GUY: Good morning, everybody... happy Thursday, May 14th, 2026, and welcome to Morning Signal.

AVA: Morning... if you're listening half-awake, don’t worry, we’ve got coffee in our veins and a slightly alarming amount of semiconductor discourse ready for you.

GUY: Slightly alarming is right. Because today’s big takeaway is pretty simple... AI is not just a tech story anymore. It’s not just a Nasdaq trade, it’s not just a chatbot story, it’s not even just a capex story. It is now a full economic system story.

AVA: Exactly. And I think that’s why today’s podcast set was so interesting. A lot of people are still arguing about whether AI is “real,” and honestly that feels late. The more relevant question now is... what kind of market structure, financing chain, labor market, power grid, and geopolitical arrangement has to exist to keep this thing going?

GUY: Right. So let’s start where we should start... markets and macro. Because over on Excess Returns, Andy Constan said something that I think is the cleanest framing of the whole week. He basically said, look, AI is real... the bubble risk is real too. Those are not contradictory statements.

AVA: And that distinction matters. Because people hear “bubble” and immediately translate it into “fraud” or “it’s all fake.” That wasn’t his point at all.

GUY: Exactly. Over on Excess Returns, Constan’s point was that bubbles tend to form around genuine regime shifts. The technology doesn’t have to be fake. In fact, the strongest bubbles usually are built on something real enough that investors can extrapolate it way too far. He framed this current period as a bubble regime, not necessarily a top-tick call.

AVA: Which is a much smarter way to think about it. A bubble regime means the character of risk changes even if prices keep going up. You can still make money in a bubble regime... but the asymmetry gets ugly fast.

GUY: Right. And he laid out a timeline I thought was useful. Over on Excess Returns, he dated AI’s “Netscape moment” to January 10th, 2023... that Microsoft deepening of the OpenAI relationship. His argument was that date made decades of statistical learning suddenly legible to markets as a monetizable platform. That was the moment investors stopped saying “interesting research” and started saying “this is a commercial stack.”

AVA: And then timing mattered. Because after the 2023 regional banking mess and the SVB episode, financial conditions eased. So you had a real technology discontinuity landing into a market environment that was becoming more permissive again.

GUY: That’s the cocktail. Real story plus easier money plus a small set of obvious beneficiaries. And then over on Excess Returns, Constan said the next acceleration was around Nvidia-related expectations in March, when semiconductor growth assumptions basically moved from something like 60 to 70 percent... to around 100 percent annual growth. That’s not a little revision. That is a duration event.

AVA: Say more on that, because I think people hear “growth went up” and don’t always appreciate why duration is the key word.

GUY: Sure. If a company beats this year, okay, that’s one thing. But if the market starts to assume the current earnings power lasts for years and compounds off a much higher base, the valuation math changes completely. You’re no longer paying for a hot year. You’re paying for a long runway of extraordinary returns. That’s why semis can look expensive even if the near-term numbers keep validating. The issue is not just this year’s EPS... it’s how much future perfection is getting discounted today.

AVA: So the debate isn’t “does Nvidia sell chips?” Obviously yes. It’s “how many years of supernormal economics are already priced into everything around Nvidia... including memory, networking, power equipment, and even national equity indexes.”

GUY: Exactly. And that brings in what Jeff deGraaf said over on RenMac Off-Script. Totally different discipline... technicals instead of macro... but weirdly the conclusion rhymes.

AVA: Right. Over on RenMac Off-Script, deGraaf said their bubble signal for a diversified sector or index is when it doubles within two years. Not a stock. A sector or index. That’s important because it filters out one-off stories and says, okay, this has become broad enough to reflect an actual regime of enthusiasm.

GUY: And he said the SOX had crossed that threshold roughly two weeks ago. That is a big statement. The Philadelphia Semiconductor Index... the SOX... doubling within two years means you’re not dealing with a normal bull market phase anymore. You’re dealing with something historically abnormal.

AVA: But he was careful too. Over on RenMac Off-Script, deGraaf was very explicit that this is not a “short it today” signal. It’s more a sizing signal. Because bubbles can keep inflating. The problem is usually six to twelve months later, not six to twelve minutes later.

GUY: That’s the part retail traders never want to hear. A bubble signal doesn’t mean “sell all your semis by lunch.” It means your expected forward reward relative to risk gets much worse. Upside may continue... but downside starts looking like an elevator shaft.

AVA: He also said something I thought was underappreciated... over on RenMac Off-Script, deGraaf flagged the KOSPI as bubble territory too, and not because Korea suddenly became a broad macro miracle. It’s because Samsung Electronics and SK Hynix now make up 43 percent of the index, versus something like 18 to 20 percent around 2000.

GUY: That’s wild. Because then the national index stops being a country barometer and becomes a memory-and-AI proxy. You think you’re buying Korea... you’re really buying a concentrated view on the memory cycle and AI infrastructure spending.

AVA: And that concentration matters globally. If AI memory demand is real, SK Hynix and Samsung deserve a re-rating. But if memory pricing or AI capex wobbles, that wobbles not just two stocks, but an entire benchmark.

