Prediction: Jensen Huang will forever change how dominant business models are created in technology companies. He took NVIDIA's compute moat and is now building an entire ecosystem around it to protect that competitive advantage
1. The customer concentration problem
During 2Q26, NVIDIA booked $96 billion of revenue. Of that, more than 50% came from hyperscalers, and the rest from AI enterprises and edge computing, which include neocloud businesses such as Coreweave, Lambda, etc.
NVIDIA is growing faster than ever before.
So what is the problem?
Five customers accounted for 70% of the money that NVIDIA is owed. The largest customer alone is 22% of their accounts receivable.
This is never the situation you want to be in as a CEO or CFO. If a single lost contract leads to you losing 1/5th of your business, you are living in fear of that customer walking away every single day.
Even scarier is that each of these large customers is a massive trillion-dollar company that is building internal alternatives for your own product. You have 70%+ gross margins as a business, and they want to find a way to bypass giving you those dollars.
Google (TPUs)
Amazon (Trainium)
Microsoft (Maia)
Meta (MTIA)
2. Financing Your Own Customers: The CoreWeave Example
NVIDIA's solution? Redirect demand for the GPUs through an intermediary instead of selling directly to the customers
In April 2023, NVIDIA participated in a $221M Series B in Coreweave, adding $100M for 17.9M preferred shares. This money helped CoreWeave buy H100s (NVIDIA GPUs) ahead of the shortage created by artificial intelligence hype from ChatGPT's launch
As part of this, NVIDIA also signed a $320M master services agreement (MSA) and became a cloud customer of CoreWeave itself. In a single move, Jensen became an investor, supplier, and customer of CoreWeave
In 2025, NVIDIA furthered the relationship by giving priority allocation to CoreWeave for their GB200 NVL72s. During March of the same year, NVIDIA put in an anchor order in the CoreWeave IPO, continuing to fund the business even as a public company
Later in September of that year, NVIDIA committed to buying any unsold capacity at CoreWeave through 2032 for up to $6.3 billion. This created a backstop for equity investors and lenders who were giving money to CoreWeave, essentially opening up the capital markets for the neocloud business
In 2026, Jensen announced that similar allocation priority will be given for the new Vera Rubin NVL72 deployment and invested another $2 billion in CoreWeave at $87.2/share (becoming a 11% holder of the equity). Thanks in large part to Jensen, CoreWeave is now a business with a revenue backlog of $104 billion and a market capitalization of over $40 billion.
3. Replicating this playbook
Since 2023, Jensen has replicated this same exact playbook across a bunch of other neoclouds and competitors.
Just a few examples
Lambda: NVIDIA was an early equity investor and is one of the company's largest customers, renting back about 18,000 of its own GPUs over 4 years for $1.5 billion, based on latest reports
Nebius: NVIDIA invested and holds roughly a 9% stake through a pre-funded warrant, based on latest reports. NVIDIA is said to be providing them with early access to the newest capabilities of GPUs and CPUs, similar to CoreWeave
Nscale: NVIDIA invested in the Series C
xAI: NVIDIA committed up to $2 billion of equity to a $20 billion SPV that buys GPUs from NVIDIA and leases them to xAI for five years. The debt is collateralized by chips and NVIDIA's equity acted as cushion for lenders to get comfortable in the financing
OpenAI: NVIDIA reportedly made a $30 billion equity investment and agreed to guarantee up to $105 billion of lease obligations at their Ohio campus, where it will also act as the compute supplier.
Anthropic: NVIDIA committed $10 billion (alongside $5 billion from Microsoft) in November 2025. In exchange, Anthropic committed to a deal with Microsoft Azure for up to 1 gigawatt of capacity on NVIDIA Grace Blackwell and Vera Rubin systems
4. Open Source as the Next Frontier
Despite this, Jensen has now correctly identified that closed-loop models still creates a series of concentrated buyers in the AI ecosystem. The GPU diversification solves the supply side of the AI equation, but if the demand side has just 2-3 large tenants, NVIDIA's relevance and moat will slowly fade away as a business
Open source is the solution to this problem. If open-source models can catch up to where the frontier models (Anthropic, OpenAI) are operating, you end up with hundreds of fragmented compute buyers
By supporting open source, NVIDIA can keep both of the supply and demand sides fragmented across many players and sit as the behemoth in the middle that controls the ecosystem.
As part of this plan, NVIDIA itself has committed to investing ~$26 billion on open-weight models.
As of today, they ship 6 model families
- Nemotron for reasoning and agents
- Cosmos for physical AI
- GR00T for robotics
- Alpamayo for AVs
- BioNeMo for biology
- Earth 2 for physics
In a brilliant move, Jensen is taking compute from the customers he supplied GPUs to and using that GPU capacity to create models that drive demand back towards those customers
In March 2026, eight labs (including Mistral AI, Reflection AI, and Perplexity) agreed to contribute data and expertise to train NVIDIA's models. The first base model, co-developed with Mistral, will be open sourced
The biggest announcement in this space came during September 2026 when NVIDIA agreed to buy the largest open model hub, Hugging Face, for roughly $13 billion.
NVIDIA paid 80x revenue to buy Hugging Face. Hugging Face is the platform where developers choose a model and also happens to be the perfect opportunity for NVIDIA to sell these developers on unused compute
5. The Five Layer Cake
Jensen Huang famously called artificial intelligence a five-layer cake
- Energy
- Chips
- Infrastructure
- Models
- Applications
He started off with an advantage in chips and is now slowly crafting an ecosystem at each layer that protects the core business of NVIDIA.
He is single-handedly shaping what AI and the future of humanity look like. Jensen Huang is playing chess, while the rest of us are playing checkers.
