Funding, Fusion, and Fine Print: Inside AI's Wild Summer

Lauren: You Derek: Hey, everyone. Welcome back to Tech Insider Weekly. I'm Lauren. Speaker 3: And I'm Derek. So Lauren, today's episode is basically a money and nuclear reactor sandwich. Derek: That's one way to put it. We've got a startup funding round that broke every record we track. Speaker 3: Right. And one ship company just jumped to an 11-figure valuation in a single round. Derek: We'll dig into what's actually driving all that cash. Speaker 3: Then, switching gears here, nuclear power is having a moment nobody... Nobody predicted. Derek: A thorium reactor you can basically print like a document? Speaker 3: Something like that. Plus a satellite, a fusion bet from Google, and NVIDIA getting into the water-saving business. Derek: NVIDIA, water, and nuclear reactors. Never thought I'd say that sentence. Speaker 3: Oh, you're going to love this part. NVIDIA is also handing startups a new compute deal. Derek: Is it a gift or a leash? We're going to argue about that one. Speaker 3: Well, I already have a take. Derek: Of course you do. And we're closing on something more personal. Who's actually getting hired at these AI startups right now? Speaker 3: Spoiler, maybe not who you'd expect. Derek: Alright, Derek, let's start with the money. Speaker 3: Let's do it. Derek: Okay, so get this, $5.10 billion, that's what flowed into startups worldwide in the first half of 2026. Speaker 3: That's $5.10 billion with a B. Derek: Joanna Teare at Crunchbase News called it a record, and AI is basically the entire reason. Speaker 3: It's not just global either. Joanna Glasner reported on Crunchbase that North America alone hit $392 billion. billion for the same six months. Derek: Okay, but what actually happened here? I sat through two hype cycles at startups I worked for, and both times everyone swore it was different. Speaker 3: And both times? Derek: It wasn't. So what's actually different now? Speaker 3: Compute. Real physical demand. Look at SambaNova. Reuters reported they just raised a billion dollars at an 11 billion dollar valuation. Derek: 11 billion for a chip company? Speaker 3: An AI chip company. Wait, back up. This isn't an app with a nice interface. This is silicon. Somebody has to build the thing that actually runs the models. Derek: Which is a very different bet than 2021 when everyone was funding a grocery app with a raccoon logo. Speaker 3: I remember that raccoon. Derek: We all remember the raccoon, Derek. Speaker 3: Fair enough. Derek: So the money's chasing actual infrastructure. Fine. But my skeptic hat says an eleven billion dollar tag on unproven margins still smells a little bubbly. Speaker 3: Maybe, but look at unicorn production: Blockster counted nearly ninety new unicorns minted just in the first half of this year, most tied straight to AI infrastructure. Derek: Ninety in six months? Speaker 3: Ninety! Chips, data centers, robotics, enterprise software-that's the mix Blockster flagged. Derek: That's a nice story, but show me the numbers behind the numbers. Is that revenue growth or investors terrified of missing the next NVIDIA? Speaker 3: Honestly, probably both. Nobody's got a clean answer. Derek: Right, right, right. Because every time I've seen this pattern up close, it's fear driving the valuation, not the fundamentals. Speaker 3: Maybe, but the fundamentals argument this round is you can't train these models without more chips and more power. That part's real, whether the valuations hold or not. Derek: And power is doing a lot of work in that sentence because all this cash has to land somewhere physical, which Speaker 3: Yeah, Derek: Speaker 3: it's got to plug into something. Derek: is where this gets a little wild because these data centers are hungry in a way regular servers never were. Speaker 3: Hungry enough that people are talking about nuclear reactors again. Derek: Wait, seriously? Actual reactors powering AI? Speaker 3: So how exactly do you keep a data center running when the grid can't keep up? Derek: OK, building on that grid problem, turns out somebody already printed a solution. Speaker 3: Oh, this one's wild. Jowi Morales at Toms Hardware reported a startup 3D-printed an actual nuclear reactor module. Subcritical solid state thorium fueled. Derek: Wait, back