GUY: Which is why deGraaf’s comments on tactical action mattered. Over on RenMac Off-Script, he mentioned Korea’s outside reversal day and the kind of messy intraday volatility you’ve seen in Micron, ticker MU, and the VanEck Semiconductor ETF, SMH. He was basically saying... this starts to look like distribution, not clean trend continuation.

AVA: Hold on though... I don’t fully buy the immediate bearish read there. Volatility around a crowded trade isn’t always distribution. Sometimes it’s just everyone discovering the same earnings duration story at once and then trading around it badly.

GUY: Fair. But this is where I side with deGraaf. When you get a parabolic move in a diversified basket and then the price action gets sloppier, you don’t need to call the exact top to say the easy money has probably been made. That’s the whole sizing point. You don’t have to become a hero. You just stop being oversized.

AVA: That’s fair. And it lines up with Constan’s framing on Excess Returns. He’s not screaming that AI is fake. He’s saying the financing assumptions required to validate these valuations are getting increasingly heroic.

GUY: Yes... and I think that’s the real market story. Over on Excess Returns, Constan’s concern was the funding chain. The hyperscalers have to keep spending enormous amounts on capex. They fund that through free cash flow, cash on hand, reduced buybacks, and increasingly debt issuance. That has second-order consequences.

AVA: And this is where the story gets more macro than people think. Because over on Excess Returns, Constan said Meta, Google, and Microsoft have already dialed back repurchases relative to what they might otherwise have done because AI capex is taking priority. That means the same companies carrying index performance may be providing less mechanical support to their own stock through buybacks.

GUY: Exactly. People have gotten used to MAG-7 stocks being this magical machine: huge free cash flow, huge buybacks, huge growth, pristine balance sheets. AI changes that mix. If Meta, ticker META, Google, ticker GOOGL, and Microsoft, ticker MSFT, are redirecting cash toward data centers instead of repurchases, then the market is losing one of its biggest bid sources.

AVA: And if they issue more debt on top of that, then you get more credit supply too. So the market doesn’t just face valuation risk. It faces asset-supply risk. More bonds from the hyperscalers, potentially more future equity supply if frontier AI firms eventually list... and less buyback support from incumbents. That’s not nothing.

GUY: Right. It’s like the market has been eating from a buffet where the restaurant also keeps buying back its own tables. Now the restaurant is saying, “Actually we need those tables for GPU clusters.”

AVA: That is such a Guy analogy... but yes, basically.

GUY: Thank you. And this is why the macro angle from Forward Guidance was so important. Over on Forward Guidance, Neil Dutta said this is the biggest capex boom of his career... potentially bigger than the late 1990s. That’s a heavy statement from a macro guy.

AVA: It is. And what I liked about Dutta on Forward Guidance was that he didn’t make the lazy argument that AI is just boosting GDP linearly through tech spending. He said the spillovers matter more than the direct national-accounting contribution. Construction employment matters. Equity prices matter. The wealth effect matters. State tax receipts matter, especially in places like California where RSUs and capital gains actually move the fiscal needle.

GUY: Exactly. If a new data center goes up, the GDP accountants capture some of that. But what they understate is the whole chain. Non-residential construction workers get hired. Specialty trade contractors get busier. Heavy and civil engineering gets support. Corporate profits improve at the leaders. Stocks rise. Households with equity exposure spend more. California collects more taxes. That’s a bigger macro footprint than the direct spend line suggests.

AVA: Over on Forward Guidance, Dutta was especially focused on non-residential construction. He basically said... outside healthcare, this is one of the places where payroll breadth has improved, and he ties that directly to the data-center buildout.

GUY: Right. And this is where the economy starts to look weirder than headline GDP suggests. Because Dutta’s other point on Forward Guidance was that the consumer is not actually that strong underneath. Real disposable income growth is around 1 percent or lower. Real consumer spending has been below 2 percent. Aggregate weekly payrolls have been negative over the last three months, which he said is highly unusual.

AVA: That number jumped out at me. Negative aggregate weekly payrolls over a three-month stretch is not what you’d expect if the labor market were reaccelerating. It tells you total wage income generation is softer than the “strong consumer” narrative implies.

GUY: Exactly. And he also said non-farm payrolls have averaged roughly 50 to 60 thousand per month over six months, which is not recessionary on its face but definitely not booming. Wage growth around 3.5 percent on average hourly earnings and the employment cost index says cooling, not reheating.

AVA: So let me frame the contradiction... over on Forward Guidance, Dutta is basically saying the household side is mediocre, but the economy hasn’t cracked because AI capex has become a support beam. That support beam shows up in jobs, earnings, wealth effects, and even public finance.

GUY: Yes. And that means if AI capex growth slows materially, this stops being a tech-sector problem and becomes a macro problem. That’s the key insight. We are no longer in a world where you can say, “Well, semis had a rough year but the rest of the economy is fine.” The rest of the economy is partly leaning on semis and AI infrastructure right now.

