Welcome to Cautious Optimism, a newsletter on tech, business and power.
📈 Trending Up: Automotive chip availability … public corruption … Nvidia sales in China? … military operations in Mexico? … sports betting … Michael Grimes
Startup of the Day: Starcloud, which just shot a GPU into space. See my interview with the company (under its old name) from January.
📉 Trending Down: MAGA polling … Russian income … Ukraine’s ability to hit Russia … Sam Altman’s operating rep (more) … GOP criticism of tech companies … academic freedom
Things That Matter
Quote of the Day:
Throwing more money into Silicon Valley doesn’t yield more great companies.
The investor also added that throwing money can instead make “it harder for us to get the small number of special companies to flourish.”
I wonder if the investor is right about the pace at which the economy generates “great companies” in the long term. Historically, Botha is correct, but if AI is going to shake up as much of the world as the venture community thinks, wouldn’t the market change faster, thus increasing the pace at which great companies are created?
But I argue that the largest tech companies are so large, so vertical, and so horizontal that they’re eating up an increasing portion of new markets, which leaves some space for new, great startups to scale into.
If that’s the case, our lack of interest in breaking up the largest tech shops is directly harming startup growth and economic disruption.
Animoca to eat Currenc: Continuing our coverage of crypto companies going public via reverse-merger deals, Animoca Brands (crypto VC, consulting, tokenization) intends to combine with Currenc Group (fintech AI).
The deal will see Currenc divesting some of its current work to make room for Animoca. The former is a small (revenue of $8.71 million in its most recently reported quarter) and unprofitable (net loss of $4.7 million in its most recent quarter) business. Animoca, in contrast, reported revenue of $314 million in 2024, with EBITDA of $97 million.
I don’t mind profitable companies going public (though Animoca’s numbers are listed as “unaudited, non-IFRS figures”). But I do find it strange when companies that appear to be in good enough shape to list, do not. Why?
Given the history of reverse mergers (SPACs and the like), I have a guess.
We’re finally seeing self-driving accelerate: Back in April, Waymo said that it was seeing 250,000 paid trips each week. That’s around 1,500 per hour, for reference. Today, Baidu’s Apollo Go service said it has reached the same milestone.
Waymo grew from around 50,000 weekly paid rides in May, 2024, to about 5x the volume in a year. However, the company isn’t talking about more recent metrics — maybe it’s the timing of its announcements; or maybe the company’s growth is lumpy because of the phased introduction of new vehicles and markets.
The horse-race reporting here has only so much value; what matters more is that we’re seeing two different companies reaching material scale. Meanwhile, Wayve is racking up positive press, Zoox is expanding from Las Vegas into San Francisco, and Tesla’s effort is set to reach 1,500 cars in two markets by the end of the year.
At this point, it’s clear we will not see any one company crack self-driving so much sooner than its rivals that it ends up owning the entire market. China, the United States and Europe (Wavye) each have their champions.
Those are the very countries/blocs with the leading AI model labs today. China has a bunch, the United States has a bunch, and Europe has Mistral, Black Forest Labs and Stability AI.
It’s probably a good time to ensure that our European allies are in good health.
What about Amazon? Our Friday issue got stuck in the gears, but we’d be remiss if we didn’t have a short note on AWS’ performance, given our focus on cloud demand:
Don’t expect Amazon’s investments to slow down. The company’s CEO Andy Jassy spent the first chunk of his earnings call spiel riffing on Amazon’s AI progress, including that AWS was growing at a pace not “seen since 2022”, and that it was “reaccelerating to 20.2% [growth] year-over-year.”
Amazon’s CFO said later in the call that the cloud platform’s quickening growth was “driven by strong growth across both our AI and core services and more capacity which has come online to support customer demand.”
AWS’ backlog is growing too — it reached $200 billion by the end of Q3, not including “several unannounced new deals in October, which together are more than our total deal volume for all of Q3.”
The traditional cloud giants are united: More capacity is needed and cap-ex is not about to decline, but we could be facing a new impediment to AI compute that isn’t spelled ‘N V I D I A.’
The AI bottleneck is changing
News that Nvidia is now allowed to ship its most powerful chips to the UAE helped boost its shares slightly in early-morning trading today. More markets equals more sales, which is good for a growth-hound like the GPU giant.
But we may be entering a new era of the AI race, one in which power, not chips, is the true gating function.
This morning, Microsoft signed a $9.7 billion deal with IREN, a neocloud, to purchase “access to NVIDIA GB300 GPUs over a five-year term,” and Redmond is prepaying 20% of the contract value. Why would Microsoft need the deal when it is spending heavily on its own data center footprint? From IREN:
[The deal] marks another major step forward for IREN as we continue to expand large-scale GPU deployments across our 3GW secured power portfolio in North America, reinforcing our position as a leading AI Cloud Service Provider.
Three gigawatts of electricity is enough to power a few million homes, and IREN says it has the juice “secured.” We care about power availability because companies like Microsoft are having a hard time finding enough.
From a recent podcast (BG2) interview, here’s Satya Nadella:
[T]he biggest issue we are now having is not a compute glut, it’s power — it’s sort of the ability to get the builds done fast enough close to power. So, if you can’t do that, you may actually have a bunch of chips sitting in inventory that I can’t plug in. In fact, that is my problem today. It’s not a supply issue of chips; it’s actually the fact that I don’t have warm shells to plug into.
This is part of why MENA is working as hard as it is to become an AI hub. The region’s power generation capacity is enormous thanks to both hydrocarbons and sunlight. Similarly, IREN has a large data center footprint in British Columbia, where hydro power is common.
The United States is behaving half-smart, half-dumb in the face of a power crunch. On one hand, we’re making state funds available to nuclear projects, and moving to make connecting to the grid simpler and faster. At the same time, we’re closing the door on wind and solar projects. We should be encouraging all of it — every last possible watt — at once. China, in contrast, is building wind, solar, nuclear and, yes, even more coal-fired power facilities.
Humans are going to run low on AI inputs across several vectors. Chips? Sure. Power? Certainly. Water? Without a doubt. We could feel disappointed by the situation, because why are we not able to bring more AI compute online faster? Or we could take a more optimistic outlook: Demand will lead to dramatically more chip, power and water resources.
The only part of that sunny take I worry about is water. We can build more chips. We can generate and distribute more power. But water? That’s tricky. Cooling is a requirement for data centers, and we can’t build them all in the Arctic Circle.
But humans are so smart we use a form of sand to think, so I don’t think anything will truly stop the great capitalistic wheels in their quest. Though if you wanted to get spicy with it, you could bet on the various companies that may ease future hyperscaler capex investments. They might do rather well.