Will Slaughter
Fixed-Income Portfolio Manager at Protective Life
Long-dated Treasuries look attractive because short-rate markets are wrong to expect the Warsh Fed to hike 50-100 bps over two years. Most of this year's rate action has been in the 2-year versus fed funds spread, which moved from -20 bps to +60 bps as pricing shifted from cuts to hikes. Real 30-year yields are at multi-decade highs, and housing, autos and consumer sectors show rates already choke demand. When the AI capex boom fades, a recession and a sharp rate repricing lower should follow. x.com x.com x.com x.com x.com x.comÂ
Potential spending cuts and AI-related deflation make long-dated US Treasuries an asymmetric opportunity. A Democratic sweep could cut spending; an AI-trade blowup could cause deflation, while the AI trade hurts employment. Ending the Iran war could reduce spending and oil prices. Further crude gains could damage the economy hard and quickly enough to outweigh the inflation threat. x.com x.comÂ
Long-dated Treasuries could benefit from less issuance of AI-related debt, which today's news may point to, even if the deal itself is minor. Separately, a free Meta Muse hurts the leading AI models at least at the fringe, and wide consumer adoption of AI agents will come when Apple launches its own or partners with a model maker. x.comÂ
Long-dated US Treasuries look unattractive when AI credit pays twice as much and the market assumes the Fed will bail it out if it breaks. Borrowers who don't care what money costs are setting the price for everyone else, so homebuyers pay through higher mortgage rates and small businesses through layoffs. JPMorgan is marketing a $5B Volta AI loan at roughly 11%, per a news post. The Fed has repeatedly protected big banks and is unlikely to let capitalism play out when banks face pain. x.com … Show more
Expensive borrowing makes debt-funded AI projects harder to justify, pressuring AI stocks and the S&P 500. Inflation keeps pressure on rates. If Republicans lose Congress after the midterms, Trump could rely on executive action to speed permits, power projects and existing funding programs, keeping AI investment moving despite expensive money. Tech and risk assets could catch a bid if earnings hold and bond yields settle; Fed cuts remain uncertain. x.comÂ
The S&P 500's concentrated, correlated AI exposure makes a large allocation too risky as rates rise. Higher rates could force data-center bondholders to sell AI equities; early AI use appears to be stagnating and unclear ROI may cap demand even if usage grows linearly. Failed lab IPOs or a credit-driven recession could also undermine the market. AI moving on-device or on-premise is another significant risk to data-center capacity demand. x.com x.comÂ
Inflation has broken long-dated US Treasuries' safe-haven role. Long-term Treasuries have returned -4.1% annually in the 2020s to date, versus +4.9% in the 1930s. Long bonds look unlikely to do well for at least the rest of this decade. x.com x.com x.com x.comÂ
Long-dated US Treasuries look like a good hedge against rising rates, which at some point could crash the stock market. Duration on the 20-year Treasury has come way down, so another 100 basis points higher in yields would mean about an 11% price decline, or roughly a -5% total return. That is small compared with the decline equities could suffer. x.comÂ
Debt-funded AI building is making tech stocks vulnerable to a 25-30% pullback as rates and energy prices rise. AI company borrowing reached $466B this year versus $101B a year earlier; weekly default-insurance trading on seven AI names rose from under $2B in July to over $9B in September. Keep cash for lower prices, though 30% earnings growth and a 19x market multiple argue this is not an AI bubble. x.comÂ
Long-dated US Treasuries could lose buyers to the AI buildout, pushing yields and government funding costs higher. Microsoft, Google and Amazon offer debt at a spread over Treasuries while generating over $100 billion a year in cash flow and 30%+ returns on capital, a better deal than lending to the government at 4.5%. Beyond about $725 billion of hyperscaler capex, debt for fabs, memory, power and the grid is about $3-5 trillion through 2030, comparable to Treasury issuance over that period. … Show more
Credit tightening and data-center delays threaten the debt-funded AI boom behind the S&P 500. Junk-bond sales fell from $51B in September to $7B in October's first three weeks; CCC yields hit a four-year high, though quality firms still borrowed. Per S&P Global, 146 data-center projects were delayed or canceled, potentially weakening buildout credit. With 40% of the index in 10 names, this is dot-com and 2008 combined; stocks have not caught up, and even good firms may be sold. x.comÂ