AVA: Which also makes the consumer-sensitive equity tape more interesting. Over on Forward Guidance, Dutta said homebuilders, home improvement retailers, and the SPDR S&P Retail ETF, XRT, all “don’t look good.” That’s the market quietly telling you the lower-quality part of the growth picture isn’t exactly healthy.

GUY: Right. This is why broad indices can mislead. The S&P 500 can look okay while the internals say consumer cyclicals are tired. If the leaders are AI capex beneficiaries and the laggards are household-sensitive names, then the macro is less balanced than the headline index suggests.

AVA: And then there’s oil. Over on Forward Guidance, Dutta said the latest jump in rates had very little to do with the inflation report and much more to do with oil markets being well bid because of the Middle East. That’s such an important distinction.

GUY: Huge distinction. Because if rates rise on stronger growth, that’s one thing. If rates rise because oil pushes up long-end yields, squeezes margins, and hits gasoline prices, that’s a much uglier tightening mechanism. It’s like the economy is getting taxed from the outside.

AVA: Right. And Dutta’s inflation view on Forward Guidance was pretty balanced. First three months of the year were bad. Last couple were somewhat better. Shelter had some weird distortion linked to the earlier BLS and OER issues around the October government shutdown. But he wasn’t saying inflation is conquered. He was saying it’s improving too slowly to pull the Fed back into a clean easing bias.

GUY: Which means the Fed is stuck. If oil stays elevated, the long end stays sticky. If the household is soft underneath, the economy doesn’t love that. But if AI capex keeps offsetting the weakness, the Fed also doesn’t get the “something broke” signal it may need to ease aggressively.

AVA: I want to add one subtle but important thing from Forward Guidance too. Dutta said as the US becomes a bigger marginal energy exporter, the WTI-Brent spread narrows, which means domestic consumers feel more of the global oil price. So the old idea that America is insulated because it produces more energy... that’s too simplistic.

GUY: Exactly. Higher exports don’t magically save the household. If Brent goes up and WTI is tracking it more closely, the consumer still gets nailed at the pump. And energy companies have been disciplined. They’re not just ramping rigs overnight because oil spikes. So the pain is immediate, the supply response is delayed, and that’s not a bullish macro mix.

AVA: So if I summarize markets and macro so far... over on Excess Returns and RenMac Off-Script, the message is that semis and AI leadership are entering bubble-regime territory. Over on Forward Guidance, the message is that the broader economy may be leaning on that same capex boom more than people realize.

GUY: That’s it. The leaders are simultaneously validating the story and increasing system fragility. That’s the paradox.

AVA: Okay... let’s pivot to the technology side because the numbers over on Invest Like the Best were honestly jaw-dropping.

GUY: Yeah, take it.

AVA: Over on Invest Like the Best, Anthropic CFO Krishna Rao made the strongest case I’ve heard that frontier AI is now an infrastructure business first and a software business second. He said compute is “the lifeblood of our business,” and that he still spends 30 to 40 percent of his time on compute procurement. Not pricing. Not branding. Not seat expansion. Procurement.

GUY: Which is not how software companies are supposed to sound. Software companies are supposed to say, “Our gross margin scales beautifully and our main constraint is sales efficiency.” This guy sounds like he’s running a utility crossed with an arms race.

AVA: Exactly. And the infrastructure numbers are insane. Over on Invest Like the Best, Rao said Anthropic has a 5-gigawatt commitment with Google and Broadcom beginning in 2027, plus up to 5 gigawatts with Amazon, which he characterized as an over 100 billion dollar commitment, with substantial capacity landing this year and next.

GUY: Five gigawatts is a country-scale number. People hear gigawatts and kind of nod... but that is not “a few extra racks.” That is industrial civilization math.

AVA: Right. And it changes how you think about AI entirely. If you’re underwriting model companies on old SaaS intuition, you’re missing the point. This is not just selling subscriptions. This is locking in power, chips, network, and data-center real estate years in advance.

GUY: And Rao’s balancing line was perfect. Over on Invest Like the Best, he said if you buy too much compute, you go out of business. If you buy too little compute, you can’t serve customers and you’re not at the frontier. That’s the whole game. It’s not a normal inventory problem. It’s existential both ways.

AVA: Which also explains why the financing issue matters so much. This isn’t like overordering office furniture. If you undershoot, you lose the model race. If you overshoot, your balance sheet gets torched. So when investors say, “Why don’t they just slow down a little?”... the answer is because frontier competition doesn’t let you slow down gracefully.

GUY: That’s right. The frontier is not a nice neighborhood. If your competitor secures better compute and you don’t, you don’t drift a little behind... you risk falling off the curve.

AVA: And the demand side seems real. Over on Invest Like the Best, Rao said Anthropic went from roughly a 9 billion dollar revenue run rate at the start of the year to around a 30 billion dollar run rate by the end of the quarter. He said they serve 9 of the Fortune 10 and are seeing net dollar retention above 500 percent annualized.

GUY: Wait really... 500 percent annualized NDR is absurd.

AVA: It is absurd. And obviously run-rate math can be noisy and frontier AI accounting is still frontier AI accounting... but even if you haircut that aggressively, the directional point is still huge. Enterprise usage is not just growing. It’s compounding as customers expand workloads once they see value.

