Nvidia invested roughly $53 billion across 170 deals spanning the entire AI ecosystem, according to PitchBook data. In 2025 alone, the company participated in nearly 67 venture rounds, up from 54 in 2024 and just 12 in 2022. Its formal venture arm, NVentures, completed 30 deals in 2025.

The recipients read like a directory of every layer in the AI stack. Model builders: OpenAI, Anthropic, Mistral, xAI, Cohere, Thinking Machines Lab. Infrastructure providers: CoreWeave, Nscale, Nebius. Autonomous driving: Wayve. Robotics: Figure AI. Chip design tools: Synopsys, where Nvidia took a $2 billion equity stake in December. Even quantum computing (Quantinuum) and nuclear fusion companies made the list.

Bears look at this and see a scheme. Short sellers Jim Chanos and Michael Burry have compared Nvidia’s investing to the vendor financing that destroyed Lucent during the dot-com bust, where the telecom equipment maker lent money to customers who used it to buy Lucent gear. When the bubble popped, those customers defaulted and Lucent’s revenue evaporated. Nvidia pushed back with a seven-page memo to Wall Street analysts in November, arguing that its customers pay within 53 days, not over years, and that its investments and revenue streams are separate.

The bear case deserves more attention than a dismissive wave, but it also misses a structural difference. Lucent was funding customers who had no other source of demand. Nvidia’s largest customers, the hyperscalers like Microsoft, Google, Meta, and Amazon, are spending hundreds of billions of their own capital on AI infrastructure regardless of Nvidia’s investments. Microsoft alone spent over $50 billion on data centers in 2025. The startup investments are a sliver of Nvidia’s total revenue, not the foundation of it.

What “Letting A Thousand Flowers Bloom” Actually Means

Huang’s gardening metaphor maps to a specific strategic logic. AI’s final architecture is not yet settled. Nobody knows whether the dominant applications will be chatbots, autonomous vehicles, humanoid robots, drug discovery tools, or something that has not been built yet. Nobody knows whether the winning model architecture will be a massive frontier model from OpenAI, a lean open-source model from Mistral, or a specialized vertical AI agent from a startup that does not yet exist.

In that environment, the worst strategic mistake is to pick a winner too early. The second-worst mistake is to let a competitor become the default infrastructure provider while you wait for certainty. Huang’s response is to fund the entire field and ensure that whatever wins runs on Nvidia hardware.

This is not charity. The Synopsys deal, for example, is not just a check. It is a multi-year partnership to shift computationally heavy chip design workflows from CPUs to Nvidia GPUs. Synopsys’ tools are used across the semiconductor industry to design nearly every advanced chip. By embedding its CUDA computing platform into that design process, Nvidia makes itself harder to replace at the most foundational layer of the technology stack. Jensen Huang called it “a huge deal” on CNBC. He was not exaggerating.

The same logic applies to the model companies. Nvidia invested up to $100 billion in OpenAI, up to $10 billion in Anthropic, and participated in Mistral’s $1.99 billion Series C. These companies build on different architectures, serve different markets, and have different philosophies about open versus closed AI. Nvidia does not need to know which one wins. It needs all of them to need GPUs.

What This Means for Your Money

If you own an index fund that tracks the S&P 500, Nvidia is almost certainly one of your largest holdings by weight, given its roughly $4 trillion market capitalization. The question for any investor is whether this ecosystem strategy creates durable competitive advantage or inflates demand that will eventually unwind.

The honest answer is that both outcomes remain possible. Bernstein analyst Stacy Rasgon told Yahoo Finance there may be “no better use of Nvidia’s cash right now,” but also flagged that circular investment flows will “clearly fuel concerns.” Chanos warned that if AI demand in 2027 or 2028 falls short of today’s projections, “you could see orders begin to be canceled, and that’s a big risk that not a lot of people are talking about.”

The comparison to legacy automakers is instructive. GM, Ford, and Stellantis all made multi-billion-dollar bets on a single outcome: that consumers would rapidly adopt electric vehicles. When that assumption broke, they collectively wrote off more than $53 billion. Huang’s strategy is explicitly designed to avoid that kind of single-point failure. By seeding dozens of companies across the full AI stack, he is building optionality, not conviction around one product or one market.

The risk is not that the strategy is fraudulent. The risk is that AI demand broadly disappoints, in which case a diversified portfolio of AI bets still loses. But in that scenario, every technology company takes a hit, not just Nvidia.

The Phase To Watch

Huang acknowledged at the Cisco summit that his approach “makes for a messy garden.” That is the point. He said Nvidia has over 1,000 active internal AI projects and argued that if the company cannot predict which of its own projects will succeed, it certainly cannot predict which external startup will.

The transition to watch is when the garden gets pruned. As AI applications mature and winners become clearer over the next 12 to 24 months, Nvidia will likely shift from broad seeding to deeper integration with the platforms that have proven their value. That phase will look very different from today’s scatter-shot investing. It will look like lock-in.

For now, the company is placing bets everywhere because it can afford to. With a balance sheet larger than most countries’ GDP and the highest-margin chip business in history, Nvidia is in a rare position: it does not need to predict the future. It just needs to make sure the future runs on its hardware.

If you liked this, you’ll love the daily dose we send to our Substack crew. It’s chill, it’s sharp, and it actually helps. Hop on here.