In partnership with

Does Nvidia have staying power?

Today, we dig into the greatest company of the AI era.

MADE POSSIBLE BY

Your Brilliant Business Idea Just Got a New Best Friend

Got a business idea? Any idea? We're not picky. Big, small, "I thought of this in the shower" type stuff–we want it all. Whether you're dreaming of building an empire or just figuring out how to stop shuffling spreadsheets, we're here for it.

Our AI Ideas Generator asks you 3 questions and emails you a custom-built report of AI-powered solutions unique to your business.

Imagine having a hyper-intelligent, never-sleeps, doesn't-need-coffee AI solutions machine at your beck and call. That's our AI Ideas Generator. It takes your business conundrum, shakes it up with some LLM magic and–voila!--emails you a bespoke report of AI-powered solutions.

Outsmart, Outpace, Outdo: Whether you're aiming to leapfrog the competition or just be best-in-class in your industry, our custom AI solutions have you covered.

Ready to turn your business into the talk of the town (or at least the water cooler)? Let's get cracking! (And yes, it’s free!)

INSIDE THE COMPANY

Jensen’s Management Style

As Nvidia employees find themselves in the media’s focal point, they repeatedly illustrate an eccentrically unorthodox work environment. There is no escaping 60-person two-day meetings, there are no one-on-ones, and no-holds-barred dialogue is just the beginning.

The Jensen method

While traditional organizations have deep compartmentalization and levels of hierarchy, Huang keeps Nvidia flat. He receives direct reports from 50+ individuals.

Such a system provides him with a direct view into the company’s operations, unfiltered by tiers of middle-level managers (a staple of comparable tech outfits).

Despite carrying the U.S. stock market and temporarily sitting as the most valuable company in the world, Jensen seeks startup-level efficiency.

With that said, the company is unequivocally centered around him.

A prospective employee might think: “Wow, I can work directly with Jensen from the get-go.” But while the man is a visionary genius, Nvidians frequently speak out about his near-abusive leadership style.

“Demanding,” “perfectionist,” and “not easy to work for” are common phrases used by outspoken employees to describe him. On forums and private media interviews, we see an even more drastic illustration of his behavior.

Jensen believes in praising and chastising individuals in front of everyone, given his well-known belief that everyone should learn from each other’s mistakes.

He doesn’t hold back, often cursing and throwing personal attacks at employees in large meetings.

Ultimately, Jensen’s management style is incredibly empowering to “low-ranking” employees. But, the dark side of working for him is often glamorized or overlooked in the face of the company’s prosperity.

This all begs the question: is Nvidia’s success attributable to this management style, or did they simply make the right bet at the right time?

Built to last?

In the classic, time-tested book Built to Last, authors Jim Collins and Jerry Porras dissect the characteristics indicative of a company’s longevity.

In Chapter 4, the authors stress the importance of establishing a core ideology implemented tangibly throughout a company. These ideologies define the company’s ethos. Importantly, they are not tied to any one individual.

A SpaceX engineer once told me something to which I frequently return: “The first thing I noticed when I started was everyone’s relentless drive towards the goal of inhabiting Mars.”

He was on the Starlink team—Mars had nothing to do with their daily operations.

While the vision is Elon’s, the vision is not Elon.

Build a company that runs itself, so you can step back and admire your creation without it collapsing.

Dependence on a single leader violates these principles. Elon Musk, while indisputably the face of Tesla/xAI/SpaceX/Neuralink/X, can not possibly micromanage each company’s operations.

What he does is incomprehensible to the typical human; but, he has structured his companies to run autonomously under his guiding ideology.

Has Jensen Huang, on the other hand, built “Nvidia” or the “Jensen Huang Show?”

EMERGING PLAYERS

Competitors

Management aside, Nvidia has undeniably released the most prolific hardware product of the AI era.

Many people forget that not long ago, Nvidia was overwhelmed with demand from crypto miners. They even had to code restrictions into consumer GPUs to prevent crypto mining and stay in the good graces of the gaming community.

AMD has long been a key Nvidia competitor, particularly in provisioning individual GPUs for gamers. Now, Nvidia stands in a class of it its own, representing at least 70% of AI chip sales.

That said, there are promising specialized hardware alternatives.

Groq

Created by Jonathan Ross and backed by Chamath Palihapitiya, Groq has created the Language Processing Unit (LPU), a custom chip designed specifically for sequential component systems like Large Language Models. In other words, they’re ridiculously good at running LLMs.

How do they compare to Nvidia’s GPUs?

An Nvidia H100 can run inference on the Llama 2 (70B) model at 30-40 tokens per second. That’s pretty fast—roughly 10 words a second. Groq’s LPU, however, runs the same model at 300 tokens per second.

(Their homepage allows you to experience the speed.)

Groq is a unicorn with an expected $300 million round incoming. Will they kill Nvidia? Probably not—they’re a niche player focused on model inference. But, they’re an example of emerging chipsets designed specifically for LLMs.

Etched

Backed by Thomas Dohmke (GitHub CEO), Stanley Druckenmiller, Peter Thiel, Bryan Johnson, Balaji Srinivasan, and a host of VC funds, Etched has over $125 million raised to date.

