1Washington moved against the long bond and the long bond moved back
The 30-year Treasury yield closed at 5.23% on 20 August, after touching 5.31% on the 17th. That is the highest the long bond has been since before the 2008 financial crisis.
Mid-week, and outside the usual quarterly refunding calendar, the Treasury announced it was upsizing its buyback programme at the long end. Joseph Wang, on Fed Guy, treats the timing as the whole story: adjustments like this are normally saved for the quarterly announcement, so an unscheduled one is a response to conditions the Treasury does not like. Secretary Bessent described the move on CNBC as a "Treasury twist" — issuing shorter paper to retire longer paper, shortening the duration of debt outstanding to push long yields down.
It worked for a day. Yields fell about nine basis points on the announcement, then gave the entire move back over the following two sessions as oil kept climbing.
Wang's conclusion is that this was a first skirmish rather than a defeat. He lists what remains unused: upsizing the buybacks further, cutting long-end issuance sizes outright, leaning on banks to hold more Treasuries, directing Fannie Mae and Freddie Mac to buy bonds, and — far down the list — involving the Fed. He notes Japan cut its 40-year auction sizes and got a sharp drop in 40-year yields that lasted weeks before fading. "I'm really not worried about yields going higher anymore," he says, with one condition attached: the cleanest fix is an end to the war in Iran, after which he could see the long bond below 5%.
Liz Ann Sonders, Chief Investment Strategist at Charles Schwab, was asked the same question on Excess Returns and gave the opposite answer. "I think it is the fundamentals," she said, calling the announcement "an attempt at jawboning" that "worked for a day, working less well today," and adding that the Treasury is "attacking the symptom, not really the problem." Her list of causes is structural: massive debt issuance from the AI buildout competing for the same capital, Japan diversifying away from Treasuries and catching down to China's twelve-year trend, foreign private investors finding better yields at home, and a term premium rising as investors demand more to lend long.
The two are not describing different facts. They agree the announcement moved yields for a day and then stopped working. They disagree about what that proves — whether an unused toolkit means the problem is solvable, or whether the tool failing on first contact means it was the wrong tool. What would settle it is the next attempt: if the Treasury upsizes again and the effect fades faster, Sonders is right. Bessent holds a press conference on Monday, which Wang framed as the test of the Secretary's resolve.
Both of them, from opposite directions, land on the same dependency. Wang needs the Iran war to end. Sonders needs debt to grow more slowly than the economy, and says plainly that there is no plan on the table to make that happen.
2A third of next year's S&P earnings growth rests on two chip stocks
Sonders put a number on the concentration that the "broadening out" story is supposed to have fixed. Citing a recent Leuthold study, she said accounts for 18% of the S&P 500's expected year-over-year earnings growth for 2026, and adds another 14% on top. Two chip stocks, 32% of the index's expected growth. Extend to the top ten earnings contributors and you reach 65% of total S&P 500 earnings — and by her account the ninth and tenth names are not tech at all, but and .
The breadth is real on one measure and absent on the other. Ten of eleven sectors saw estimates improve through second-quarter reporting, and the blended growth rate went from 24% expected at the start of the quarter to more than 50%. Sonders calls that a parabolic acceleration in both the growth rate and the surprise factor, with only two precedents — coming out of the global financial crisis and out of the COVID recession. Both of those were base effects, arithmetic rebounds off compressed earnings. This one is not: last year's earnings were already strong.
Her concern is what happens when the arithmetic stops. The surge is tied to AI spending, hyperscaler capex is pointed at roughly $900bn next year, and "at some point depreciation is going to catch up and we're going to have an inflection point in the growth rate."
She declined to give a view on any individual stock, noting she does not cover them. But she offered a mechanism that matters this week. Samsung beat sell-side consensus by a handy margin and still underperformed the unpublished buy-side expectation — and because Samsung and SK Hynix are roughly half the index, the KOSPI drew down around 40%. The spread between published sell-side consensus and unpublished buy-side hope is wider now, she argues, than anything she has seen. Where a report lands inside that spread is what determines the reaction, not whether it beats.
reports on Wednesday after the close, against $2.13 of expected earnings and $93.6bn of expected revenue. On Sonders' framing, the consensus number is not the bar that matters.
One more distortion worth flagging: she noted that the recent listing may be inflating headline earnings-growth statistics, and said it is hard to parse how much.
3Electricity is the binding constraint on AI, and the three companies legally able to sell it have been marked down
Felix, of Felix & Friends, built an episode around a laboratory in Melbourne where roughly 200,000 lab-grown human neurons wired to a chip learned to play Doom. His argument is that the experiment is not a science story but a scarcity signal: the entire biocomputing industry has raised less money than Microsoft spends on a single year's electricity contract, so the interesting fact is not the brain but what the richest companies are willing to try.
The numbers he assembles are the substance. A single rack of AI servers draws about 100 kilowatts, as much as a couple of hundred homes. Gartner projects power shortages will restrict — not slow, restrict — 40% of AI data centres by next year. In PJM, the largest US electricity market, the capacity auction price went from $28 to $329 in two years, and would have gone higher had the regulator not capped it on inflation grounds.