GUY: Which means the bears who say this is all demos and no revenue are getting less and less credible. You can still be bearish on the market structure and bullish on the adoption. In fact, that’s probably the intellectually honest position right now.

AVA: Exactly. And Rao added another detail on Invest Like the Best that I think operators should really pay attention to. Anthropic actively uses Amazon Trainium, Google TPUs, and Nvidia GPUs, and spent years building the orchestration layer to move workloads across them. That’s strategic flexibility, not just vendor diversification.

GUY: That’s huge. Because the company with the best ability to translate dollars into usable compute may win even if its model quality isn’t always number one. If you can arbitrage architectures better, schedule workloads better, and avoid single-vendor lock-in, your effective economics improve.

AVA: He even said over on Invest Like the Best that Anthropic works from the chip level up with Amazon’s Annapurna Labs. That suggests real co-design, not just “we are a happy cloud customer.”

GUY: Right. This is where software people underestimate how industrial this has become. The differentiation layer is no longer just model architecture and researchers. It’s procurement, orchestration, chip collaboration, and power access. It’s almost like AI leaders are becoming mini sovereigns.

AVA: That’s a good way to put it. And pricing strategy is evolving too. Over on Invest Like the Best, Rao said Anthropic kept pricing relatively stable across Haiku, Sonnet, and Opus, but cut the price of Opus with Opus 4.5 because customers were underusing it relative to capability. Then usage increased by more than the price cut would imply, which he explicitly framed as a Jevons paradox dynamic.

GUY: That’s super important. Because one lazy bear case has been “model prices will fall, therefore margins die, therefore the whole thing commoditizes.” But if lower prices unlock much larger workload volumes, total revenue can still expand. Oil got cheaper to extract over history and we used a lot more of it. Same idea.

AVA: Exactly. Demand elasticity matters. If a better model gets cheaper and suddenly becomes embedded in more workflows, then price cuts aren’t deflationary in the simplistic sense. They can be market-expanding.

GUY: So over on Invest Like the Best, the bull case is crystal clear... massive real demand, giant enterprise penetration, infrastructure commitments measured in gigawatts, and usage expanding as prices come down. That’s not vaporware.

AVA: Which is why over on 20VC, Patrick Forquer’s comments about Legora were so interesting. Because Legora gives you the vertical software layer beneath the frontier model race. He said the company reached 100 million ARR at unusual speed, grew from about 40 employees to more than 500, and generated more than 50 million dollars of qualified pipeline in one month after a big brand campaign.

GUY: That tells me awareness, not product-market fit, may be the gating factor in some vertical AI markets. If one campaign can produce that kind of pipeline, the market may have been more demand-constrained by trust and familiarity than by actual usefulness.

AVA: And the pilot conversion number was the real headline. Over on 20VC, Forquer said pilots convert to closed-won at 78 percent, and strong enterprise reps can close deals within 90 days. In legal tech. A historically conservative, slow-moving profession.

GUY: That’s the part that matters. Lawyers are not exactly famous for reckless software adoption. If legal AI can close in 90 days and convert pilots at 78 percent, then we are moving beyond novelty. That starts to look like budgeted operational spend.

AVA: He also defended Legora’s 5.5 billion post-money valuation by saying the true TAM isn’t just the roughly 40 billion dollar legal-tech software market. It’s the one trillion dollar legal services market. In other words, the upside comes if agentic workflows don’t just replace software tools... they eat service revenue.

GUY: I buy that directionally, but I’m a little skeptical on timing. Saying software can attack services is always seductive. The real question is where supervision, liability, and workflow integration slow that process down. Legal is lucrative precisely because errors are expensive.

AVA: That’s fair. But I do think the conversion metrics make the thesis harder to dismiss. And I loved the self-awareness from Forquer on 20VC when he said quota attainment averaged 280 percent last year and comp ratios ran 8x to 12x, which he framed partly as a management miss.

GUY: That’s actually bullish in a weird way. It says the demand wave is moving faster than traditional SaaS planning systems can handle. You don’t accidentally pay people 8 to 12 times comp unless reality is outrunning your spreadsheet.

AVA: Right. It also means investors should be careful with standard software metrics right now. Some of these businesses are operating in environments where forecasting, territory design, and comp plans are all lagging the actual demand curve.

GUY: Which makes comparisons to mature SaaS kind of useless. Okay, but let’s do the consumer side, because this is where the AI narrative gets a lot less clean.

AVA: Totally. Over on the Big Technology Podcast, Joanna Stern was much more skeptical about consumer AI hardware. Her test was very simple and very good... a device needs a compelling use case besides being an AI assistant. She used the Humane Pin as the cautionary example of a single-purpose AI device that basically did nothing people really needed.

GUY: I think she’s right. Consumers do not wake up wanting “an AI device.” They want a camera, music, messaging, glasses, earbuds... something with obvious utility. AI can enhance that. It usually can’t be the entire proposition.

