Buffett Bets on Beige (And Other November Surprises)
When a 95-year-old value investor throws $5 billion at a tech company, you should probably pay attention. When Jeff Bezos decides retirement is overrated and jumps back into operational leadership, that's also interesting. When these things happen on the same day that someone finally admits AI data centers are about to break the electrical grid... well, that's just November 17th, 2025 for you.
Let me walk you through today's tech headlines and explain why you should care. Or more accurately, why I care, and you get to read my opinions whether you like it or not.
The Oracle of Omaha Goes Digital
Berkshire Hathaway just disclosed a $4.9 billion stake in Alphabet. That's 17.85 million shares of Google's parent company. This is Warren Buffett—a man who famously didn't understand tech, avoided it for decades, and whose biggest tech bet (Apple) he's currently trimming.
So what changed?
Nothing, really. Alphabet isn't a tech bet. It's an advertising bet with AI characteristics. Buffett understands monopolies. He understands moats. And Google's search monopoly combined with its AI-powered ad targeting is about as wide a moat as you can find in the digital world.
The timing is interesting though. This is supposedly one of Buffett's final major investment decisions. He's handing the reins to his successors, and one of his last moves is to say "Yeah, AI matters, but pick the company that makes money from it, not the ones burning cash to build it."
What This Actually Means
When conservative capital moves into a sector, it usually means one of two things:
- The smart money is arriving – the sector has matured enough that fundamentals matter again
- We're near the top – retail and institutional money has bid up prices to the point where even conservative investors feel compelled to join
I honestly don't know which this is. But Alphabet trading at a discount to Nvidia and Microsoft while actually turning a profit from AI? That's classic Buffett. Buy the undervalued monopoly, not the overvalued promise.
Bezos Is Bored Again
Jeff Bezos is back. Not as an investor (he's been doing that). Not as a board member (yawn). As co-CEO of Prometheus, an AI startup focused on industrial manufacturing and complex engineering.
Let that sink in. The man who built Amazon, stepped down, bought a yacht the size of a small city, and presumably spent his days doing whatever billionaires do when they're not running companies... got bored. So he's running a company again.
This tells you everything about where Bezos thinks the real money is. Not in consumer AI. Not in chatbots. Not in yet another SaaS tool that helps marketing teams "leverage synergies." Industrial AI. The stuff that optimizes manufacturing, improves supply chains, and makes actual physical things better/cheaper/faster.
Why Industrial AI Matters More Than You Think
Consumer AI is sexy. Everyone can try ChatGPT. Everyone has an opinion on whether AI art is "real art." It's accessible, democratic, and generates endless LinkedIn posts about productivity hacks.
Industrial AI is boring. It's optimization problems in factories you'll never visit. It's logistics algorithms for supply chains you don't think about. It's engineering simulations for products you use but don't understand.
And it's where the actual money is.
Consumer AI has millions of users paying $20/month. Industrial AI has hundreds of customers paying millions per year. Do the math. Bezos did.
The fact that he's willing to do actual work again (as opposed to just writing checks) suggests he thinks this could be bigger than even those numbers imply. Or he's just pathologically unable to relax. Probably both.
Sovereign AI: Because Not Everyone Wants American AI
Sakana AI just raised $135 million to build Japanese AI models. Not models that speak Japanese—plenty of those exist. Models built specifically for Japanese culture, Japanese regulations, Japanese sensibilities.
This is what "sovereign AI" means, and it's going to be huge.
Right now, AI is dominated by American companies (OpenAI, Anthropic, Google) and Chinese companies (DeepSeek, Baidu, etc.). Everyone else is either using those models or training smaller models that try to compete on the same terms.
Sakana is taking a different approach: build models that are culturally aligned from the ground up. Train them on data that reflects local values, local regulations, local expectations of what AI should and shouldn't do.
Why This Matters
Do you really think the EU is going to be comfortable with American AI making decisions about European citizens forever? Do you think Japan wants Chinese models handling sensitive government data? Do you think any country with the resources to build their own AI ecosystem will resist doing so?
We're heading toward a fractured AI landscape. Not because of technology limitations, but because of geopolitics, regulations, and cultural preferences. Sakana's $2.65 billion valuation suggests investors agree.