up. That's supposed to be my line. OK, subcritical for those of us who skipped nuclear physics. Speaker 3: So it means it can't run away into a meltdown. The reaction needs an outside driver to keep going, factory-built module by module. Dan Robinson at the Register says it's rated for up to 30 megawatts, up to 30 years. Derek: Thirty years without refueling? That's a reactor you install once and forget. Speaker 3: Which is the pitch for data centers. Nobody wants a fuel swap crew next to their server racks. Derek: Okay, but why now? Why is everyone racing toward tiny reactors instead of, I don't know, more solar? Speaker 3: Solar doesn't run at 2 a.m., and AI training doesn't stop at sundown. Constant load needs constant power. Derek: Fair? Switching gears, this next one made me actually gasp reading it. Speaker 3: I know exactly which one. Derek: SpaceX launched the first ever nuclear-powered commercial satellite. Josh Dinner covered it for Space Yesterday. Speaker 3: A satellite Derek: Wow. Speaker 3: that doesn't need sunlight to keep running. Derek: No solar panels at all. How does that even work up there? Speaker 3: Radioisotope power, most likely. Heat from decay converted straight into electricity. NASA's flown that on deep space probes for decades, but a commercial satellite is a different bet entirely. Clearly. Derek: Ah, so in one week we've gone from reactor in a shipping container to reactor in orbit. Speaker 3: Same week. And then there's Proxima Fusion. Derek: Yes, Bloomberg reported Google and RWE just backed this German fusion startup at a 2.4 billion euro valuation. Speaker 3: Google betting real money on fusion, which, let's be honest, has been 20 years away for something like 60 years. Derek: That's serious money for something science fiction writers used to dream up. Speaker 3: Or physicists having a very expensive argument for sixty decades. Derek: That joke never gets old, huh? Speaker 3: Never does. But Google's not funding this out of curiosity. They need power for their own data centers. And fusion never runs out of fuel and never melts down. Derek: So it's not charity, it's supply chain management. Speaker 3: Basically. And speaking of resources beyond power, Reuters reports Speaker 4: that Google is working on a new battery technology that could give electric cars more than twice the distance of the new Tesla. Speaker 3: Traders reported nuclear startup Valar partnered with Nvidia on a data center built specifically to save water. Lauren: Water? I thought this whole thing was about electricity. Derek: It's both. Data center cooling drinks millions of gallons. Pair that with a plant that recycles its own cooling loop, and you solve two resource problems with one build. Lauren: Okay, so reactors handle the power side, but you still need chips to actually do anything with it. Derek: Which is the other half of the equation, and Nvidia just made a pretty interesting move on exactly that front. Speaker 3: music Lauren: Okay, flip that on its head for a second, because reactors in fusion solve the power problem. Derek: Right, but chips are the other half of the math. Lauren: Exactly, and Nvidia just made a move I need you to translate for me. Derek: So Storyboard18 reported this. NVIDIA's rolling out a revenue-sharing model for AI startups and cloud providers. Lauren: In English, please? Derek: Okay, for the non-engineers in the audience, instead of paying full price up front for GPUs, A startup gets access to NVIDIA's compute, in an exchange NVIDIA takes a cut of whatever revenue that startup makes down the road. Lauren: So it's less buy the hardware and more rent now, split the winnings later. Derek: Basically. And their CFO, Colette Kress, framed it as giving startups faster access to infrastructure they couldn't otherwise afford. Lauren: Faster access or NVIDIA just found a way to get paid twice. Derek: Well... Lauren: No, seriously, what does NVIDIA actually get out of this? this. Derek: Bloomberg's take was pretty direct. They called it a play aimed at quote aspiring AI startups, basically the companies that can't write a nine-figure check for chips today. Lauren: Which sounds generous until you realize Nvidia's now got equity-like exposure to every company it powers. Derek: Yeah, BeInCrypto actually used the phrase deepening its