Etched has “burned” the transformer architecture into their new chip, Sohu. In a quasi-manifesto press release, they explicitly state that “if transformers are replaced by SSMs, RWKV, or any new architecture, [their] chips will be useless.”

The company relies on the ongoing use of the transformer model—a logical bet, given that the transformer architecture is the reason LLMs became viable half a decade ago.

Because the transformer is “etched” into Sohu, they claim a whopping 500,000 tokens per second of Llama 70B inference. This opens the door to incredible applications powered by incomprehensible amounts of reasoning.

Both Groq and Etched are yet to release active, self-service enterprise access. When they do, I expect a frenzy.

Overall

It seems Nvidia’s GPUs are improving by growing, rather than by becoming more efficient. The latest B200 is only 15% more size-efficient than the H100 of 2022.

The key question: how much market share will specialized players like Groq and Etched take from Nvidia?

INDUSTRY TRENDS

The AI Landscape

Most of Nvidia’s post-2023 revenue has come from large GPU cluster purchases. Look no further than Meta, purchasing 350,000 H100s. The Middle East, a burgeoning region, is also getting its hands dirty with GPU clusters.

Plain and simple, these clusters are used to train foundational models. As large models converge in performance and small models outclass large models in domain-specific applications, will companies still be training huge foundational models?

The most impactful advances in large model performance have come from Reinforcement Learning with Human Feedback (RLHF) and better datasets. Now, small models fine-tuned with distillation (superimposing large models’ outputs) are outperforming large models in general intelligence!

Nvidia did launch NIM, a suite of inference microservices designed to accelerate models using their GPU hardware. While the performance benefits are nonnegligible (on the order of 2-3x), they pale in comparison to competing players like Groq and Etched with dedicated LLM chipsets.

LONGEVITY

Killing Your Darlings

Learn to kill your darlings.

The startup equivalent of killing your darlings is the pivot. Two of the greatest pivots of all time? Netflix and YouTube.

• Netflix began as a mailbox DVD rental service. Realizing the impending internet revolution, they pivoted entirely into a digital media company. To date, Netflix’s stock is up 56,978%.

• YouTube began as a video dating site called “Tune In, Hook Up.” Realizing they were part of a digital paradigm shift, and recognizing the demand for a video sharing platform, they pivoted into the life-essential product they are today.

Killing Nvidia’s darlings

Like the biggest tech companies, Nvidia invests heavily into research adjacent to their core product offerings. Their goal has always been to spearhead the next revolution, as they did for online gaming, crypto mining, and AI.

Will Nvidia remain at the top by the time the next revolution comes around? Will they be prepared, or are they too attached to their darlings?

Staying power

Up over 3,000% in the last 5 years, Nvidia has an EV/Revenue multiple of 39.24x. As a public company.

According to the latest edition of Jamin Ball’s Clouded Judgement (highly recommended publication), the best EV/Revenue multiples from top private companies are in the 16-21x range.

Amongst high-growth private companies projecting over 27% growth, the median EV/Revenue multiple is still almost 3x smaller than Nvidia’s.

To be clear, I am not saying Nvidia’s stock is overbought and due for a correction. Rather, that Nvidia’s short-term multiples reflect a market frenzy in dire need of price discovery. (Is that the same thing? Oops.)

In all seriousness, I’m clueless and opinion-less about the stock’s impending moves. Fundamentally though, Nvidia’s staying power depends on:

• Their ability to beat, compete with, or evade the threat of alternative chip architectures

• Penetration of software like NIM and CUDA, given the necessity of CUDA to build applications for GPUs (a monopoly)

• Employee retention, given the aforementioned work dynamic and the fact that stock grants can’t rise forever (currently, employees are in a perpetual cycle of waiting for the next vest)

• Their willingness to kill their darlings, keep employees happy, work on long shot bets, and position themselves in front of wherever computing will be in the next decade

I’m incredibly excited to see where Nvidia ends up over time. At the very least, they live in the Hall of Fame of companies and are a case study in modern history.

If you read this far, I just want to say thank you.

Please send me an email at [email protected] (subject: “Read Everything”) introducing yourself and letting me know if you agree with the takes in this edition.

WHO WE ARE

Return the Fund 🚀 

• One startup a week poised for 10x growth; market deep dives with actionable insights for builders and investors.

• Technical breakdowns of advanced new tech to empower informed decision-making

• Find your next prospect, product, job, customer, or partner. 🤝 Written by undercover pioneers of the field; trusted by builders and investors from Silicon Valley to NYC. 💸 

Last week, we uncovered the AI cloud.

As you know, we’re hell-bent on uncovering future unicorns cruising under the radar. Preeminent companies are lean, quiet, and driven before reaching their watershed moments. By the time people start talking about them, it’s too late.

You won’t get editions from us very often. Weekly at best. Two reasons:

1. We’re balancing full-time VC/startup work with Return the Fund.

2. We prioritize depth, insight, and value. This is not a daily news publication… We hope that when you do get an email from us, it’s dense, valuable, actionable, and worth saving.

Thanks for reading today’s RTF. Feel free to reach out to us at [email protected]. 🤝 

Psst: None of our company picks are ever sponsored. All research and opinions are completely held by the Return the Fund team.