Late August 2026 Random Ramblings

Summary

• Walker’s macro worry is that today’s market may be running the 2010s in reverse: the 2010s gave stocks a “beautiful double whammy” of strong earnings growth plus multiple expansion off a suppressed starting multiple, but now Treasury rates are “the highest they’ve been in 20 to 30 years” (10-year ~4.5%, 30-year higher, “the Treasury Secretary is intervening and buying the long end”). The equity risk premium is currently ~4.5%—roughly in line with history—but if rates keep rising and the ERP re-widens toward the 6% it hit in the 2010s, “P/E multiples are going to fall a heck of a lot.”
• He also questions whether the ERP should be additive rather than proportional: in his examples, a 6% equity return over a 2% Treasury is 3x the risk-free rate, while 10% over 6% is only ~66% more. He wonders whether that would mean the ERP should compress when rates are low and expand when they rise—an amplifier on multiple-contraction risk—while stressing that this is “more macro than I normally talk about” and not foolproof.
• The rate move directly attacks AI data-center economics: the generally described structure is a 15-year lease with two 5-year tenant options, where the lease NPV roughly covers the build cost (his example: $5 billion) plus a little risk capital for the developer fee—the real bet is the terminal value, which is “quote-unquote free.” If rates go from 4% to 5%, lease rates need to rise ~5–10% to compensate, and the discounted terminal value shrinks because it sits 15–25 years out.
• The downside case hits data-center developers twice: “none of these companies were here 15 years ago,” so a 15-year lease can die in year six via tenant bankruptcy, and releasing risk is brutal—a $100 million/year NOI contract with AMD might relet to a Bitcoin miner or other next-best player at roughly 2023 economics for “$10 million per year,” while the prices these data centers charge are up about 10x from Bitcoin-mining levels. Standard terminal-value math (inflate NOI, apply a 10x multiple, discount back) depends on AI demand continuing.
• Finance 101 closes the loop: rising rates crowd out investment, and since data centers and power are a material portion of the AI buildout, higher rates could slow it—which would itself trigger the credit and terminal-value risks. “I don’t think it’s going to get there if interest rates tick up another 20 basis points, but at some point it would have an effect.”
• Two case studies of CEOs running capital allocation for personal balance-sheet needs: UWMC (Q2 hedging loss on a deal not under contract, a distressed raise, and short-seller allegations covered by Hunterbrook that dividends were needed to pay for the Phoenix Suns) and Cogent (CCOI), where Walker thinks David Shafer kept paying a huge dividend as leverage rose to help fund taxes on his Cogent stock and RSUs and provide cash for his margin-called Washington, D.C., real-estate portfolio. Both were founders; Walker flags management misalignment as rare but a major red flag he’s now screening for.
• On spotting “the next Mark Leonard”: early track records often mask hidden risk—“you can get great returns in 1 year by levering up and yoloing something… I’m looking at you, Situational Awareness”—and many admired managers simply “found one great theme and rode it,” like the 2000–2008 oil geniuses you never heard from again. Mark Brad Jacobs stands out to Walker for having succeeded with multiple roll-ups in different industries, while he asks how much of Mark Leonard’s success reflects Leonard versus his vertical-software insight. If the “SaaS apocalypse” fears of April 2026 had arrived in 2014, “does history look a lot different?”

Deep dive

1. If the 2010s re-run in reverse, the multiple is the casualty

• Walker’s setup, from his equity-risk-premium post that morning (recorded Wednesday, August 26): in the mid-2010s, 2% Treasuries plus a historical ~4% ERP should have implied ~25x P/E, but stocks “hovered in the mid-teens”—the ERP had blown out from 4% to 6%. Stocks then did “low teens annualized for 10 to 12 years” on strong earnings plus expansion off that suppressed multiple.
• He attributes some of the decade’s earnings growth to the Trump 1.0 tax cuts, which he thinks took the corporate rate from roughly 35% to about 21%, with additional investment effects as marginal projects became more profitable.
• Today’s mirror image: the 10-year is “in the 4 and 1/2% range,” the 30-year higher, the Treasury Secretary “intervening and buying the long end to kind of suppress rates,” and the ERP sits ~4.5%—around average. His question: if rates keep rising and the ERP also re-widens back to 6%, “P/E multiples are going to fall a heck of a lot.”
• His secondary puzzle—should the ERP scale with rates at all? A 6% equity return over 2% Treasuries is 3x the risk-free rate; 10% over 6% is only ~66% more. He wonders whether that would mean low rates compress the ERP and rising rates expand it, potentially worsening the downside. He hedges throughout: “I’m not going to say that’s foolproof… this is more macro than I normally talk about.”