ecosystem grip. That's their read, not mine. Lauren: Deepening its grip. I mean, sure, call it a lifeline. I'd also call it a leash. Leash. Derek: Wait, back up. It's not that different from how VCs already work, right? Cash for a piece of the upside. Lauren: Except Nvidia is not just an investor here. They're also the landlord, the supplier and now the equity partner. That's three seats at the table. Derek: When you put it that way. Lauren: Right? And remember the funding numbers from earlier. Startups are sitting on more cash than ever. Derek: Sure, but a huge chunk of that cash gets eaten by compute costs before it even touches paper. Such as payroll. Lauren: Which is exactly why this deal is tempting. You're cash rich on paper but GPU-poor in practice. Derek: So do founders take it or do they see the trap? Lauren: Honestly depends on how desperate they are for compute versus how much control they're willing to give up. Derek: Yahoo Finance framed it as Nvidia making it easier to get compute power. Benzinga used almost the same language. Easier access, advanced GPUs, no giant check. Lauren: Easier for who, though? The startup or Nvidia's balance sheet? Derek: Luckily both, probably. That's kind of the point of a good deal. Lauren: I'll believe it's good once we see a founder walk away from one. Derek: Fair. Nobody's turned it down yet. Lauren: The thing that gets me, startups are burning cash on chips instead of people. Derek: Yeah, which raises a whole different question: who's actually getting hired at these companies right now? Lauren: Ooh, now that's a conversation I want to have. Derek: Because it turns out the answer might not include very many junior engineers at all. Lauren: Great, another thing for new grads to panic about. Let's get into it. Pivoting Pivoting a bit, let's talk hiring because the compute obsession has a headcount cost too. Derek: Right. In Harvard Business School and INSEAD just put a number on it. Their study found AI-native startups run about 25% smaller than traditional companies at the same valuation. Lauren: Twenty five per cent smaller; that's a different org chart entirely. Derek: Owais Sultan wrote about this for Hackread: "Startups are building global teams with contractors and EOR partners instead of stacking headcount in one office. Lauren: Lean and distributed. Okay, I get the instinct-I ran ops at a startup where we hired eight people for a job three could do, and it nearly sank us. Derek: So the Lean logic makes sense on paper. Lauren: But there's a catch. Fortune's Emma Burleigh reported these companies are skipping entry-level hires almost entirely, chasing senior engineers with top-tier degrees. Derek: Business Insider covered the same Harvard data: fewer junior hires, flatter teams, more people who already know what they're doing. Lauren: And that's exactly the mistake I made early on. Hire only proven people, and you never build anyone yourself. Derek: Which raises the real question for me as an engineer: If nobody's hiring juniors, who trains the next generation of engineers? Lauren: The senior people we all fought with didn't start senior. Derek: Somebody gave them a shot on day one. Lauren: So if every AI startup skips that step, where does the next Derek come from? Derek: Don't put my name on that." Fair, Lauren: "but seriously, five years from now, does the industry have a bench or just a shrinking pool of people who already made it? eat it. Speaker 3: Mm-hmm. Lauren: Okay, so today went from reactor in a shipping container to reactor in orbit, and honestly, my brain is still catching up. Derek: Same. And underneath it all, one thread. Compute needs power, and power needs new thinking. Lauren: Right, and that funding number we opened with isn't just hype. Somebody's actually building the physical stuff behind these models. Derek: Which loops back to hiring. Leaner teams, fewer entry-level seats. Worth watching where that goes. where that goes. Lauren: Definitely a thread we'll keep pulling. Derek: If you enjoyed this one, subscribe wherever you listen and leave us a review. It helps more than you think. Lauren: Got a founder we should talk to or something we missed? Send it our way. Tag us online. Derek: New episodes every Wednesday. Lauren: Thanks for spending this time with us. Derek: See you next week.