AVA: Exactly. Over on the Big Technology Podcast, Stern said glasses or earbuds can work if they already do something valuable like photos, video, or audio. That implies the consumer AI monetization curve is very different from enterprise AI. In the enterprise, ROI can justify the product. In consumer, AI often has to hitch a ride inside a product people already desire.

GUY: Which is why Apple may be more durable than people think. Did she talk about that?

AVA: She did. Over on the Big Technology Podcast, Stern argued the smartphone isn’t disappearing anytime soon and floated the idea of AirPods with cameras as a plausible “Siri eyes” interface. That’s interesting because it suggests Apple doesn’t need to own the best frontier model to preserve hardware relevance. It just needs the best product wedge.

GUY: That’s exactly right. Consumer hardware is about form factor and distribution. If models commoditize, the device ecosystem owner can still win. Apple, ticker AAPL, has never had to invent every underlying technology first. It has to integrate it into something people already want to carry.

AVA: So there’s a sharp split forming. Over on Invest Like the Best and 20VC, enterprise and infrastructure AI look real, budgeted, and scaling. Over on the Big Technology Podcast, consumer AI hardware still looks like a “show me the actual use case” category.

GUY: And that split matters for valuation. The market should not price “AI” as one homogeneous revenue stream. Selling GPUs into data centers, selling legal workflow automation, and selling ambient AI wearables are three completely different businesses.

AVA: Exactly. Okay... let’s move into geopolitics, because this is where the AI story gets even more physical and frankly more fragile.

GUY: Go for it.

AVA: Over on Geopolitical Cousins, Jacob Shapiro and Marko Papic had a really useful conversation about why markets keep fading Middle East risk. Their basic point was not that oil is harmless. It was that the shock may be delayed rather than immediately catastrophic. They cited an oil path of about 120 dollars in May, 130 in June, and only potentially 150 to 200 in August or September if disruptions persist.

GUY: And that timeline matters a lot. Because 120 is painful. 150 to 200 is where you start getting systemic damage. At that point you’re not just talking about inflation prints. You’re talking about demand destruction, rates pressure, political stress, and broad earnings compression outside energy.

AVA: Right. Over on Geopolitical Cousins, they basically argued markets are willing to look through current tension because inventories and oversupply may be delaying the real crunch, and because the dominant narrative is still AI capex and adoption. In other words, until oil gets into that crisis zone, investors are choosing to focus on the stronger story.

GUY: Which is rational... until it isn’t. Markets can ignore a threat while it’s gradual. They struggle when the same threat starts hitting rates, margins, and households at once.

AVA: They also got into China and sanctions. Over on Geopolitical Cousins, Shapiro said China had instructed domestic companies not to comply with US sanctions on five Chinese refiners linked to Iranian oil, using a 2021 Chinese blocking measure against foreign laws deemed unjustified. He viewed that as a pretty notable signal ahead of a possible Xi-Trump summit around May 14th to 15th.

GUY: Today and tomorrow, basically. That’s not a small calendar item.

AVA: Not at all. But Papic’s pushback was important. Over on Geopolitical Cousins, he said this isn’t best understood as “China versus America” over physical oil access. China buys something like 15 to 30 percent of its crude from Iran and Venezuela, but it could replace some of that with other Gulf barrels if needed. The issue is losing cheap, discounted barrels... not being cut off from oil entirely.

GUY: That’s a really useful distinction. Discount risk is not the same as supply risk. China doesn’t want to lose bargain barrels, but if you actually remove Iranian supply from the global system... he cited roughly 2 million barrels per day... everybody loses because the global price spikes.

AVA: Exactly. Which is why Papic’s bigger point on Geopolitical Cousins was that the US and China are, in practice, on the same side of wanting to avoid a true Gulf oil shock. Despite all the rhetoric, neither side can comfortably absorb a sustained Hormuz-style disruption.

GUY: That’s a fascinating alignment. Because geopolitically they’re rivals, but in crude markets they’re involuntary roommates. And if oil explodes, it hits the same AI-heavy equity complex that’s currently carrying the tape.

AVA: Yes... and Shapiro’s higher-priority risk was actually not oil itself but US-China trade. Over on Geopolitical Cousins, he argued the AI buildout depends on imported components and materials. So even if the US has the capital and even if some domestic power gets built, renewed trade conflict can choke off the physical inputs that data centers still require.

GUY: That’s the hidden bottleneck. People keep talking as if AI is just code and capital. It’s not. It’s copper, transformers, cooling systems, specialized motors, networking gear, and a global chain of obscure industrial inputs. You can have all the money in the world and still get stuck waiting for the wrong magnet.

AVA: Which is exactly where Jacob Helberg’s argument on No Priors comes in. Over on No Priors, Helberg said AI is already “fueling over a third” of US GDP growth, and that the buildout depends on things like copper, cobalt, electricians, precision reducers, server motors, rare-earth magnets, and actuators spread across thousands of supply-chain nodes.

GUY: “Fueling over a third” of GDP growth is a big claim, but directionally I think the message is right. AI is no longer marginal to growth. It’s becoming central to how policymakers think about growth, national competitiveness, and industrial strategy.

