The Factory Is Running: Jensen Huang, Vera Rubin, and the Moat That Compounds

Opening Bell: The AI Factory

Last week we explained how the AI infrastructure stack was sorting into three permanent layers run regulated on premise by Dell, in private clouds by Oracle, and in space by SpaceX for low-cost, broad public use. This week, the Founder of the hardware that powers all three just confirmed his latest step change, the Vera Rubin chip, produces 10x more inference at 1/10th the token cost.

Founder Jensen Huang has spent 33 years building an "AI factory," an industrial machine that turns electricity and data into intelligence. NVIDIA uniquely makes GPUs for every factory in every layer of the stack.

NVDA is our #1 largest holding in FFF. Nvidia is a high-conviction core holding that we plan to own long term. This week the investment case materially strengthened.

Founder Spotlight

Jensen Huang at NVIDIA (NVDA)

From Denny's to the infrastructure of human intelligence.

In late 1992, three engineers sat in a Denny's diner on Berryessa Road in East San Jose, drinking weak coffee and making a courageous bet: that the next wave of computing would be built on graphics-based parallel processing units (GPUs). Jensen Huang, Chris Malachowsky, and Curtis Priem had just quit LSI Logic, Sun Microsystems, and IBM, respectively, to start NVIDIA in Priem's Fremont townhouse with $40,000 in seed capital and virtually zero odds of success by Huang's own admission.

33 years later, Jensen Huang remains Founder CEO, providing continuity that matters more than most investors understand. The five eras of NVIDIA: GPU invention, CUDA platform, deep learning, AI training, and generative AI were deliberately sequenced and built on top of the last by a Founder thinking in decades, not quarters. A CEO hired in 2006 may have declared victory in gaming and distributed the cash, but Jensen Huang saw more: a GPU platform waiting to be born. He bet NVDA's balance sheet on it.

"The buildout of AI factories, the largest infrastructure expansion in human history, is accelerating at extraordinary speed."

- Jensen Huang

One of the common qualities of great Founders is their vision. While hired CEOs manage the current arc, great Founders are boldly positioning for the next one.

The Business

NVIDIA reported Q1 27 sales of $82B, +85% year-over-year (a +3% beat), led by data center sales +92% year-over-year, equivalent to 92% of total company sales. Adjusted EPS was $1.87, a +5% beat. Jensen Huang was right: Compute is revenues. However, the numbers we find most compelling are the ones the market is still not pricing in, in our opinion.

The Vera Rubin Thesis

While Blackwell made NVIDIA a $3.4 trillion company, Vera Rubin is a step change in innovation.

The Vera Rubin platform, six co-designed chips operating as a single AI supercomputer, delivers 10x higher inference tera operations per second per watt (TOPS/W) compared to Blackwell, at 1/10th the cost per token, and trains large language models (LLMs) with 1/4th the GPUs. It is in full production, and manufacturing partners across 350+ factories in 30 countries are ramping systems. Microsoft's next-generation Fairwater AI superfactories are being built around Vera Rubin NVL72 rack-scale systems. CoreWeave (CRWV), another FFF holding, is among the first to offer it commercially.

Jensen Huang's framing at GTC Taipei: "One prompt can launch a thousand-step journey of reasoning, retrieval, tool use, and response generation."

Uniquely designed for the Agentic AI workload, Vera Rubin makes workloads that were previously unaffordable possible.

While every NVDA analyst covers the hardware, the deeper moat is the software. In 2006, Jensen Huang made CUDA closed-source, running only on NVIDIA GPUs. CUDA created an ecosystem that created researchers that created deep learning that created the AI boom that created $75 billion in quarterly data center business.

Founders create new industries. Today, an estimated >80% of AI training workloads run on CUDA with high switching costs that would require years of retraining, tooling rewriting, and institutional knowledge to replace. Competitors have unsuccessfully tried to break NVDA's moat for 15 years.

NVIDIA is entering the CPU market at scale to test and validate model outputs. The ratio is changing: as agentic workloads scale, data centers need >4x the current CPU capacity per gigawatt of power. NVIDIA is on track for $20 billion in CPU revenue near term, adding to its $75 billion run rate and directly challenging Intel and AMD.

The Agentic Inflection

In the Q4 FY2026 earnings release Huang shared, "The agentic AI inflection point has arrived. Enterprise adoption of agents is skyrocketing."

We believe this is a durable structural shift. Training AI models was a phase. Inference was the next phase. Agentic AI models that reason, act, use tools, and run multi-step tasks is the phase that transforms compute from a capital expenditure into a direct revenue-generating input. When Jensen says "compute is revenue," he means it literally: every agent task is a billable unit of AI compute.

