Speaker · Updated
Erik Townsend: latest market views
MacroVoices
Summary
Erik Townsend is the host and co-founder of the MacroVoices weekly financial podcast, which interviews macro investors and economists for professional and sophisticated investors.
Chorus has 16 views from Erik Townsend across 5 episodes on MacroVoices since September 25, 2026. Their latest calls: bullish on Oklo; bearish on Anthropic.
Where they stand
Their latest bullish or bearish call on each company.
- OkloBullish ·
- AnthropicBearish ·
What they talk about most
The latest views
Newest first. Every quote is checked against the episode's transcript.
Refinery damage from strikes may keep refining capacity impaired for years after the war ends.
Strikes on Russian, Saudi and Iranian refineries are removing refining capacity. Refineries take about ten years to build, so the damage could last well beyond the war. Townsend thinks the news is not picking up on this and that the cracks and product prices could stay tight for years after the conflict ends.
“If we take refineries out and we reduce the global capacity for refining products, ... Refineries take ten years to build. ... the damage we're doing to refining capacity could be permanent. Not permanent, but persistent for many years after the war finally ends.”at 5:01
MacroVoices: MacroVoices #556 Luke Gromen: US Midterms & Beyond ·
An AI credit bust would let China overtake the US in AI, a real national security risk.
Credit-financed AI spending is giving US frontier labs their lead, but as Matt Berry has noted, that lead is only about six months, and China is not far behind. A credit bust in AI would stall US progress and could hand China dominance in AI globally, which Townsend sees as a genuine national security risk and as the policy reason Washington would step in to prevent it.
“if we had a credit bust in AI, it would legitimately risk a... China taking over and basically becoming the dominant power in AI globally, and that really is a genuine national security risk in my opinion.”at 32:57
MacroVoices: MacroVoices #556 Luke Gromen: US Midterms & Beyond ·
Refinery strikes could trigger a global crack-spread crisis that new refinery builds, taking about ten years, cannot quickly fix.
The West already lacked enough refineries before the war, and crude is only useful once refined into finished products. Erik argues that if refineries keep being targeted, the world faces a crack-spread explosion. A new refinery usually takes about ten years to build from scratch, and the global economy cannot go ten years without refining capacity. He says he is disgusted by people cheering refinery destruction in Russia, and notes that EU leaders are openly cheering it.
“That means the whole world is going to have a crack spread explosion problem that I don't think can be solved until you build new refineries... usually it takes ten years to build a new refinery from scratch.”at 22:28
MacroVoices: MacroVoices #554 Dr. Anas Alhajji + Louis-Vincent Gave: Developing Crises in Middle-East and France ·
Backwardation narrowing while flat prices rise signals the market expects the tightness to last a long time.
The back of the futures curve is catching up to elevated front-month prices, so backwardation is narrowing even as the flat price rises. Erik says that is a very unusual way for backwardation to resolve. He reads it as the market realizing the disruption will last a long time, which he links to the refinery problem and the Paul Sankey interview.
“the back of the futures curve is really starting to catch up to the elevated front month prices... It says to me that the market is realizing this is going to last a long time.”at 31:38
MacroVoices: MacroVoices #554 Dr. Anas Alhajji + Louis-Vincent Gave: Developing Crises in Middle-East and France ·
A China-first export ban on refined products could put the rest of the world in a very hard position.
Paul Sankey says most remaining spare refining capacity is in China. China has signaled a China-first approach that would ban exports of finished products and keep them for domestic use, exporting only what keeps the global economy from crashing. Erik doubts China would crash the global economy on purpose, since it depends heavily on exports. Still, he argues such a policy would leave the rest of the world in a very hard place, given that the crack spreads are globally set.
“China has just signaled their willingness to take a China first approach that says, We're going to ban all exports of finished products and keep them for our own use in China... If China were to adopt that policy, it seems to me that they could really put the rest of the world in a really hard place.”at 32:20
MacroVoices: MacroVoices #554 Dr. Anas Alhajji + Louis-Vincent Gave: Developing Crises in Middle-East and France ·
OkloBullish
Oklo's planned 1 GW/yr is woefully insufficient for near-term AI data center demand.
