore.
Its technology is becoming part of the basic infrastructure behind the AI economy.
Read the Reuters analysis on how financiers are starting to treat NVIDIA powered compute more like a commodity: Financiers are set to turn Nvidia into an AI baron
The AI Conversation May Be Focusing on the Wrong Thing
Most conversations about AI still focus on models.
Which model is smarter?
Which chatbot is better?
Which AI agent can do more?
But perhaps another competition is happening underneath.
Who controls the compute?
AI can become incredibly intelligent, but intelligence still needs somewhere to run.
And as organisations become increasingly dependent on AI, compute capacity starts becoming a strategic dependency rather than simply another IT resource.
That dependency also creates a cybersecurity question.
An earlier discussion on AI agents and security boundaries highlighted how AI security increasingly depends on controlling identity, permissions and access.
Perhaps infrastructure needs to be added to that conversation too.
Because the AI race may not ultimately be decided only by who builds the smartest model.
It may also depend on who owns, controls and can afford the machines running it.
AI might be software.
But the AI economy is becoming very physical.