2. Rising rates rewrite the data-center lease math

• The structure Walker lays out, used by many former Bitcoin miners that flipped to data centers: a 15-year lease to a hyperscaler, NVIDIA, AMD, CoreWeave, or another tenant, with two 5-year tenant options, where the lease NPV covers the build cost (“if the plant is going to cost 5 billion to build”) plus a little risk capital for the developer fee. The real prize is the terminal value—releasing in 15–25 years is “kind of quote-unquote free for them.”
• Rates going roughly from 4% to 5% do two things: lease rates need to rise “somewhere between 5 to 10%” to preserve NPV, and the discounted terminal value shrinks because it is 15–25 years out and “you’re discounting it back to today.” Nobody builds for cost-of-capital recovery alone: “what person is going to do that?”

3. The double hit if the AI trade rolls over

• Credit risk first: “none of these companies were here 15 years ago… you might have built this big project thinking you had a great tenant for 15 years and in year six, they’re bankrupt.”
• Then releasing risk. When AI demand remains strong, the tenant is likely to be in the money on its option. The terminal-value exercise—a $100 million NOI contract with AMD, inflated to $120–200 million, at a 10x multiple, discounted back—depends on those economics continuing. If the “bubble” bursts (Walker’s qualification: “I don’t think it’s a bubble, but obviously it gets pretty frothy sometimes”), the next-best tenant might be a Bitcoin miner or other player paying roughly 2023-level rates: “it would have been going for 10 million per year.” The prices charged by these data centers are about 10x what Bitcoin miners had charged.
• The circularity that fascinates him: rising rates crowd out investment as a matter of finance 101; data centers and power are a material portion of the AI buildout; and a slowdown could activate both credit and terminal-value risks. “We’re not there yet… I don’t think it’s going to get there if interest rates tick up another 20 basis points, but at some point it would have an effect.”

4. When the CEO’s balance sheet drives the company’s capital allocation

• UWMC’s Q2: a “massive bath on a hedging loss” tied to a deal the company was never under contract for at any point in Q2, followed by a distressed raise. Short-seller allegations covered by Hunterbrook said the large dividends were being paid because the founder needed them to pay for the Phoenix Suns. Walker says, “I don’t know if that’s true or not,” and suspects there might also have been “a little bit of cowboy” behavior.
• Cogent is the clearer disclosure case: as it integrated the Sprint deal and leverage rose, CEO David Shafer continued paying a huge dividend long past the point when it made sense. Shafer disclosed margin loans and a Washington, D.C., real-estate portfolio on which he was being margin-called. Walker thinks the dividends helped fund taxes on his Cogent stock and RSUs as they vested and provided cash for that portfolio. “Go pull up the Cogent stock chart… doesn’t look great.”
• The pattern he’s now hunting is “classic management misalignment”: rare, but a major red flag when management appears to be running the balance sheet for its own needs rather than shareholders. Both leaders were founders, which can bring the attitude, “this is my company, this is my baby, I can run it the way I want to.” It’s also, he notes, one of the reasons activists exist.

5. How would you actually spot the next Mark Leonard?

• The prompt: a friend’s text about a CEO everyone called “the next Mark Leonard” three to five years earlier. The friend, who knew him—“good guy, really sharp”—graded him “somewhere between a B and an A-minus,” while Mark Leonard would be an A-plus, and laughed at the comparison.
• Walker’s warning is hidden risk. Like “next Buffett” profiles based on the first 10 years of a career, early results can fail to show the risk being taken: “you can get great returns in 1 year by levering up and yoloing something… I’m looking at you, Situational Awareness.” With Leonard, even a decade can be hard to judge because a stock may have started very low and become inflated, and such records are volatile.
• He connects this to judging management more generally. Management teams are great salespeople, and investors—including Walker—can accept an explanation such as “this was a one-time thing” until a year later, when they conclude, “Oh, these guys are liars.” Personal rapport and looking management in the eyes are not enough to separate a durable track record from a good story.
• Many admired operators “just found one great theme and rode it”—the 2000–2008 oil geniuses stopped being heard from when the factor stopped. Mark Brad Jacobs, who Walker says is struggling at QXL’s building-products roll-up, stands out because he succeeded with multiple roll-ups in different industries.
• Even with Leonard, Walker asks how much of the genius was Leonard versus the vertical-software insight. It was a great bet; “if he had chosen coal, it wouldn’t have been as great.” By the end of August 2026, many SaaS companies were bouncing back from the April fears, but Walker still has “real terminal-value questions.” If the “SaaS apocalypse” had happened in 2014 instead of 2026, “does history look a lot different?”