Erik says he has done the arithmetic and sees considerably more demand from AI data centers alone over the next few years than the 1 GW a year Oklo's factory under construction in Texas would supply (about 100 reactors at 10 MW each). Critics view that as overkill, but he thinks it will be woefully insufficient. He also argues that replacing fossil fuel energy would require far more output: at 1 GW a year it would take about 9,000 years, and a 25% share would require about 112 GW a year. He discloses that he is an early investor in Oklo and that it is his largest private equity position.
“I think there's already considerably more demand just from the AI data centers alone over the next few years... some critics think that's overkill. I think it's going to be woefully insufficient.”at 8:53
MacroVoices: MacroVoices #554 Yasir Arafat: Mass Manufacturing Nuclear Power Plants ·
Demand for nuclear would explode once it costs less than coal and gas and deploys as fast.
Erik argues the key threshold is making safe, clean nuclear cost less than coal and gas and deployable as quickly. Because the whole world wants to move off fossil fuels, demand would surge immediately once that is achieved. Today nuclear costs about twice as much as coal and gas, takes about ten years to build, and is hard to finance. A modular plant built on an assembly line for less than a coal plant would be a 'nuclear Henry Ford moment' that could break humanity's addiction to fossil fuels. Erik says this is a long-term view and not something he expects in the next few years.
“when somebody figures out to make safe, clean nuclear energy cost less than coal and gas and be deployed just as quickly, I contend demand explodes overnight”at 17:31
MacroVoices: MacroVoices #554 Yasir Arafat: Mass Manufacturing Nuclear Power Plants ·
Structural enriched-uranium supply deficit is coming in the 2030s, especially for HALEU.
Erik says uranium is cheap today, under $100 a pound, but that will change in the next few years. Centrifuge enrichment plants cost almost as much as oil refineries and take almost as long to build and bring online. Every uranium expert he has interviewed agrees there is a structural supply deficit in the 2030s, before any mass production of reactors. The deficit would be most acute in the HALEU market now forming. He concludes that scaling to hundreds of gigawatts a year would be hard without a smarter approach to enriched fuel.
“Every uranium expert that I've ever interviewed on this program seems to agree that there's a structural supply deficit coming in the twenty thirties, not for a few years, but in the twenty thirties”at 27:20
MacroVoices: MacroVoices #554 Yasir Arafat: Mass Manufacturing Nuclear Power Plants ·
Spent nuclear fuel is wasted energy; reprocessing and burning it is the answer.
Erik says the world has about 250,000 metric tons of spent nuclear fuel in storage and that public fear of it is misplaced. He argues the real problem is wasting perfectly good fuel. Reactor designs from before he was born can burn this material after reprocessing, so the answer is to stop calling it waste, use it as fuel, and build reactors that can burn it.
“What I had no idea of is the crime here is wasting that perfectly good nuclear fuel.”at 36:23
MacroVoices: MacroVoices #554 Yasir Arafat: Mass Manufacturing Nuclear Power Plants ·
Reprocessing spent fuel is proven viable, so shelving it for disposal is a policy failure.
Townsend argues that skeptics should look at the history. The US built a reprocessing facility in the 1970s, but President Carter ordered it shut down just before it was ready, because the issue became politicized. He says the UK has shown reprocessing works: it recovered a large quantity of reactor-grade plutonium that can be burned as fuel as effectively as the HALEU that the US struggles to source. The UK government then decided to spend billions of pounds disposing of that valuable material underground. He presents this as wasteful policy rather than a technical barrier.
“they proved that you can reprocess it and get it to something that can burn in a reactor.”at 42:23
MacroVoices: MacroVoices #554 Yasir Arafat: Mass Manufacturing Nuclear Power Plants ·
China's state-funded nuclear buildout threatens US competitiveness and energy security unless US policy acts.
Townsend argues the main competitor for advanced nuclear is China, not another reactor company. China has a state-funded plan to outspend the sector and already has more light-water reactors planned or under construction than the entire US fleet. It has built a high-temperature gas-cooled reactor and has announced thorium molten-salt reactors for container ships. He thinks China is likely to move into factory mass production of these designs, which is the path Aalo is pursuing. He says the technology was invented in the US, so a competitive disadvantage is forming, and that the energy security of the US is at stake. He conditions the outcome on whether US policy prioritizes the steps Chris Wright has started.
