Energy Tech Meets Timescales: Notes on MacroVoices #548 and the Real Bottleneck in the Nuclear Renaissance

Though the road is short, you will not arrive without walking it; though the task is small, it will not be done unless you do it.

—— Xunzi, “On Self-Cultivation” (Warring States period; my own translation)

What the episode is about

MacroVoices episode 548 (2026-09-03) features Dr. Carly Anderson, a chemical engineering PhD who now runs an early-stage energy technology fund. Host Eric Townsend is a fission partisan, and within minutes he tells his guest that he can’t see how fusion gets cheap or arrives in time. They argue about it for the rest of the hour.

I like interviews like this, because disagreement forces the criteria into the open. When two people who both understand the technology argue, the argument isn’t about who knows more — it’s about what standard you use to decide whether something will actually happen. The standard that came out of this one travels well beyond nuclear.

The main points

1. What opened this market wasn’t technology — it was a buyer. Anderson puts it plainly: a nuclear renaissance needs somebody willing to sign a 20-year power purchase agreement at $100 per megawatt-hour, and that somebody “didn’t exist until a couple of years ago.” American utilities are small and hyper-local; one might operate two or four plants. Hyperscalers were the first entity able to create an order book for a dozen plants at once and make the projects financeable.

2. “Factory production” and “mass production” are different words. Eric stops to draw the line: when nuclear people say factory production they often just mean assembling off-site and shipping it in. He wants assembly lines, robotics, repeatable quality control — what Henry Ford did to the automobile, done to the reactor. Anderson doesn’t fully take the bait. She puts her weight one step earlier: get a reference facility running. Only then does the order book become writable and the cost curve become something you can slide down.

3. The fuel chain is stuck somewhere most people don’t look. After uranium is mined it has to be converted into a form that can be enriched. There are about five facilities in the world that do this. The American one is old, and much of the capacity sits in Russia. Executive orders call for getting off that supply by 2028, and in Anderson’s words there is currently no alternative, with multiple companies racing to fill the gap. Order books solve demand; this one is a physical constraint.

4. Lasers may be next in enrichment, but it isn’t settled. From gaseous diffusion to centrifuges, and now to lasers that excite uranium-235 without touching 238 — in principle far fewer units. Anderson mentions a Texas company starting from lithium metal, which sidesteps fluorine chemistry. Her line about fluorine stuck with me: people know it from the dentist, but the other way fluorine interacts with human bone is nasty. The interesting part is the spillover — the same processing chemistry improves rare earth separation.

5. Supercritical CO2 turbines: the near-term value is bolting them onto old equipment. Steam turbines run around 30% thermal efficiency. Swap the working fluid for supercritical CO2 and you can reach 50%, with much smaller hardware — and small is what makes mass manufacturing and fast iteration possible. But Anderson points to a nearer use: attach one to existing equipment as a bottoming cycle and unlock another 10 to 20% of generation from the same asset. That matters more than the new-build case, because it doesn’t require waiting a decade.

6. The fusion argument comes down to tritium. Eric swings the market price: $30 million per kilogram, so the cheap-fuel claim collapses. Anderson’s answer is closing the fuel cycle — a lithium blanket around the plant, where the products of fusion hit lithium and break back down into tritium. The actual external input becomes deuterium, which is in seawater. You can disagree with her, but note the timeline both of them land on: early-to-mid 2030s, because that’s also when the larger American fission plants come online.

7. Data centers turned the question from “is there enough power” into “is the power the right shape.” Anderson describes a 40 MW data center as roughly six Sam’s Clubs, and now the industry wants a megawatt — enough for about a thousand homes — delivered into a single cabinet. The harder part is shape: the grid side needs a smooth curve, while AI training loads swing from 80% to 20% on a one-second timescale. The box that reconciles those two is where solid state transformers and silicon carbide come in.

Going further

”These technologies all sound great — how do I tell which one is actually coming?”

This is my recurring problem with shows like this. Every idea has a PhD behind it, a successful demo, and a five-year timeline to changing the world. Five years later, one or two happened.