AVA: Helberg’s solution on No Priors was what he called “Pax Silica.” Keep state-of-the-art fabs in the US because they’re already in flight, intensely capital-heavy, and talent-constrained. But distribute the rest of the trusted supply chain across allies. So don’t try to put every node in one country. Build an aligned network.

GUY: That strikes me as much more realistic than pure autarky. Full domestic replication of every AI input is fantasy. But a trusted-allies architecture... that’s at least a coherent plan.

AVA: He also stressed over on No Priors that rare earths aren’t especially scarce geologically. The real chokepoint is processing and refining capacity outside China. That’s a very important nuance. The bottleneck isn’t “the atoms don’t exist.” The bottleneck is “the midstream economics have been ceded.”

GUY: Exactly. Mining narratives are often too upstream. What matters is who can refine, separate, process, and price these inputs at scale. If China controls the midstream, then owning a mine somewhere else doesn’t fully solve your problem.

AVA: Helberg said on No Priors that the administration is trying to solve both the supply problem and the pricing problem that makes non-Chinese projects uneconomic, and he said he was incredibly confident that the pricing issue would be resolved before the end of the administration.

GUY: I like the ambition. I’m less convinced on execution speed. Industrial policy loves nouns. The market cares about tonnage, yield, timelines, and return on capital.

AVA: Fair. But he did give some evidence of actual movement. Over on No Priors, Helberg referenced a State Department critical minerals summit on February 4th with more than 55 countries, plus MOUs with dozens of countries. So at least the coalition-building is happening.

GUY: That’s good. Still, MOUs are the LinkedIn endorsements of industrial policy. Useful signal... not the same as operating cash flow.

AVA: That is brutally fair.

GUY: Thank you.

AVA: Helberg also tied AI competitiveness to energy abundance, especially nuclear. Over on No Priors, he said the administration wants to quadruple domestic nuclear supply and is pushing deregulation and incentives to support that buildout. So again, power is not a side quest to AI. It is the quest.

GUY: Exactly. AI bulls who only look at software multiples are missing the picks-and-shovels. Grid equipment, transformers, utility interconnects, uranium, nuclear services, cooling, industrial automation... these are all increasingly part of the same trade.

AVA: And he added one more thing that connects to labor. Over on No Priors, Helberg said any serious reindustrialization has to be highly automated because unemployment is already around 4 percent. He cited Singapore’s autonomous ports and factories. So the next manufacturing buildout will not look like the 1950s. It’ll be robotics-heavy, AI-assisted, power-intensive.

GUY: Which is where I think people are talking past each other. On Forward Guidance, Dutta is saying the manufacturing renaissance is not visible in the data yet. On No Priors, Helberg is saying it needs to happen and here’s how. Those aren’t actually contradictions. One is descriptive. One is aspirational.

AVA: Exactly. And that same industrial sovereignty theme showed up in a completely different context over on In Good Company with Aliko Dangote. His whole strategy was basically... find essential goods Africa imports and produce them locally through backward integration.

GUY: I loved that interview. It’s one of those conversations that sounds obvious after you hear it, which usually means it’s built on something fundamental.

AVA: Totally. Over on In Good Company, Dangote said the early cement opportunity was basically structural scarcity. In Zambia and Congo-Brazzaville, cement sold for around 250 dollars per ton, and in Nigeria customers could pay and still wait three months for delivery. That’s not a cyclical opportunity. That’s a broken market.

GUY: Right. You don’t need a fancy spreadsheet when people are waiting three months after payment for a basic construction input. That’s pure supply failure. And he said Dangote Cement now operates in 14 countries and is targeting 100 million tons of capacity.

AVA: Then the refinery story. Over on In Good Company, Dangote described the 20 billion dollar refinery as a macro bet against imported fuel dependence, FX leakage, and subsidy arbitrage. He said Nigeria endured fuel queues for 52 years despite being an oil producer. That’s the curse of exporting crude and importing refined product.

GUY: And the detail that blew me away was the infrastructure burden. He said they had to build a new port, roads, and water systems capable of treating 440 million liters per day because the logistical base didn’t exist. That sounds extreme until you remember this is what industrial bottlenecks look like in the real world. Missing midstream kills everything.

AVA: Exactly. And that’s why the Dangote conversation matters for AI too. Different geography, different sector... same strategic principle. If you don’t control the processing and logistics layer, your natural-resource endowment doesn’t save you.

GUY: Which takes us nicely into the cross-currents... because the deepest theme today is that validation and vulnerability are arriving together.

AVA: Yes. Over on Invest Like the Best, Anthropic’s numbers make it much harder to dismiss the current AI wave as fantasy. But over on Excess Returns, Constan’s thesis actually gets stronger if the technology is real. The best bubbles are built on true regime shifts.

GUY: Exactly. Pets dot com was silly. Railroads were real. The internet was real. Housing demand was real. The nastiest bubbles are usually not scams. They’re realities financed in a way that becomes unstable.

AVA: And over on RenMac Off-Script, deGraaf’s technical rule is basically the chart version of the same idea. Once a diversified sector like semis doubles in two years, the issue isn’t whether demand exists. The issue is whether the market structure around that demand has become too enthusiastic to absorb itself cleanly.

