You don't build a five-trillion-dollar empire just by selling hardware. You build it by controlling the entire ecosystem.
Nvidia CEO Jensen Huang is currently orchestrating one of the most aggressive corporate investment sprees in tech history. Over the last couple of years, Nvidia has deployed tens of billions of dollars—part of a massive deal pipeline approaching $90 billion to $100 billion in total commitments—into the very startups that use its graphics processing units (GPUs).
If this looks like a closed-loop economy, that's because it is. Skeptics call it a circular recycling of cash to inflate revenue. But if you look closer, it's a defensive moat designed to starve competitors of market share before they even get off the ground.
The Inside Track on the $90 Billion Investment Spree
Nvidia isn't acting like a passive venture capital firm. When its investment arms, like NVentures, write a check, the cash usually comes with a clear understanding: the startup will be buying Nvidia compute capacity.
Look at the sheer scale of the biggest deals fueling this boom:
- OpenAI: Nvidia participated heavily in OpenAI’s massive $110 billion fundraising round with a direct $30 billion investment. This formalized a deeper infrastructure partnership where OpenAI anchors its future on Nvidia’s next-generation Vera Rubin and Blackwell platforms.
- Anthropic: Nvidia poured billions into Anthropic’s recent $30 billion funding round. In return, Anthropic committed to purchasing up to 1 gigawatt of Nvidia GPU capacity to power its enterprise Claude models.
- CoreWeave: Nvidia invested roughly $2 billion directly into this specialized cloud provider, becoming its second-largest shareholder. Nvidia even signed a $6.3 billion backstop agreement to purchase any unused cloud capacity CoreWeave couldn't sell.
By funding the buyers, Nvidia ensures its order book remains completely full. Wall Street analysts estimate Nvidia has roughly $1 trillion in confirmed chip demand visibility through 2027.
The Real Reason Behind the Circular Cash Flow
The biggest criticism of Huang's strategy is that it creates an artificial demand loop. Nvidia hands cash to a startup like Mistral AI or xAI, and that startup immediately hands the money right back to Nvidia to reserve clusters of H100s or Blackwell chips.
But thinking this is just an accounting trick misses the structural brilliance of the play.
First, it locks developers into Nvidia’s software architecture, CUDA. Once an engineer builds an AI model using Nvidia's proprietary software stack, switching to cheaper hardware from competitors like AMD or Intel becomes a massive, expensive headache.
Second, it allows Nvidia to dictate the pace of the market. Startups aren't just buying chips; they're buying prioritized access. In a world where waiting six months for chips can kill a startup, getting on Jensen's good side is a matter of survival.
Moving From Generative AI to Agentic Compute
The nature of the AI boom is changing rapidly, and Nvidia’s deal spree is adapting to it. The initial phase of generative AI was all about training models—feeding text and images into massive supercomputers. Now, the market is shifting toward inference and what Huang calls "agentic AI."
"The compute needed for agentic AI has increased 1,000% compared to generative AI over just the last two years," Huang noted at a recent industry event.
AI agents don't just give a single chatbot answer. They read, plan, write code, use external tools, and check their own work over hours or days. Every single one of those internal steps requires heavy computational power.
To win this next phase, Nvidia is investing heavily beyond traditional foundation models. They put money into Anysphere, the creator of the Cursor AI coding tool, which is now used by 30,000 of Nvidia's own engineers to triple their programming output. They’ve backed Mira Murati’s new Thinking Machines Lab and Ilya Sutskever’s Safe Superintelligence.
Nvidia is even investing heavily in the physical world required to build these data centers. They've backed Crusoe, a company focused on energy infrastructure for mega-scale AI data centers, because chips don't run without massive amounts of power.
How to Navigate the AI Concentration Risk
If you own an S&P 500 index fund, you're already a major participant in Jensen Huang's master plan. Nvidia alone makes up roughly seven cents of every single dollar inside a standard index fund. When you add Microsoft, Alphabet, and Meta, over a quarter of your savings are tied directly to the exact AI buildout Nvidia is bankrolling.
The trade is no longer optional. If you want to manage your risk or find the actual upside in this next leg of the cycle, you need to look at where Nvidia’s money is flowing.
Stop focusing exclusively on the foundational model companies like OpenAI or Anthropic, where valuations are already sky-high. Instead, pay attention to the infrastructure layer that makes the data centers possible. Keep an eye on high-bandwidth memory makers, advanced liquid cooling suppliers, and specialized "neocloud" providers like CoreWeave that are receiving direct backing from Nvidia. The physical buildout is where the execution risk lies, and it's where the next wave of capital is being deployed.
