> ## Content Index > Fetch the complete content index at: https://oddbits.ai/blog/llms.txt > Use this file to discover other available public pages before exploring further. # I Raised From Nvidia: What Circular Financing Gets Wrong - URL: https://oddbits.ai/blog/i-raised-from-nvidia-what-circular-financing-gets-wrong/ - Published: 2026-11-13T21:35:08.000Z - Updated: 2026-11-13T21:42:24.000Z - Description: With Jensen stepping down, what happens to open-source AI? Wall Street thinks NVIDIA invests in AI companies to sell more GPUs. I think they have it backwards. I raised from NVIDIA as a founder and worked with them as a neocloud CEO. What I saw: NVIDIA is filling a funding gap nobody else will. - Author: Will Falcon Wall Street thinks NVIDIA is investing in AI companies to get them to buy more GPUs. I've raised money from NVIDIA as a software company and later worked with them as a neocloud. I've seen how they invest from both sides, and I think Wall Street has it backwards. **NVIDIA isn't manufacturing demand for GPUs. It's filling a funding gap that nobody else will.** Of all the raises I've done, NVIDIA was one of the most rigorous investors. Their diligence isn't done by generalists. It's done by AI experts who literally built the tools in your space, and they can tell whether your tech is real and valuable. In my experience, NVIDIA is a deeply specialized investor with more vision about where AI is going than most VCs. ## **The funding gap** In AI, there’s no such thing as SaaS margins. Whether you buy GPUs or pay per token, you’re paying for GPUs. A token is just a GPU dressed up as software. That breaks the way AI companies get funded. VCs invest in high-margin software and shy away from capital-heavy businesses. Traditional buyout P/E wants predictable cash flows, and infrastructure capital wants contracted revenue from investment-grade customers. Most AI companies sit somewhere in the middle: they grow like software companies, spend money like infrastructure companies, and aren’t investment grade yet. Funding them takes deep pockets, risk appetite and vision, and no single type of investor today has all three. ![](https://oddbits.ai/blog/content/images/2026/11/image-3.png) That’s the gap NVIDIA fills. Some call it circular financing. I think it’s more accurate to call it a new kind of investor, one with P/E-sized pockets, VC-level risk appetite and the deep knowledge of AI to take big bets with conviction. ## **Why it looks like circular financing** Wall Street is confusing strategic investment with circular financing. The story goes like this: NVIDIA funds a company, that company turns around and buys NVIDIA GPUs, and NVIDIA turns balance-sheet dollars into revenue. But investing in a company that buys your products doesn't make the financing circular. Nearly every AI company buys NVIDIA GPUs, whether NVIDIA invests in them or not. The real question is whether those businesses have independent demand and real customers. In my experience, they do. ![](https://oddbits.ai/blog/content/images/2026/11/image-4.png) NVIDIA benefits twice, once as an investor and once as a supplier. But here's the thing: NVIDIA's interests and ours happen to line up. NVIDIA doesn't want a world with three cloud providers and three model providers, OpenAI, Anthropic and Google, because a handful of buyers would have all the leverage. Neither should we. Fewer choices mean higher prices, slower innovation, and a few companies deciding who gets access to AI. When NVIDIA funds alternatives, it's protecting its business, and it's doing the rest of us a favor. NVIDIA isn't investing to manufacture demand for GPUs. It's investing in companies building businesses around demand that already exists. ## **How the deals actually work** If NVIDIA invested in every startup that buys its GPUs, it would own a piece of every AI company on earth. It doesn’t. I can’t speak for every deal, but here’s how I’ve seen NVIDIA approach them. For startups that aren’t neoclouds, NVIDIA invests one of two ways: through NVentures or through corp dev. NVentures invests on VC-style terms. Corp dev invests when there’s a product partnership. “Strategic” doesn’t mean “you buy GPUs,” since every AI startup buys GPUs anyway. NVIDIA rarely leads. It usually joins rounds that Tier 1 VCs have already structured, and those rounds tend to be competitive. More often than not, the round has to make room for NVIDIA, not the other way around. Where NVIDIA differs from a traditional VC is scale. Once a round gets to around $1B, most VCs drop out, and P/E firms lack the context to understand what they’re investing in. NVIDIA in the round is the bridge that gets the P/E world comfortable. Bears compare this to Lucent [\[1\]](https://tomtunguz.com/nvidia%5Fnortel%5Fvendor%5Ffinancing%5Fcomparison/?ref=oddbits.ai), which lent cash to thinly capitalized startups. NVIDIA mostly takes equity, usually alongside other