What the hyperscalers have done in response is the evidence: signed a 20-year contract to restart Three Mile Island, the first commercial nuclear restart in US history, reportedly at about twice the current rate; signed a 17-year deal with Talen; has locked up as much as 6.6 gigawatts through 2035. You do not sign a 20-year power contract for a two-year problem.
The trade he draws from it rests on a legal fact rather than a technological one. Most US nuclear plants are owned by regulated utilities that cannot sell output directly to a private buyer. Only a handful sit in deregulated markets where private 20-year contracts are possible, and by his count three companies own a meaningful fleet: , and . Competing with that would require changing legislation in about a dozen states.
The part that makes it interesting rather than obvious is the price action. All three have been marked down — roughly 31% off its high, about 33%, around 20% — while the contracts were being signed. His reading is that the story got less exciting without getting less true: the customers did not cancel, the law did not change, and the turbine queue got longer.
He attached four risks himself, and they belong with the call. A data centre announcement is a claim on power, not a fact, and much of what has been announced will never be built. If AI spending slows, the demand curve flattens. The moat is regulatory, and Microsoft is lobbying. And nuclear timelines slip — the Three Mile Island restart is targeted for 2027, and a delayed restart is delayed revenue.
He also named as a way to own the constraint without picking the winning generation technology: its gas turbine backlog has gone from 83 to 116 gigawatts against roughly 20 gigawatts of annual production, and its grid business carries a $42bn backlog into a global transformer shortage.
This claim and the one above it pull in opposite directions, and the tension is not resolved. Sonders' warning is that AI capex is what is inflating earnings growth and that depreciation eventually bites. Felix's thesis needs that same capex to keep arriving for twenty years. If she is right about the inflection, the power contracts still pay — they are signed — but the marginal new contract stops coming.
4The ceasefire is holding and the Strait is still not reopening
Prediction markets are pricing peace and blockade at the same time, and the gap between them is the most informative thing on the board.
On Polymarket, the US-Iran ceasefire holds through 31 August at 93.5%, through 15 September at 84.5%, and through 30 September at 75.5%. The ladder decays the way you would expect, but every rung is priced as more likely than not.
The Strait tells a different story. "Strait of Hormuz traffic returns to normal by December 31" trades at 31.5% on $9.0m of volume, and has fallen 13 points over the past week — the sharpest move in anything checked. The September 15 version of the same contract sits at 1.75%, effectively ruled out.
So the market expects the shooting to stay stopped and the waterway to stay broken. A ceasefire is not a settlement, and cargo does not move on the absence of hostilities alone.
That gap is what connects this section back to the first one. Wang's route to a long bond below 5% runs through an end to the Iran war — and the contract that would have to reprice for that to happen is the one moving hardest in the wrong direction. Meanwhile oil rose on Friday as sanctions on Iran's partners were threatened, Reuters reported NATO members discussing options for the Strait without alliance involvement, and Iran granted passage to a number of Iraqi tankers.
The administration's own posture points the same way. Truth Social posts over the window promoted an "economic D-Day" operation against Iran's shadow economy, alongside an EPA emergency fuel waiver framed as increasing domestic supply and lowering prices — pressure on the supply side abroad and relief sought at the pump at home.
Separately, and reversing the direction of the past week, the trade picture with Canada broke. CNBC reported on 22 August that US-Canada trade talks collapsed and a wave of new tariffs follows, and on Sunday the President posted that Canada "wants the benefits of being a State, without being one" and has charged US farmers tariffs for years — "No more!!!" A week ago the same file was being described as a pause and a pending deal.
What the sources recommended
Four single-name calls came out of three videos. All four came from one source, all four sit inside a single thesis, and the person making them attached four separate risks before he finished — so this is one argument with four expressions, not four independent signals.
Felix & Friends named , and as the only companies able to sign private long-term power contracts with hyperscalers, explicitly on a regulatory moat rather than on fundamentals — he described the underlying businesses as unimpressive on paper and the case as "a story on future earnings." He added as a way to own the electrification build-out without picking which generation technology wins. Each call carried his own caveats: unbuilt data centres, an AI spending slowdown, lobbying against the regulatory moat, and slipping nuclear timelines.
There were no sell or avoid calls from any source this window.
Fed Guy and Excess Returns made no single-name calls at all. Sonders was explicit that she does not cover individual stocks and could not comment even if she wanted to; her tickers appear above as evidence, not as recommendations.
The thesis gets marked to market on Wednesday
Bessent holds a press conference on Monday. Wang framed the coming week as a test of whether the bond market challenges the Secretary's resolve, so Monday is the first read on whether the buyback was an opening move or the whole hand.
reports Wednesday after the close, and on Sonders' framing the question is not whether it beats the published consensus but where it lands against the unpublished buy-side number — the same gap that took roughly 40% out of the KOSPI on a Samsung print that technically beat. Given that and together carry a third of next year's expected index earnings growth, the report is a test of the concentration argument above more than of one company.
The rest of the week: and Monday, and Tuesday, , , and alongside on Wednesday, then , , , and on Thursday.
Nothing on the macro calendar rivals those. The September Fed meeting is on 16 September, and the market has already moved a long way toward pricing a hike into it.
