I think it’s kind of worth wondering how one can protect oneself in the face of an AI revolution.
There’s a lot of big questions around it. Much of it goes like:
If everything’s automated, how will I get a job?
If I don’t get a job, how will I get money?
If no one has money, how will companies get any revenue?
And will money even mean anything in a time of automated abundance?
Right now I don’t have the answers to exactly how all of this will play out. But I figured it’d be worth sharing how I’m making long-term bets with what I have.
Disclaimer: this is not financial advice, these are just my notes on how I’m thinking about the tech and AI sector’s future
If you invested in an index that tracked the S&P 500 at the start of 2023, you’d have gotten great returns; roughly 23%. But if you break down the top performers in this list of 500 companies, there’s a funny pattern:
Most of the gains were concentrated in the top 7 tech companies, coined “The Magnificent Seven”. These tech companies rose 75% in 2023. If you removed those companies from the S&P 500, its gains were actually a modest 12%.
These seven companies were:
Apple (AAPL)
Microsoft (MSFT)
Alphabet (GOOG)
Amazon (AMZN)
NVIDIA (NVDA)
Tesla (TSLA)
Meta Platforms (META)
I think that AI drove a lot of movement here. At the same time, I don’t think a lot of nuance was applied. I am going to go through each item on the list and explain my position for why I’m picking or dropping the items on this list.
If you want to stop reading here, these are the four companies I’m calling “The Magnificent / Chad Four”:
NVIDIA
Tesla
Apple
Microsoft
First, the rejects:
Alphabet / Google:
What? Google? Aren’t they all about AI? Why wouldn’t they do well?
There’s several chronic issues at Google. The elephant in the room is that they should have been the ones to make ChatGPT and didn’t, but I’ll get to that in a second.
First, they kill projects left and right. Note, these are often projects with users. Maybe even millions of users! On top of that, they were often paying users. Nevertheless, these services get axed. Of the projects they do keep alive, it feels like delivery of new or needed features are slow or non-existent. I basically can’t in good conscience recommend google’s services for most developers these days.
Second: A chronic inability to ship. Everyone saw that for the past year, Google was promising to ship Gemini, its answer to ChatGPT. But as the months went on and on, its delivery date kept getting pushed back.
OpenAI would deliver a new feature or modality or improvement on ChatGPT, and seemingly Google would go back to the drawing board and figure out how to try match it, but never quite meet it. Finally they did release Gemini, but its intelligence and capabilities mostly don’t exceed GPT-4.
Third: Internal obsession with form over substance. In 2017 Google published a research paper titled “Attention is All You Need”. It describes the neural architecture that kicked off this current AI revolution that ChatGPT relies on.
The paper had 7 authors. 6 out of these 7 authors went on to found billion dollar companies or become worth billions through some other fashion.
Naturally, with odds that high, you’d want to get your name on Google’s next big AI paper, right? That’s what happened. Gemini’s paper had over 900 authors on it. Keep in mind, Gemini at its time of release wasn’t even state of the art; still underperforming GPT-4, maybe only outperforming some open source models.
It makes the whole situation look funny and a little bleak. As it stands I have no reason to be optimistic for Google’s future in AI.
Meta / Facebook:
Zuckerberg is taking AI extremely seriously. He recently announced that Meta has acquired 350,000 H100 NVIDIA GPUs, which are the best AI training chips that money can buy.
With other GPUs, Meta owns the equivalent computational power of 600,000 H100s. Reports state they’re spending between $9 Billion - $20 Billion on GPUs and model training.
So why wouldn’t I invest in Meta long-term?
Basically, I don’t see any real strategy for them, aside from attempting to become an AI-cloud company.
Meta’s approach, as far as I can tell, is to run a war of attrition. For the first time in 20 years, a unipolar silicon valley’s power balance has tipped significantly away from Google. Microsoft is pouring its arsenal into AI models and services it can run and sell. OpenAI has become a new ascendant force, another player to contend with.
So Meta is dropping bombs. They are spending billions to train massive, advanced models, and then turning around and open-sourcing them immediately.
The stated plan from Meta is to “develop AGI and then open source it”.
If they are lucky, they might bankrupt OpenAI and Microsoft. The issue at hand is that they aren’t the only ones open sourcing large AI models, and the models they’ve released so far aren’t even the best by a long-shot — for now.
Who knows, maybe Meta becomes an AI cloud provider, and maybe AI massively enhances the metaverse experience. Maybe it boosts sales of their meta glasses.
I don’t know. But what’s not clear to me is that their strategy for what they’re doing with AI now really lines up with getting them to wherever they’re trying to go.
Amazon:
Amazon’s a lot of things. They uphold most of the infrastructure for the web and cloud. And they’re a massive logistics company.
But do they stand to be a major beneficiary of AI?
They can certainly sell more cloud compute. But they’re already using robotics in their warehouses.
Maybe they can lean on AI more to enable autonomous drone-based delivery of individual packages. Maybe they can replace all the humans in the warehouses.
I’ve seen AI-generated product review summaries on the Amazon store. But also there’s tons of AI-generated reviews too, and it seems far from being solved.
Basically, I’m not sure they’ll have a huge comparative upside, though they could have the infrastructure to.
