enesptjazh-Hanszh-Hantfrdeitkoruarhiidtrnlplvithsvhemsukcselrohudafinobn
The 10-year hit a 2002 high and El-Erian put part of the bill on AI borrowing
The 10-year Treasury touched 5.34% on October 1. El-Erian blames hyperscaler debt, CoreWeave says its margins outrun rates, Ackman sits in T-bills.
The 10-year Treasury yield touched 5.34% on October 1, the highest since 2002. The day before, Mohamed El-Erian went on Fox Business and said something most bond commentary has stepped around. Part of the reason the government is paying more to borrow, in his telling, is that the companies building AI data centers are borrowing too, and they are bidding for the same buyers. Over the same two days the people on the other side of that trade gave their answers. A neocloud CEO said his margins outrun the coupon. A hedge fund manager said he only wants the risk-free end. A bank CEO said her borrowers have not noticed yet. I think the answers add up to a picture worth spelling out, and one of the numbers in it does not hold up.

El-Erian says the AI build is crowding the bond market
Larry Kudlow wanted to argue that a capital goods boom is disinflationary. El-Erian agreed over the long run and then moved the conversation to supply. In September alone, he said, yields rose 50 basis points across the curve. Kudlow asked whether that was just real growth. El-Erian said yes, and something else.
Just to give you a sense, hyperscalers borrowed $200 billion last year, they're going to borrow a trillion this year. And it comes at a time when the Gulf countries are buying fewer Treasuries. Why? They're spending more at home. Japan and China are buying fewer Treasuries. That's the imbalance we have to handle.
Mohamed El-Erian, September 30, 2026, source
The direction is right. The size is not, at least not on any public tally I could find. Amazon, Alphabet, Meta and Oracle sold about $108 billion of bonds in all of 2025 and about $194 billion in 2026 through early July, and Goldman Sachs expects the five hyperscalers including Microsoft to reach roughly $250 billion this year, according to a July report. A trillion only works if you count loans, private credit and off-balance-sheet data center vehicles, and he did not say that. So treat his number as a guess at total AI financing, not bond issuance. The same report has the part that supports him. Order books on hyperscaler deals fell from nearly five times covered in February to under two times in July. Buyers are getting full.
He wants the fix to come from Bessent, not Warsh
The Fed already acted. It raised rates 25 basis points on September 16 to a 3.75% to 4% range, the first move under Kevin Warsh, and Joe LaVorgna, sitting next to El-Erian, expects another hike in December. El-Erian thinks that is the wrong tool for a supply problem.
The right response to what's going on here is not to increase interest rates to damage the housing market, to damage the car loan market. The right response is fiscal containment, and it also has the advantage of lowering interest rates.
Mohamed El-Erian, September 30, 2026, source
That is a clean argument and a hard one to act on. Fiscal containment means Congress, and nobody on this panel expected Congress to move first.
Gundlach had already written off the buyback
The Treasury's actual response so far has been to buy back long bonds with short issuance, a move Bessent called a Treasury Twist and analysts waved off as a weak form of the real thing. Jeffrey Gundlach sat down with his own DoubleLine channel the day after the Fed hike. The video went up September 30, so his comments are two weeks older than El-Erian's, but nothing since has made them stale.
I don't know if they have or not, but $6 billion, which is what they're talking about on buybacks, is utterly pointless. I just don't see how that's supposed to work. I think I'd have to go much, much higher, but at some point they might.
Jeffrey Gundlach, recorded September 17, 2026, source
In the same conversation he repeated that AI bond spreads have widened by 50 to 60 basis points over the past year while the rest of investment grade sits still, the point we wrote up on September 17. Put that next to El-Erian and you get the same story told from two ends. AI paper is getting more expensive to sell, and the Treasury is selling into the same market.
Intrator says his margins outrun the coupon
If anyone should feel higher rates, it is a company that finances GPUs with debt. Michael Intrator of CoreWeave was asked about it on CNBC from his Fully Connected conference in San Francisco, and his answer was that the question is framed wrong.
What is important to take away from this discussion is that the margins that we are earning are expanding faster than the rates are going up. So the overarching profitability of the infrastructure that we are bringing online is higher than it was a year ago.
Michael Intrator, September 30, 2026, on CNBC
He did not give a margin number, so this is a claim and not a disclosure. He also said there has "absolutely not been any reduction in orders" and that data center moratoriums only move where capacity gets built. The best outside support for him landed a few hours later. Micron reported fiscal fourth quarter revenue of $54.23 billion at an 86.8% gross margin and guided to $61.5 billion for the next quarter. At those margins a few hundred basis points on the debt is noise. That is the uncomfortable part of El-Erian's argument. The borrowers he worries about crowding out the Treasury are the ones least bothered by the price.
Aditya Bhave made the same point on Bloomberg the next morning, calling AI issuance relatively rate-insensitive. On that same show Max Junestrand of Legora, booked to talk about legal software as the bond selloff ran on screen, put it in revenue terms.
We have more revenue this quarter than we did in all of 2025.
Max Junestrand, October 1, 2026, source
Ackman wants the two ends and nothing in between
Bill Ackman was on Bloomberg on September 30 to talk about Howard Hughes bringing in outside capital. Asked whether the bond market is untouchable right now, he gave the shortest answer in the window.
So I'm not really a fixed income investor. I never found the risk-reward of owning a longer-term fixed income instrument interesting. Just not my... I sort of like risk-free or high return. I don't like stuff in the middle.
Bill Ackman, September 30, 2026, source
His insurer, Vantage, will keep most of its assets in short-term Treasuries and put the rest in stocks, the Berkshire playbook. Later in the same interview he said he is less likely to own a fast-growing business consuming a huge amount of capital where you are betting on the lines crossing later. He was talking about frontier labs. It also describes most of the issuers in El-Erian's trillion.
Fraser's borrowers have not felt it yet
Jane Fraser was on Fox Business the morning of September 30, the morning after the 10-year closed at its highest since 2002, mostly to talk about a cyber alliance. The host pushed her on mortgages at 7% and asked about delinquencies.
You know, it's remarkable, no. We've been seeing improvements in the delinquency rates. I have to say though, we do have more of a prime portfolio at Citi, so we do tend to see the better side of that.
Jane Fraser, September 30, 2026, source
Credit her for the caveat. A prime card book is the last place rate pain shows up. El-Erian named the housing market, car loans and the low-end consumer as the people who pay for the wrong policy mix. Citi does not lend much to the third group, so her answer tells you the top of the distribution is fine and nothing about the bottom.
So who pays for the higher coupon?
Not the AI builders, on this week's evidence. Micron's margins and CoreWeave's order book say the companies issuing the debt can absorb a 5% handle and keep going, and Junestrand's quarter says the software on top is still compounding. The Treasury can't absorb it the same way, because Gundlach is right that a $6 billion buyback does not move a market that just passed $40 trillion of federal debt. That leaves the borrower who has to refinance a car or buy a house at 7%, the person none of these executives lends to directly. El-Erian's trillion is overstated. His point about who gets squeezed is the part of his argument that holds.