Blackwell Ultra, already shipping, delivers 50x better performance and 35x lower cost for agentic AI versus the Hopper platform. Vera Rubin delivers another 10x on top of that. The cost curve of intelligence is moving fast enough to create demand at every price point.

Our Position

NVDA is our #1 largest holding in FFF. Jensen Huang is the rare Founder who has reinvented his company 5x across three decades without missing a wave. The CUDA ecosystem is the most durable software moat in semiconductors. The Vera Rubin platform is in production and accelerating. And the agentic AI inflection has just begun.

We are patient, long-term shareholders in the infrastructure of human intelligence. Founders build what lasts.

A Stat That Travels

Industry Insight

CUDA at 20, the Moat That Compounds

The most underappreciated dynamic in technology investing is NVDA's vertical integration.

CUDA launched in 2006. At the time, it was a bet: general-purpose parallel computing on graphics hardware, for markets that barely existed. Jensen Huang made it closed-source and optimized exclusively for NVIDIA GPUs. He then did something rare for a hardware company: he went aggressively after developers, funding research grants, building academic partnerships, and distributing free tools. He subsidized the ecosystem because he understood that the ecosystem would become the moat.

The researchers who built the early CUDA applications are now the CTOs, AI researchers, and engineering leads at every hyperscaler, AI lab, and enterprise software company in the world. Their default is CUDA. Their stack assumes CUDA. Their tooling was written for CUDA. The switching cost is institutional, not contractual.

AMD's ROCm platform has made real technical progress. Custom ASICs from Google (TPUs), Amazon (Trainium), and Microsoft (Maia) are real products serving real workloads. After 15 years of well-funded efforts, NVIDIA still commands >80% of AI training.

The question we always ask at FFF: who built the moat, and why does it compound? Jensen Huang built the CUDA moat, and it compounds because every new researcher, every new AI application, and every new foundation model trained on CUDA deepens the ecosystem further. Vera Rubin is not a disruption to CUDA. It is CUDA's latest expression, with 10x better inference economics behind it.

Portfolio Pulse

Founders in Their Own Words

Conviction Corner

RIA Q&A on FFF's Role in a Portfolio

Q: "What do I say when my clients ask what AI means for their portfolio?"

A. FFF owns the Founders building all three layers of AI. Jensen Huang's NVIDIA chips are the mind that makes AI think. Larry Ellison's Oracle and Michael Dell's Dell data centers are the bodies that house the data. Marc Benioff's Salesforce and Tobi Lutke's Shopify, among others, are the soul: the software that puts AI to work. FFF's portfolio reads like a who's-who of the people building the next 30 years. That is a conversation your clients will remember. AI is not just a Silicon Valley story. FFF holds 100 founder-led companies across sectors and market caps that will benefit from AI: energy, financials, industrials, healthcare, and consumer businesses. Founders are building the next era of every industry, not just tech.

Macro Minutes

U.S. Economic Releases · U.S. Economic Releases, June 23–27

What We're Watching

Washington Wire

Post-Peace Deal, What the Strait of Hormuz Means for AI Infrastructure

“"This agreement paves the way for lasting peace and allows the reopening of the Strait of Hormuz. It is an important step in the right direction for our compatriots that will soon enable a decrease in energy prices."”

French President Emmanuel Macron, June 18, 2026

Our take. The peace deal is done, but the global economic benefits are just beginning. Falling energy prices are good for FFF. Data center electricity directly repriced when the Strait of Hormuz reopened. Oil sets the price of natural gas and electricity. The May 4.2% CPI print was driven higher by energy costs. If oil prices remain at lower levels in June and July, CPI will decelerate in Q3, giving the Fed room to hold or cut rates at future meetings. For NVIDIA specifically: every dollar reduction in per-megawatt-hour energy cost improves the economics of the AI factory buildout. Lower energy prices mean more data centers get built, more Vera Rubin systems get deployed, and NVIDIA's addressable market expands. The hard work of geopolitical reconstruction is ahead, but today, the macro setup for our Founders looks materially better than it did a month ago. Lower oil prices could translate into interest rate cuts in the future. Housing remains the most rate-sensitive sector in the economy. Affordability constraints have kept new home sales volatile, with only pockets of strength. Warsh struck a cautious tone in his first press conference last week, and a purchased home inspires a wave of downstream spending. Any signs of life from housing strengthen the soft landing case, the macro backdrop where our Founders tend to thrive.