Amazon, Microsoft, Google, CoreWeave, Nvidia, and even Tesla - AI news

Investment Analysis Clubs Saul’s Investing Discussions
,
1

Don’t let the Legos on his bookshelves in the background fool you - Nate Jones has insights on AI that are useful to us investors (and anyone in a career path that develops or uses AI, or might use AI). Case in point are his two most recent YouTube videos:

Here are my take-aways:
Amazon:
• Revenue up 13%YoY to $180B, Net income up 38%, AWS growing 20%, etc.
• But, Amazon’s cash flow actually went negative!
• Why? Their yearly Capex hit $125Billon - the highest of any company on earth. Ever. And 75% ofthat going directly to AI infrastructure. ($83B in 2024 Capex, btw)
• Amazon even took on $12B in debt.
• So when CEO Jassy claims the massive lay-offs of 10% of their white collar workforce, 30,000 people, is about “company culture” and Pandemic hiring bloat, that’s true, but it’s hiding the real reason.
• Jones claims this shows us “what’s coming for the entire tech industry.”
• 30K white collar jobs saves the company $6B year, which turns their currently negative FCF for -$4.8B into postiive FCF.
• Jones cites a BofA study of the top 5 hyperscalers (Amazon, Microsoft, Google, Meta, and Oracle) and found that their combined Capex is 94% of their Operating Cash Flows.

  • “Human headcount is being converted to compute capacity.”
  • “Salaries are being transformed into silicon.”

• Bottom line: “Amazon is under financial pressure not because the business is weak, but because the AI infrastructure race demands capital at a scale that even Amazon’s profit engine struggles to generate.”
• The perception is that Amazon was slow to AI with AWS, and has lost ground MS’s Azure, Google’s GCP, even Oracle. I’ll add that the whole rise of the NeoClouds (Coreweave, Nebius, etc.) is due to legacy cloud providers not jumping whole heartedly into providing AI cloud services well.
• CEO Jassy talked about company culture and Pandemic hiring bloat, but they already did 27k layoffs in 2022, so why wait 3 more years to do more? Jones says Jassy doesn’t want to admit financial pressure to Wall Street.

But, this isn’t just an Amazon problem, it’s an every hyperscaler problem. How do you fund massive Capex while maintaining profitability? But cutting everywhere else. Amazon and others are cutting jobs because they’re losing money, they’re some of the most profitable companies in history. They’re cutting to fund AI Capex, since that’s what management sees as necessary to be profitable in the future.

Microsoft and Meta (Facebook) :
• Microsoft itself cut 15,000 jobs
• Meta continues its “year of efficiency”
• Both beat Wall St. expectations. Both announced high Capex spending will continue (Meta nearly doubling Capex, for instance).
• Yet Meta was up 10% while MSFT was down 10%.
• Jones thinks the difference is that Meta owns its own AI and can point to how it improves its business, while Microsoft gets its AI from partners like OpenAI (Oh, and Azure revenue was down 1% last quarter).
• “Spending on AI is fine; spending on AI you don’t own is a different story.”
• By not owning their own AI, they’re exposed to other companies’ execution. Which is riskier.

Apple and Google (and Anthropic) :
• Apple chose Google, somewhat because Anthropic was nickel and diming them with powering Siri with Claude, whereas Google was a good prior partner for search, so getting Gemini on good terms was the better choice.
• Jones thinks investors will like the Gemini deal since they haven’t had faith Apple could do AI on their own.
• Google’s huge search profits have helped their AI development.
• Kind of ironic that a year ago, many were worried about Google getting disrupted with search due to AI, but instead we see Google’s doing just fine with AI AND search.
• Anthropic closed a funding round valuing the company at $350B - double what it was just last Sept. That’s half the market cap of Walmart.
• Anthropic’s model is more trusted than many others, which should be sticky. They have a “safety focused development.”

Coreweave and Perplexity:
• Nvidia invested $2B in Coreweave at $87.20/share (now $93).
• Coreweave using the money to add 5 GW of capacity by 2030.
• Perplexity gets $750M from Microsoft
• Jones says the question for investors is whether the AI Infrastructure build out that’s happening is sized correctly - is it too small, too large, or just right? We won’t know for sure for years, but “none of the hyperscalers want to risk” not building out enough.

Tesla is now an AI company selling cars:
• Ending Models S and X, converting the assembly lines to build Optimus robots, starting at the end of this year.
• Grok is deployed across Tesla’s vehicles.
• Tesla investing $2B in xAI.
• More than a million paid FSD users
• Robtoaxi service expanding in H1 of this year.
• Tesla’s “future lies in AI and robotics, not premium sedans.” They have higher margin opportunities in robotics."
• Tesla spending $20B in Capex for AI, autonomy, and robotics.
• “A transition that most companies are just beginning to navigate.”
• "We’ll see whether Tesla can execute, but they have never been afraid of jumping 5 years into the future making a bet on where they think the market is going.

I found both videos to be worth watching twice.

63 Likes
2

Thank you for the summary, very enlightening.

How do we use this information to inform our investment decisions?

  • The AI infrastructure build out is still white hot. What is the best way to be invested going forward?
  • What else besides AI infrastructure should we be investing in or getting out of?
8 Likes
3

Well, that’s up to everyone’s approach and views on what the future will be. There’s nothing here that’s directly actionable, in my view, but some thoughts:

  1. Is the reaction to Microsoft right? Do investors really think MSFT won’t be managing their partner relationships properly? I suspect it’s really just an OpenAI question - if you believe in OpenAI being successful, then Microsoft will be successful. If you think Sam Altman will screw things up, then avoid MSFT, too. But, if you think the Azure growth last quarter was a blip, maybe it’s a buying op.

