September 16: Warsh Will Decide Who Pays for High Rates

The main question in markets right now is whether the Fed will raise rates. Most people expect it will not. Warsh was nominated by Trump, Trump has repeatedly said America pays too much interest, and the midterm elections are on November 3. In the BofA fund manager survey, 72% said the Fed would not move before the election.

After the September 4 payrolls report, the market put September hike odds at 58% to 62%. Before Jackson Hole, that number was 35.4%.

A thirty-minute speech doubled the probability, and most retail positioning has not adjusted.

That view leaves out two developments that have been in place all year. The Iran war has raised the cost of energy and transportation. The AI infrastructure build-out is now competing for the same pool of long-term capital that the Treasury needs to fund itself at auction. Neither one appears on the economic calendar, so investors who follow payrolls, CPI and Fed speeches tend to miss them.

The September meeting will not reduce how much AI projects need to borrow, repair a refinery, or shrink the deficit. What it determines is narrower:

whether the Fed is willing to let equity valuations and the Treasury's interest costs absorb the impact in order to keep the market's confidence in the dollar and in 2% inflation.

Start With a Chart Anyone Can Pull Up

Search “10-year TIPS” and you’ll get the CNBC chart of the real yield on inflation-protected Treasuries, currently around 2.43%.

Here is what that means. If you lend to the US government for ten years through TIPS, your principal adjusts with the price level, so inflation does not erode it. Even with that protection, investors currently require an additional 2.43% per year. That is the cost of lending to the United States after inflation is removed. I will refer to it as the real yield.

There is no published chart for expected inflation. You calculate it:

Take the ordinary 10-year Treasury yield and subtract the TIPS real yield. The difference is the average inflation rate the market expects over the next ten years.

Using the Fed’s own numbers as of September 1:

Real yields rose 57 basis points. Nominal yields rose 51. The spread between them actually got smaller.

Which means the real story is this:

Long-term Treasuries are being sold and yields are rising because the real cost of capital has increased, not because the market expects higher inflation.

Core PCE is running at 3.3%, while the market prices only 2.35% average inflation over the next decade. The bond market does not expect the United States to inflate away its debt.

It expects the price of money (capital) to remain expensive. The question is why?

AI Is Outbidding the Treasury for Its Own Buyers

Debt issued this year to finance AI infrastructure is running at roughly 1% of global GDP, slightly more than one trillion dollars a year.

The typical buyers of 30-year Treasuries are pension funds, insurers and sovereign wealth funds. They have obligations to retirees and policyholders decades ahead, so they need assets that are long-dated, highly rated, and yield somewhat more than government bonds. They do not normally buy technology company debt.

But AI infrastructure debt has been structured to meet their requirements.

Meta’s Hyperion data center is financed through a separate project company at approximately 27 billion dollars. The key term is Meta’s 16-year guarantee: if the lease is not renewed or the project ends early, Meta pays the project under the agreed terms. Rating agencies relied on that guarantee to assign an investment-grade rating, and PIMCO purchased most of the paper.

For a pension fund, this debt has a maturity close to a long Treasury, a comparable credit rating, and a higher yield. Investors who would previously have absorbed the final portion of a Treasury auction now have a better-paying alternative. To keep that money in the government bond market, the Treasury has to offer a higher yield.

Bessent has acknowledged this. On August 20 he said the Treasury is competing with a large volume of corporate debt issued at higher yields, including AI infrastructure debt. At the G20 he said AI investment is absorbing savings that previously flowed into Treasuries and helped keep US borrowing costs low.

I’ve written a separate piece on AI issuance, check it out:

And the Old Buyers Are Walking Away

While supply is increasing, the buyers who held down long-term yields for the past decade are stepping back.

The Fed has ended quantitative easing and paused reserve management purchases on August 14. The Bank of Japan has ended its policy of suppressing domestic yields, and Japanese institutions are moving capital home. Japanese holdings of Treasuries fell from 1.239 trillion dollars in February to 1.116 trillion in June, while the 30-year JGB yield reached 4.14%, near the 4.20% peak in May. Central banks are also changing reserve composition, buying a net 288.9 tonnes of gold in the second quarter, up 62% year over year.

Supply continues to grow. Federal debt passed 40 trillion dollars in August, the deficit is close to 6% of GDP, and net interest costs will exceed 1.1 trillion dollars this fiscal year, more than the defense budget. Germany issued 30-year bonds on August 18 at the highest yield since 2011, with the 30-year reaching 3.85% and the 10-year at 3.36%, a 15-year high. Japan’s 10-year reached 2.93%, the highest since 1996. The US 30-year hit a 19-year high on August 18.

A decade ago there was more savings than there were investment opportunities, which held rates down. Today governments, data centers, energy security and defense are all seeking capital at the same time.

The War Ended the Case for Waiting

If AI capex were the only factor, the Fed could wait. The build-out increases demand for data centers, power, chips, land and labor, and it may eventually raise productivity and lower unit costs. Facing a technology investment boom in the 1990s, Greenspan chose to assume productivity would improve.

Warsh doesn’t have that option in 2026.

The Iran war started in late February. Hormuz closed on March 4, Brent briefly cleared 120 dollars, and by March 12 output across four Gulf producers was down at least 10 million barrels a day. March CPI rose 0.9% on the month, with gasoline up 21.2% in one print, close to three quarters of the monthly increase on its own. By July, headline CPI was 3.4% year over year with energy up 14.7%.