“China is set to win if we don't take policy action to put more priority on doing the things that Chris Wright has already set out”at 56:18
MacroVoices: MacroVoices #554 Yasir Arafat: Mass Manufacturing Nuclear Power Plants ·
Free Chinese open-source AI models will soon be good enough for most corporate problems.
Erik cites Matt Berry's point that Chinese open-source models are only three to six months behind the frontier labs. He says that for most AI users, models like Opus are already strong enough, so the gap matters little. He argues Chinese models may not match US frontier models but will be good enough for most corporate problems, and they are free to run on your own hardware. Running them air-gapped would mean firms never send decision-making data through Anthropic or OpenAI clouds, which he sees as a possible major upset for US labs.
“Opus is plenty strong enough, and it says to me that it's not gonna be very long before open source free AI from China, maybe it's not as good as The US frontier models are, but it's still good enough that it can solve most corporate problems.”at 38:15
MacroVoices: MacroVoices #553 Brent Johnson: Disparate Housewives ·
China's nuclear buildout outpaces the US, and state funding leaves private US nuclear firms struggling.
Erik says China has more conventional nuclear plants planned and under construction than the entire US fleet. He adds that China has a strategy of advanced research into thorium-fuelled molten salt reactors, which almost no private-sector players have attempted. The few private firms trying are struggling against a state-funded Chinese competitor. On this evidence he says the assumption that the US wins a cold war with China is hard to support, since the facts look daunting.
“China's got more [conventional] nuclear power plants planned and under construction than [the] entire US fleet.”at 43:53
MacroVoices: MacroVoices #553 Brent Johnson: Disparate Housewives ·
AI is a white-collar tractor: long-run job-loss fears are ludicrous, though near-term displacement is real.
Erik compares AI to the tractor, arguing that fears AI will destroy the economy or jobs are as misplaced as fears tractors would destroy farming. He calls AI the biggest productivity tool ever and says it will make white-collar workers far more productive. He concedes the near-term harm: as with the field workers replaced by one tractor driver, people can be out of work for a while, which can bring political and local unrest. In the long run, displaced workers learn new jobs.
“I think in the long term, that is ludicrous. That's like saying that the invention of tractors is gonna destroy farming.”at 44:57
MacroVoices: MacroVoices #553 Brent Johnson: Disparate Housewives ·
Free Chinese open models run on own hardware could upset US frontier labs' business.
Erik relays Matt Berry's argument. US frontier labs such as Anthropic and OpenAI produce the best models, but they are expensive and their payback is unclear. Chinese open-source models can be run on a company's own hardware and air-gapped from the lab's cloud, so the company's data is not used to train the next model. Berry expects this to upset the market. Erik says the free Chinese models are about as good as US frontier models were three or four months earlier, and he sees China possibly running a strategy of letting US labs spend heavily on research, copying their models, and releasing them for free to undercut the US AI boom. He calls it a clever argument and says it is speculative.
“what you end up with open source for free out of China is a model that's about as good as the frontier labs in The US were delivering three or four months ago.”at 39:10
MacroVoices: MacroVoices #551 Michael Every: Decoding The Global Geopolitical Puzzle (Part 2) ·
AnthropicBearish
Anthropic's training on user data could push enterprises toward free Chinese open models.
Erik relays Matt Berry's view that Anthropic's terms let it keep and train on personal and business information shared with Claude, such as strategy discussions. Berry thinks that once users tire of their data being mined, they will move to free Chinese models run on their own hardware. Berry expects this to eventually break the business model of Anthropic and of US frontier labs more broadly. Erik presents this as a clever argument and asks what it does to the US AI industry.
“So Matt thinks we get to a point where people get sick of having their data mined and Anthropic is training on your... Matt thinks eventually that blows up, and everybody says, I'd rather just get these Chinese models that are free and put them on my own hardware.”at 39:33
MacroVoices: MacroVoices #551 Michael Every: Decoding The Global Geopolitical Puzzle (Part 2) ·
Shows
- MacroVoices · 5 episodes
Ask your AI what Erik Townsend thinks
Chorus gives Claude, ChatGPT and other assistants every view Erik Townsend has shared on the shows it covers, next to everyone else's. Ask where they stand, what changed, and who disagrees.
Chorus reports what speakers said. It isn't investment advice.