  2. Is the reaction to Meta right? Zuck has been wrong on the Metaverse, but he’s still throwing money at it. Does what the company gets out of AI fully compensate?

  3. Google has been a good partner for Apple for Search - if they’re the same for AI, will Apple benefit enough to not be in control of their own AI?

  4. Everyone is worried about AI hallucinations and the ramifications. Does Anthropic’s “safety focus” means they will eventually be more successful?

  5. Coreweave and Nebius have talked up their AI native cloud offerings, but most of their deals are plain ole infrastructure. Infrastructure isn’t sexy, but it’s hard - permitting for power alone is a gauntlet to be run. Are these companies getting enough for their troubles, or are they locking in the picks and shovels for the AI gold rush? AND, if you think the latter, then maybe a company like IREN is the real golden play since it has no pretentions and has locked in power.

  6. Is Amazon doing what it has done in the past - which is, spend lots on infrastructure that looks to hurt profits, only to repeat huge profits from those investments later? Maybe it’s a buy on major weakness?

  7. Finally, while the debate on TSLA will not be settled for years, that the company is turning the aircraft carrier to a new heading does mean that investing in the company means investing in autonomy, AI, and robotics. If you were investing because of EVs, you should get out now. If you’re not comfortable investing like a VC in some future business that may or may not pan out, don’t get in. In my view, Elon is almost always right directionally, but is always wrong on the timeline.
    When he’s wrong directionally, such as with the Cybertruck, he fails big-time. Cybertruck is both the wrong product (what was delivered is way below the promised spec and way above the promised price), and was delivered years late. A robust EV pickup business, which would have happened if Musk hadn’t thrown all sorts of new tech and new styling into the product, and instead had just concentrated on range, charge time, and towing capacity, the CT would be selling well and the transition narrative wouldn’t look as bad as it is. Tesla’s not going to be putting up good financial numbers for a couple-few years. Are you really ready to invest like a VC?

29 Likes
4

These are great questions. The problem is, for the average investor who doesn’t work in these industries, we really have no way to make anything more than an educated guess. All we can rely on are the actual numbers, and what management tells us. Would love to hear from anyone here who has direct experience working with these businesses.

8 Likes
First-Hand Industry Knowledge and its Lawful Application
5

SemiAnalysis is a great site for information on semiconductors - mostly now AI chips and memory. They built an in-depth benchmark and ran it across lots of servers and software versions. It’s quite technical, but here’s the link if you’re technical and brave:

For more of a layman’s bottom line take, here’s the head of SemiAnalysis, Dylan Patel, being interviewed. I suggest watching the whole thing, but I’ve cued to a good spot if you don’t have time comparing Nvidia to AMD and others, like Cerebras:

“AMD will stay in single digit market share.”
“You have to specialize because you’re never going to beat Nvidia at its own game.”
The danger with that is “it’s not clear if AI models will still be in that realm” when their silicon comes out. IOW, Nvidia has the general purpose chip, whereas companies specializing to gain some performance advantage are depending on AI models and implementations being what they are today, which is certainly not guaranteed.
“That’s why Nvidia has a portfolio of chips now - not just a single GPU line.”

Interesting tidbit later on - the 2 largest GPU renters are:

  1. OpenAi
  2. ByteDance (Chinese company behind TikTok and Seedance AI video generation). ByteDance can’t buy Nvidia GPUs, but they can rent them from server farms that Oracle runs in Malaysia. So, maybe not impacting Nvidia as much as might be thought (but still a lot).

Patel sees Gen AI revenue hitting $100Billion this year (about 39:30 in the video he gives a break-down), up from $1B in 2023. He cites a stat that 2% of GitHub commits are Claude Code (and that underestimates since that’s only commited by Claude Code, not including the commits by developers using Claude Code), and later says 5% of code is AI generated - today - so with $2T of software wages each year, that’s still $100B right there.

“As long as the models keep getting better we’re not in a bubble.” The Capex spend results in better models, which leads to more adoption and more value earned.

“Why would I hire an L4 engineer? I’d just tell Claude to do it. Low level knowledge work just doesn’t matter. Why would I use Excel when I can just tell Claude to manipulate CSVs.”

And then, he captures the state of constant change in the AI world right now:
“That’s just [Anthropic’s] Opus 4.5. I think OpenAI’s new model will be better than Opus 4.5 and it’s coming in March-ish. OpenAi has a better RL stack [Reinforcement Learning] than Anthropic, it’s just their pre-trained models suck compared to Anthropic’s. … Flip-side, Google has a better pre-trained model than Anthropic or OpenAI, but their RL stack sucks, so if they catch up on RL…These models are going to get ridiculously [good]… and then Anthropic obviously advancing, everyone’s advancing.”

“It’s a whole new world.”

Yeah, it’s like he can’t talk fast enough to get it all out.

BTW, at 47 minutes in, he talks about how people blamed Microsoft and Nebius’ data center for power prices going up, but the reality, Dylan Patel says, is simply that was the fallout from Hurricane Sandy many years ago with NJ infrastructure that had to be rebuilt and paid for.

20 Likes