GUY: That’s it. The market isn’t debating existence anymore. It’s debating how much duration and how much capital intensity can be carried without something snapping.

AVA: Another big cross-current is that AI has become a physical economy story. Over on Forward Guidance, Dutta sees it in construction payrolls and wealth effects. Over on No Priors, Helberg sees it in minerals, processing, and nuclear. Over on Invest Like the Best, Rao sees it in gigawatts, procurement, and chip architecture. Over on Geopolitical Cousins, Shapiro and Papic see it in trade frictions and oil transmission.

GUY: Right. This is the part I think Wall Street still underestimates. People are used to software scaling with almost no bill of materials. AI at the frontier has a giant bill of materials. Power. Land. Cooling. Chips. Packaging. Networking. Substations. Transmission. Skilled trades. Credit. This is not just code on a server someone else manages. This is industrial capitalism in a hoodie.

AVA: That line should be on a T-shirt.

GUY: I’d wear it.

AVA: The other cross-current I keep coming back to is the wealth effect loop. Over on Forward Guidance, Dutta’s point was that households don’t look amazing on income growth alone. But spending has held up because equity appreciation acts like a substitute income stream, especially at the higher end. And that matters because higher-income households drive a disproportionate share of consumption.

GUY: Exactly. AI capex supports earnings. Earnings support stock prices. Stock prices support consumption and tax receipts. That stabilizes the broader economy. It’s a circular system.

AVA: But over on Excess Returns, Constan warns that buybacks are being displaced by capex and debt issuance. So the same companies creating the wealth effect may eventually provide less equity support through repurchases while asking markets to absorb more credit supply.

GUY: Which is why the loop can reverse awkwardly. On the way up, it feels magical. On the way down, it feels like everything breaks at once. Less buyback support, slower capex growth, weaker semis, weaker household confidence, weaker spending... and suddenly the “strong economy” was just a narrow flywheel.

AVA: Hold on though... I don’t think that means collapse is imminent. One thing today’s set made clear is that enterprise demand is genuinely broadening. Over on Invest Like the Best, Anthropic says 9 of the Fortune 10 are customers. Over on 20VC, Legora says 78 percent of pilots convert. Over on Geopolitical Cousins, Papic even mentioned companies already worrying about their OpenAI bills. That sounds like utilization, not theater.

GUY: I agree. And that’s why I’m not in the “AI is over” camp at all. My concern is narrower. Real utilization does not immunize an asset class from overpricing. In fact, real utilization can enable overpricing. If everyone agrees the demand is real, then financing discipline often weakens because no one wants to miss the buildout.

AVA: Exactly. That’s the bottleneck thesis. The risk has shifted from “AI disappointment” to “AI bottleneck.” Maybe demand is there... but power, chips, trade, oil, debt markets, or valuation tolerance become the limiting factor.

GUY: Yes. And oil is the wildcard. Over on Forward Guidance, Dutta said rates have recently been moving more on oil than on CPI. Over on Geopolitical Cousins, Shapiro and Papic think the oil pain may get serious in August and September if crude pushes into 150 to 200. If that happens while semis are already in bubble-regime territory... that’s not a comfortable setup.

AVA: Especially because the consumer doesn’t look robust enough to absorb a big gasoline shock cleanly. Real disposable income around 1 percent, real spending below 2 percent, negative aggregate weekly payrolls over three months... there isn’t a ton of cushion there.

GUY: Exactly. And if the Fed stays constrained because headline inflation gets energy pressure, then higher oil is like a two-for-one problem. It hurts consumers directly and keeps financial conditions tighter.

AVA: Another cross-current I liked was enterprise strength versus consumer weakness in AI. Over on Invest Like the Best and 20VC, enterprise AI looks immediate and monetizable. Over on the Big Technology Podcast, consumer AI hardware still looks like it needs an existing form factor and utility wedge. That’s analytically useful because investors keep lumping these together.

GUY: Right. You should not be valuing AI glasses like legal workflow automation or hyperscale compute. Different adoption logic, different margins, different replacement cycles, different customer behavior. Enterprise buys on ROI. Consumers buy on habit and desirability.

AVA: And then concentration. Over on RenMac Off-Script, deGraaf’s KOSPI point shows how a real memory boom turns into an index bubble when Samsung and SK Hynix are 43 percent of the benchmark. Over on Excess Returns, Constan’s broader concern is that if semis and frontier AI capture too much profit share or GDP share, then “everything else” either has to lose share or the overall pie has to grow in extraordinary fashion.

GUY: Exactly. The market can be right that there will be giant winners. The danger is assuming there can be an infinite number of giant winners all priced for the same perfect future. AI can be big enough to justify Nvidia, Broadcom, some cloud names, some software names, some power names... but probably not every adjacent story at every multiple.

AVA: Okay... let’s finish with what we’re watching, because the next few weeks and months actually have pretty clear catalysts.