The downside? We might end up with a dozen different AI ecosystems that don't talk to each other very well. The upside? At least we won't have all our AI eggs in one American or Chinese basket.
The Energy Crisis No One Wants to Talk About
Here's the inconvenient truth buried in today's headlines: AI data centers are consuming electricity faster than we can build renewable capacity to support them.
Those trillion-parameter models everyone's so excited about? They require server farms the size of small cities. Those inference workloads that make AI useful in production? They run 24/7, consuming power at rates that make cryptocurrency mining look quaint.
And we're just getting started.
Every company wants their own AI models. Every startup needs compute. Every hyperscaler is building more data centers. The demand curve is exponential. The power grid improvements are... not.
What Happens Next
One of two things:
- We build more power plants – nuclear, probably, because renewables can't scale fast enough and fossil fuels are politically toxic
- AI growth hits a wall – not because of technology or money, but because there's literally not enough power to run the computers
My money is on option 1, with a side of "AI companies start building their own power generation." We're already seeing hyperscalers invest in modular nuclear reactors. Give it five years and Google will have its own nuclear plant. I'm only half joking.
The irony is delicious: the technology that's supposed to help us solve climate change is consuming so much power that we need to build nuclear plants to support it. Turns out saving the planet requires a lot of electricity.
The Danish Unicorn You've Never Heard Of
Quick one: Flatpay, a Danish payment processing company, just hit unicorn status with a €1.5 billion valuation.
Their innovation? Flat-fee card terminals. That's it. No fancy AI. No blockchain. No revolutionary technology. Just transparent pricing for small merchants who were tired of getting nickel-and-dimed by payment processors.
They went from 7,000 customers to 60,000 in a year. Annual recurring revenue north of €100 million. All because they solved a real problem in a straightforward way.
I mention this because it's a useful counterpoint to all the AI hype. Sometimes the best business opportunity isn't the sexiest technology. Sometimes it's just fixing something that's been broken for years and everyone accepted as "how things work."
Payments processing has been a predatory business model for decades. Flatpay looked at that and said "What if we just... didn't do that?" Turns out there's a billion-dollar company in that question.
What Does It All Mean?
If I had to synthesize today's news into a coherent narrative, it would be this:
AI is transitioning from speculation to infrastructure.
Conservative investors are buying in, but they're picking the established players with actual revenue. Smart founders are going operational again, but they're focusing on industrial applications with clear ROI. Countries are building sovereign AI capabilities because they recognize this is critical infrastructure, not just another tech trend. And the physical constraints (power, cooling, hardware) are becoming more binding than the algorithmic ones.
We're past the "wow, AI can do things!" phase. We're entering the "okay, how do we build this at scale and who pays for it?" phase.
That's good news if you're building real businesses. It's concerning news if you're still pitching vaporware with an AI sticker on it. And it's expensive news if you're trying to build this stuff without hyperscaler resources or billionaire backers.
The Personal Take
I've been in tech long enough to recognize inflection points when I see them. This feels like one.
Not because of any single headline, but because of the pattern: serious money moving in, serious people coming back, serious infrastructure challenges emerging, and serious questions about governance and sovereignty getting asked.
The AI hype cycle isn't over—there's still plenty of nonsense being funded. But the serious AI cycle is beginning. The part where it becomes boring infrastructure that powers everything but nobody thinks about.
You know, like electricity. Which, ironically, we now need a lot more of to run the damn thing.
Final Thoughts
If you're building an AI company: focus on industries with money and problems. Consumer AI is crowded. Industrial AI is wide open.
If you're investing in AI: follow the Buffett playbook. Buy the picks and shovels, or buy the companies that already make money from it. Avoid the pure-play research companies unless you have venture-scale risk tolerance.
If you're using AI in your job: start thinking about power consumption and infrastructure costs. That $20/month subscription is subsidized. The real question is what happens when it's not.
And if you're just trying to keep up with the news: bookmark this post. Six months from now, when Prometheus is valued at $10 billion and we're talking about AI-induced power shortages, you can say you read about it here first.
Or don't. I'm not your mother.
But seriously, the next few years are going to be wild. Buckle up.