investors, in businesses whose demand comes from their own customers, not from NVIDIA. That’s the test that matters: would those customers keep paying if NVIDIA disappeared tomorrow? ## **Neoclouds need a co-signer** Imagine you’re a college grad trying to sign a seven-year lease on a luxury apartment with no credit. That’s every new neocloud. Before a neocloud can rent out a single GPU, it has to secure datacenter leases. Datacenter landlords come from real estate. They want credit history and a lot of money up front. For a college grad, the answer is simple: a parent co-signs. In Silicon Valley, the promise of AI is enough to raise money. Neoclouds are where that promise meets the rigor of the capital markets, and it isn’t enough. Infrastructure investors want creditworthy tenants, and VCs can put their money to work more efficiently elsewhere, so neoclouds have no parents. Your options are to rent a smaller apartment (sign smaller datacenter leases and grow slowly) or find someone to back you. NVIDIA understands this problem better than almost anyone, and it has the balance sheet and industry knowledge to help neoclouds get off the ground. When NVIDIA backs capacity or infrastructure commitments [\[2\]](https://www.sec.gov/Archives/edgar/data/1045810/000104581026000075/nvda-20260726.htm?ref=oddbits.ai) [\[3\]](https://www.sec.gov/Archives/edgar/data/1769628/000176962825000047/crwv-20250909.htm?ref=oddbits.ai), it isn’t manufacturing demand for GPUs. It’s helping promising companies get past a financing gap that traditional capital markets won’t fill. But NVIDIA can’t co-sign for everyone. And that’s the opportunity nobody is talking about: there’s still a massive financing gap for whoever is willing to underwrite the infrastructure commitments young AI companies can’t make on their own. ![](https://oddbits.ai/blog/content/images/2026/11/image-5.png) Bears will say that's just NVIDIA funding startups to buy its GPUs. But NVIDIA isn't creating demand for GPUs. It's helping companies build the infrastructure to serve demand that already exists. It also turns out that neoclouds that manage their debt well can become money-printing machines and genuinely good businesses. So NVIDIA may keep investing simply because they're good investments. ## **What NVIDIA’s money built** In 2019, the math on our business didn’t work. The demand was there. What changed was the price of a GPU. When we started running cloud infrastructure, there were only three options: AWS, GCP and Azure. At their prices, roughly 5x what neoclouds charge today, we couldn’t have been profitable, even though the demand was clearly there. Once neoclouds launched, that gap collapsed. That gave birth to a wave of startups, including coding startups like Cursor, that helped the first enterprises and developers get real value out of AI. It let open-source developers train and fine-tune models that have closed much of the gap with the big three. And it built an open-source ecosystem that’s starting to let enterprises and developers own their own AI, and with it, their sovereignty. Today, NVIDIA is the biggest backer of American open-weight models at the frontier. Without it, Nemotron wouldn’t exist, and Reflection wouldn’t have gotten off the ground. ## **Own your intelligence** Owning your intelligence is like securing freedom of speech. As AI becomes the way we think, write and communicate, whoever controls the model controls what we can say. ![](https://oddbits.ai/blog/content/images/2026/11/image-6.png) Jensen stepped down this week, but his strategy brought us to the edge of personal models becoming a reality. I’ve worked with many senior leaders at NVIDIA since our partnership began in 2019, and I can tell you the mission of open-source AI is bigger than any one person. I hope the new leadership team keeps that mission going and keeps supporting the ecosystem. Imagine if Steve Jobs and Bill Gates hadn’t made the personal computer a thing. Everyone would still be sending their work to an IBM mainframe and waiting days for the result. There’d be no smartphone in your pocket. Personal computing gave everyone privacy and freedom. The next stage of AI is breaking up the monopoly of a few centralized models, so every person on earth can own a personal model they control. If we come to rely on AI the way we rely on electricity, we can’t live at the whim of a few companies that might flip the switch and plunge the world into an AI dark age. \[1\] [Tomasz Tunguz, “Circular Financing: Does Nvidia’s $110B Bet Echo the Telecom Bubble?” ](https://tomtunguz.com/nvidia%5Fnortel%5Fvendor%5Ffinancing%5Fcomparison/?ref=oddbits.ai) \[2\] [CoreWeave Form 8-K, September 15, 2025\. ](https://www.sec.gov/Archives/edgar/data/1769628/000176962825000047/crwv-20250909.htm?ref=oddbits.ai) \[3\] [NVIDIA Form 10-Q, Q2 FY2027\. ](https://www.sec.gov/Archives/edgar/data/0001045810/000104581026000075/nvda-20260726.htm?ref=oddbits.ai)