Macquarie's Wizman Offers an AI-Financing Reading of the Treasury Buyback

When the US Treasury began buying back long-dated government bonds, the market's first question was where the money would go. Thierry Wizman, a foreign exchange and rates strategist at Macquarie, offered the reading that the space it cleared would be filled by corporate debt financing AI infrastructure and data center capital expenditure. This is the first time a Treasury buyback has sat in the same sentence as AI capital spending.

The Treasury issued its announcement on August 19. The announcement METAL read says the department is increasing, by at least double, the size of liquidity support buyback operations for longer-dated nominal coupon securities in the 10-year to 20-year and 20-year to 30-year sectors. The maximum of $2 billion per operation rose to at least $4 billion, effective September 9 and running through November 4, when this refunding quarter ends. The Treasury cited consistent strong sponsorship from market participants in the longer-dated sectors and the significant volume of high-quality offers it routinely receives.

Several numbers overlap in the background. Thirty-year Treasury yields approached a near-20-year high last month, the US national debt just passed $40 trillion, and Treasury interest payments are expected to exceed $2 trillion in fiscal year 2026. Long-dated yields serve as the benchmark for borrowing costs across the economy, from mortgages to business loans, so pulling yields down also lowers the government's own cost of funding. Weeks before the buyback announcement, Treasury Secretary Scott Bessent announced an intervention to buy Japanese yen, and Japan holds more US government debt than any other country.

The criticism came from close by. The investor Stan Druckenmiller, Bessent's friend and mentor, wrote in an opinion piece that a credible fiscal package would have had more impact on yields than artificially suppressing them through price management. That is where the suspicion spread that the Treasury was trying to manage prices in the bond market. The Treasury itself never said the buyback was a price-setting exercise, yet the label of failure still attached to it.

Christina Parajon Skinner, a professor at Wharton, thinks that reading goes too far. Having served at the Treasury under Bessent from July 2025 until August this year, she called the move market plumbing. She said, "From the outside looking in, this very clearly does look like liquidity management, a market functioning exercise, which I don't at all perceive to have been anything remotely close to a failure." Regular Treasury repurchasing operations were already introduced in May 2024, she points out, and what changed is only the size per operation. She also said, "The Treasury has never been a passive buyer of government debt."

Thierry Wizman, Macquarie's global foreign exchange and rates strategist, read a different grain in the same material. He noted that Bessent spoke only about liquidity and did not raise the deficit or a yield target. He then said, "If there's a pressing need to allow AI infrastructure to get built out and financed, you certainly wouldn't want all of that government debt issuance to crowd out the corporate issuance, and therefore we need to make space." When government yields fall, so does the cost of the corporate debt competing with them, which makes money for AI capital spending cheaper.

Wizman placed that intention not with the Treasury secretary but above him. While saying it is not the Treasury secretary's job to promote one sector over another, he said, "It's implicit by what the president is saying that they want to run the economy hot for AI, and then it's the job of the Treasury to execute on that broader intention." Industrial policy is set by the president and executed by the Treasury.

Demand on the AI side is certainly large enough to need space made for it. In an August 7 report, Goldman Sachs Research projected that global AI-related investment will exceed $1 trillion in 2026, with $581 billion of it occurring in the United States. The commonly cited hyperscaler capital expenditure consensus of $794 billion understates global AI capex by roughly $200 billion while overstating US investment by roughly $200 billion, the team found. Cumulative investment since 2022 will reach $1.8 trillion by the end of this year, and AI capex as a share of US gross domestic product rises from 1.8 percent in 2026 to 2.5 percent in 2027 and 2.8 percent in 2028.

Joseph Briggs, who co-leads the global economics team, wrote in the report, "The AI capex growth outlook, including how high AI investment ultimately rises as a share of GDP and when capex growth slows, is a key source of uncertainty for macro markets right now." He added that all the leading indicators rank near the top end of their range since 2022, suggesting robust near-term growth. Wizman urged holding off on judgment: whether everyone got financed will only be clear at year-end, and whether AI lifts productivity enough to grow the economy and bring inflation down will take until next year.

What the move actually taught the market is something else. Yiming Ma, a professor at Columbia Business School, thinks the market has now learned at what point the Treasury moves. She said, "This kind of communication sometimes works in the short term because it signals commitment to the market that a big buyer is going to step in in times of need," but added, "sometimes it also can backfire because the market will look at this and say, 'Well, the fact that you need to come out and say and do these things implies that this market has already lost the confidence of investors.'" She also said the intervention has something of a Pandora's box about it, because once the market expects the Treasury to step in and it does not, confidence falls sharply.

Rising long-dated yields are not only an American matter. Ten-year yields have been tracking higher in the United Kingdom, Japan and France as well, and the fundamentals of government borrowing and inflation expectations are likely to keep them high. So the precedent this buyback set is not about size but about a reaction function. The practical result of this round is that the market now knows under what conditions the government is willing to intervene in the bond market.

AI capital spending has only recently come down into the bond market. METAL has reported that SoftBank is issuing junk bonds to fund its investment in OpenAI, and has also covered Alphabet raising $3.6 billion in its first Australian bond sale. A trillion dollars of capital spending has to be borrowed somewhere, and the room to borrow it widens by exactly as much as government debt steps aside. Whether Wizman's reading holds will be answered not by the Treasury but by the corporate issuance ledger at the end of this year.