The episode handed me a usable filter: find the layer that’s stuck, then check whether anyone is working on it.

Take nuclear apart and the constraint isn’t the reactor. The order book gap got filled by hyperscaler contracts. The regulatory gap has a new pathway. The power conversion side has supercritical turbines. And conversion? Five plants worldwide, an aging American one, capacity in Russia, a 2028 deadline, no replacement yet. That layer will set the pace for the next several years.

The same question travels. If everything around a story is moving and one layer is unsolved, that layer is your indicator. If the person telling you the story only describes the destination and can’t name the constraint, you’re probably too early to judge. The mistake I’ve made runs the other way: buying because the headline story was compelling, then spending two years waiting on a step nobody had started.

”I keep getting the technology right and the timing wrong”

The thing that made me stop the episode was this: two people arguing opposite sides gave the same date.

Eric backs fission, Anderson backs fusion, and they go at each other — but lay out the timelines. A large fission plant starting today: seven years if lucky, honestly ten. First microreactors online around 2028 to 2029. Fusion plants, in her view, mid-2030s. The gap between the two positions is smaller on a calendar than it sounds in an argument.

What I take from that: “is this technology right” and “when does this happen” are separate questions, and the second one decides more of your outcomes. A technology can be entirely correct and still lose you money by arriving after your patience runs out. Anderson named her firm TimeScale for exactly this reason — she says you need to get technology to market fast for it to matter.

So when I hear a long-horizon story now, I add a question: between today and actual cash flow, what has to happen, how long does each piece take, and is any piece something I can’t track? The untrackable piece is the reason to size the position smaller. Her reference-facility logic is the same idea from the other side: she isn’t asking for proof that the technology works, she’s asking for proof it can be repeated — that’s when timelines start to converge.

”The episode named a lot of companies — should I be buying them?”

Most of them are private, early-stage venture positions that you and I can’t buy. Which I think is the most useful thing about this episode: it forces attention away from “which ticker” toward “which layer benefits.”

Anderson demonstrates the shift herself. Talking fusion, she slides into magnet materials, laser modules, simulation software. Talking geothermal, she slides into drill bit materials — and notes the same hard, thermally conductive material shows up in mining. Talking silicon carbide, she lands on wafering, solid state transformers, and rack-level power management. She likes these adjacencies: the market grows the market, you learn faster, and you reach new customers without rebuilding a manufacturing base.

She also drops a line I expect to remember for a long while: simulation and modeling compressed design cycles from years to hours, and that is what makes deep tech an investable sector at all. That isn’t a comment about one company. It’s a claim that the risk profile of an entire category got rewritten.

Worth a look

• MacroVoices episode 548 itself, published 3 September 2026
• TimeScale Ventures — public writing on their site and Substack
• The NRC’s Part 57, the alternative licensing pathway for advanced reactors; the documents are public
• For the power-shape section, public technical material on the move toward 800-volt distribution inside data centers

The one thing to take with you

What I kept from the whole hour is a single question: before committing, find out whether anyone has completed the full version once.

When Anderson talks about reference facilities, she isn’t really talking about technology. She isn’t trying to prove a reactor core produces heat — that was settled long ago. She wants one plant, built end to end, connected to the grid, with books you can read. Before that exists, every cost estimate is inference. After it exists, you learn which step actually takes longer than expected and which one turns out easy.

Here’s something I’ve tried, and it has nothing to do with nuclear: take one thing you’re pushing on right now — a career change, a skill, starting to exercise, repairing a relationship — and go find one person who has completed it end to end. Don’t ask for advice. Ask about the process. How long did it take, which step took the longest, and was there something they expected to be hard that wasn’t. If you can’t find a single completed example, that information is useful too: either you’re doing something genuinely novel, or your path is stranger than you assumed.

When I do this, the step that turns out to be the bottleneck is usually not the one I prepared for.

Raw Markdown

This article is an educational discussion of investment method. It is not advice to buy or sell any individual security, offers no target prices, and does not analyze any current holding. Investing carries risk; make your own decisions or consult a qualified professional.