Next, the Magnificent Four
NVIDIA:
This one’s kind of obvious. They’re making most of the chips powering the entire AI revolution we’re going through. Companies can’t get enough of these chips. Their chips are being bought and then sold on eBay for 60% markups.
I don’t see them going away any time in the next 2 years and gains will probably continue to aggregate to them.
Their main con is geopolitical: If China attacks Taiwan and TSMC, their goose is cooked.
TSLA:
Tesla is one of those companies that is weird but well positioned.
Do I think Musk is the most stable person ever to run a company?
No.
But is he executing on a futuristic vision? It seems so.
What’s unique about Tesla is they’ve spent the past decade setting up the infrastructure for training autonomous machines to act upon their environment using just vision from cameras, and neural networks.
It’s possible this ends up being one of the few approaches that generalize well.
It doesn’t make a lot of sense to invest if Tesla is only a car company though. Their work with autonomous humanoid robots like Optimus is very interesting.
If they nail the robot’s behavior and autonomy, they might end up creating an extremely helpful household and warehouse robot.
They’re not the only ones in this space. But they’re probably the only ones that:
have a well defined camera - vision - deep learning training pipeline
extremely automated manufacturing process
have huge state-of-the-art battery manufacturing facilities
All of these combined will probably result in a highly cost-competitive, intelligently autonomous humanoid robot.
My other factor here is I took a Tesla for a demo-drive recently. I turned on full-self-driving before leaving my parking space, and the car handled everything flawlessly, including pulling out of the space, pulling into traffic, handling onramps, merging lanes, stopping at red lights, etc.
The feeling I got was that while it’s probably not 100% perfected yet, it’s obviously come very close, and if it does get perfected, millions of people are probably going to want it.
Apple:
Frequently termed the Dark Horse in AI, Apple has a lot going for it.
Your first response to that should be “But Siri sucks!”
And you’re right. Siri does suck!
But I don’t think that’s a reason to count Apple out yet. They have an incredible distribution on their hardware, and have been very forward-looking.
First, every device Apple has shipped for the past several years has had an onboard Neural Engine integrated into it.
This is a part of the chip optimized to run calculations specifically for neural network operations.
Apple has also recently released a new library called mlx, similar to PyTorch that automatically leverages the neural engine more directly, and has enabled incredible improvements in running AI models on Apple devices.
Second: Their unified memory architecture is unbeatable. It allows programs to run AI models in the RAM as if the RAM is VRAM. This means that the only options available today for running massive AI models with tens of billions of params is to either:
Spend thousands on a dedicated multi-GPU rig with enough VRAM
Buy a MacBook
Third: A near-monopoly on TSMC’s most advanced lithography process.
When it comes to chip manufacturing, Apple seems to have negotiated top priority with TSMC.
What this means is that when TSMC puts out a new process that enables the equivalent of smaller transistors, with less energy usage, and more computational power packed in, it’s Apple’s devices that get it first.
That might not seem important, until you consider that there’s a barrage of AI hardware devices coming down the pipeline from various competitors, including Meta.
This isn’t the first time something like this has happened. But what’s likely to happen is that Apple will wait and see which AI device interfaces do the best, and then release their own.
Since they also have access to the best chip process, their device will have the best energy usage and the best onboard neural execution, and will end up being a superior experience.
Hardware is also hard to get right. This is why I’m not fearful of the recent AI hardware upstarts, and I’m not willing to bet on Meta’s glasses – especially in light of the Oculus’ poor performance. Apple’s integrated software + cloud environment stands unique amongst all phone / laptop / watch / tablet providers.
Apple was recently shown to have purchased the least GPUs amongst top tech companies while still pushing their AI research and models forward. I suspect they’re developing their own dedicated neural processing blocks for scaled up training and inference to compete with NVIDIA. If anyone’s well positioned to do it, it’s them, though this is also speculation.
And with iPhone recently getting the transformer neural architecture installed for auto-complete, plus on-device voice cloning. I expect one day Apple will release an “improved Siri” that is obviously far more capable, and at that point people will recognize that Apple has taken the AI revolution seriously.
Microsoft:
This one’s simple in its reasoning as well. They’re major investors in OpenAI. They’ve aggressively pursued placing AI Copilot tools in all of their products.
If OpenAI does well, they get a lot of upside in it. And if a company is building a massive AI future that’s defining a new industrial revolution, then the opportunity for “doing well” is probably larger than most of us can imagine – even when a company is already worth trillions.
Microsoft has executed their AI vision well. They also get early access to OpenAI’s models, and self-host them to resell through their cloud.
They also don’t sleep on their products. Some of the biggest developer tools in the world today are built by Microsoft:
VS Code
GitHub
TypeScript
And GitHub Copilot, their in-editor AI auto-complete tool, has over 1 million paying subscribers — grossing over $100 million per year. For a slightly intelligent coding autocomplete, that’s really impressive.
Basically, I believe OpenAI will continue to do well and beat the competition on model quality and speed, and Microsoft stands to have the biggest upside that the public can benefit from.
So that’s my whole rundown of thoughts here and how I’m positioning myself.
Is it a little sloppy? Yes
Is this mostly based on my everyday experience in dealing with these companies and their products? Yes
Is it very vibes-based? Hell yeah.
If a war with China breaks out for control of Taiwan will I be bankrupt? Probably yeah
anyway, follow this at your own peril.