GUY: First one is obvious. Over on Geopolitical Cousins, Shapiro flagged the Xi-Trump summit expected around May 14th to 15th. So basically now. If that produces even a workable framework on trade, the market probably breathes a little easier because one of the hidden risks to AI buildout is supply-chain interruption, not lack of demand.

AVA: And if it fails, I think that matters more than people realize. The market is still trading as if AI inputs can mostly be sourced eventually. If trade tensions harden and specific nodes of the bill of materials get caught in retaliation or compliance fights, that can slow actual deployments even if capex budgets stay high.

GUY: Second... oil into late summer. Over on Geopolitical Cousins, the path they cited was roughly 120 in May, 130 in June, and potentially 150 to 200 in August or September. That late-summer window matters because below crisis levels, AI can probably remain the dominant narrative. At crisis levels, oil starts rewriting everything.

AVA: Right. If crude merely stays annoyingly high, markets may tolerate it. If it gets into the 150-plus zone, then you have to think about higher long-end yields, weaker consumer demand, and political pressure on central banks all at once.

GUY: Third... the bubble-risk window in semis and Korea. Over on RenMac Off-Script, deGraaf said the danger period is usually six to twelve months after the doubling signal, not immediately. So for the SOX and KOSPI, I’d watch failed breakouts, breadth deterioration, and more evidence of distribution in names like MU or vehicles like SMH.

AVA: I’d add concentration metrics too. If the leadership narrows even further while broader participation drops, that’s usually not a healthy sign. Bubbles can keep rising, but they often get thinner before they break.

GUY: Fourth... hyperscaler capex growth into 2027 planning cycles. Over on Forward Guidance, Dutta said the growth rate of hyperscaler capex is unlikely to accelerate next year at the same pace as this year. Over on Invest Like the Best, Rao laid out commitments that stretch into 2027. So the key question is whether real workloads keep validating these spend levels... or whether the industry outruns monetization temporarily.

AVA: And related to that, over on Invest Like the Best, Rao said a large portion of Anthropic’s Amazon-related capacity is arriving this year and next. That is a concrete operational test. Can demand absorb the capacity without crushing unit economics? If yes, the bull case strengthens. If not, the market starts asking harder questions about return on invested capital.

GUY: Fifth... the inflation mix for the rest of 2026. Over on Forward Guidance, Dutta expects rents to keep cooling, though more slowly. So one thing to watch is whether shelter disinflation can offset oil enough to stabilize long yields. If it can’t, the Fed stays pinned and duration-sensitive leadership gets less comfortable.

AVA: Sixth... buybacks and debt issuance from the leaders. Over on Excess Returns, Constan’s warning was that AI capex is already crowding out repurchases at Meta, Google, and Microsoft. I want to watch every incremental signal there. If capex keeps climbing and buybacks keep stepping down, that changes market plumbing even before fundamentals change.

GUY: Seventh... critical minerals and processing policy. Over on No Priors, Helberg said he’s incredibly confident the pricing problem for non-Chinese processing gets solved before the end of the administration. If that starts to happen for real, it could rerate an entire set of businesses around magnets, refining, allied processing, grid gear, and maybe nuclear-adjacent infrastructure.

AVA: Eighth... labor spillovers. Over on Forward Guidance, Dutta highlighted non-residential construction as one of the clearest channels where AI is showing up in employment. I’d watch that closely. If those payrolls stay strong, AI remains a macro support beam. If they soften while the consumer is already weak, the economy loses one of its best buffers.

GUY: Ninth... enterprise adoption quality. Over on 20VC, Legora’s 78 percent pilot-to-close and 90-day sales cycles are incredible if they persist. Over on Invest Like the Best, Anthropic’s 500-plus percent annualized net dollar retention is incredible if it persists. The market needs proof that these are durable operating realities, not just first-wave land grabs.

AVA: And tenth... consumer AI product-market fit. Over on the Big Technology Podcast, Joanna Stern’s test was brutally practical... is the device useful without the AI? I think that becomes the filter for every new gadget this year. If the answer is no, it’s probably noise. If the answer is yes, then AI can be additive.

GUY: So if you want the one-line summary for today... AI is real enough to matter to GDP, labor, power, trade, and credit. And that’s exactly why it’s dangerous to treat it as just another growth story.

AVA: Right. The validation is real. The revenues are real. The infrastructure buildout is real. And because all of that is real, the system is getting more dependent on it... which means the bottlenecks matter more than the skeptics expected.

GUY: I think that’s the whole thing. Stop asking whether AI is fake. Start asking what happens if oil spikes, trade frictions harden, buybacks shrink, capex growth slows, or the grid can’t keep up.

AVA: And maybe the most important mindset shift is this... the next major break, if we get one, probably doesn’t look like “chatbots are useless.” It looks like “the financing chain or the physical supply chain hit a wall.”

GUY: Exactly. Different problem. Bigger problem.

AVA: All right... that’s where we’ll leave it for this morning.

GUY: We’ll be watching the Xi-Trump headlines, oil, yields, semis, and every capex datapoint we can get our hands on.

AVA: Thanks for starting your morning with us.

GUY: This is Morning Signal... we’ll see you tomorrow.

AVA: See you tomorrow.