The conflict has become a stalemate that could extend into 2027, and Iranian oil exports fell roughly 85% in August. The effect on refining capacity is the more persistent problem. A shut-in well can be restarted relatively quickly. A damaged refinery takes much longer to repair. Reuters expects diesel and gasoline prices to stay elevated for several years even if an agreement is eventually reached.

AI is a possible improvement in supply in the future. The war is an increase in costs that has already occurred. AI has raised the cost of capital, and the war has removed the Fed’s ability to wait on the argument that AI will eventually be deflationary.

In an ordinary cycle, an investment boom that raises inflation from 2% to 2.5% is is tolerable for a central bank. But today, the war has already used up that margin.

Hikes Don’t Reach the Long End

After the September 4 payrolls print, the 2-year yield rose 7.6 basis points. The 30-year rose 1.

1994 was the same shape, scaled up.

The Fed raised rates seven times in twelve months that year, taking the funds rate from 3% to 6%. By mid-May the 2-year yield had risen about 280 basis points while the 30-year had risen a little over 110. The 30-year went from 6.17% on January 12 to 8.16% on November 4, a 199 basis point increase for the year, and the New York Fed's long Treasury index lost approximately 7.5%.

The front end follows the Fed. The long end has its own agenda. The curve flattened substantially and never inverted.

A rate hike changes the expected policy rate over the next few years. What is driving long-term yields in this cycle is the real cost of lending to the United States for a decade, and the funds rate does not affect it.

Treasury intervention has not worked, and the UK and Japan have had the same experience. On August 19 the Treasury raised its long-bond buyback cap from 2 billion dollars per operation to at least 4 billion, with a window from September 9 to November 4. The 30-year yield fell nearly 10 basis points within minutes, reversed the entire move the next day, and by September 1 was at 5.27%, six basis points from its level before the announcement.

Four billion dollars is about 0.33% of average daily Treasury volume.

The UK and Japan have both shortened issuance maturities and reduced long-bond supply, and 30-year yields in both countries remain at multi-decade highs.

Druckenmiller made this argument in the Wall Street Journal on August 24: the Treasury is not addressing a market malfunction, it is supporting a price level. A genuinely impaired Treasury market produces specific signs, including failed auctions, forced dealer liquidation, and a sudden widening of bid-ask spreads. None of those occurred. The Treasury’s own announcement described the larger buybacks as liquidity support for securities where market participants have shown consistently strong demand. If demand is strong, that segment was functioning normally.

If long-term Treasuries need to yield 5.5% to attract buyers, that is the return investors currently require, not evidence of market failure.

What Warsh Can Actually Control

Real yields are beyond the Fed’s reach. Inflation expectations, currently 2.35%, are not.

For the sake of the argument, if ten-year inflation expectations rise from 2.35% to 3.5% while the real yield stays at 2.44%, the ordinary 10-year Treasury yield goes from 4.79% to nearly 6%.

The cost of that outcome would be broad. Mortgage rates, corporate borrowing costs, commercial real estate valuations and equity multiples would all reprice against a 6% risk-free rate rather than 4.79%. That repricing would apply across the entire economy and would persist as long as expectations stayed elevated.

Warsh is not raising rates to lower long-term yields. He is raising rates so the market continues to expect the Fed to act before inflation becomes entrenched.

There is a cost to hiking. Treasury bills are 22.1% of marketable debt, about 7 trillion dollars, so each 25 basis point increase adds roughly 17.4 billion dollars per year to Treasury interest expense. Highly valued equities also face multiple compression from a higher discount rate.

There is also a cost to holding. If the market concludes the Fed will accommodate the Treasury’s funding needs, the adjustment appears in the dollar and in inflation expectations, long-term yields rise further, and equities reprice regardless.

The question is not whether there is a cost. It is who bears it.

Warsh Set a Standard, Not a Decision

At Jackson Hole, Warsh said responsibility for 65 months of elevated inflation sits with the central bank. He restated that the 2% PCE target is firm and fixed, said price stability is not self-executing and inflation is not necessarily mean-reverting, and said short-term interest rates are the predominant tool while unconventional policy should be used sparingly, if at all. He also said he would be hard-pressed to describe broad financial conditions as restrictive, citing capex, corporate profits, consumption, credit spreads and asset prices.

He also said the opposite of a commitment. He restated the July FOMC view that a good majority preferred to wait for new information before deciding whether a policy change was advisable, particularly given developments in supply chains, investment flows and geopolitics. He said the Fed's AI and productivity task force will report later and has no bearing on current policy decisions. And he closed by saying he was committed to a discipline, not a decision

What he did make public is his test:

“We must be confident that underlying inflation is moving to our objective clearly and at sufficient speed. Otherwise, we have work to do.”

That standard has two conditions. Inflation has to be moving toward 2% clearly, and it has to be moving at sufficient speed. He also said the summer’s better PCE and CPI readings did not indicate that underlying trends had meaningfully improved, and that 54% of the 199 PCE components rose more than 3% over the past twelve months, against about 32% in the two decades before the pandemic.

So Warsh did not signal a September hike. He set conditions that are currently difficult to satisfy.

The committee is a separate question. On September 3, Governor Waller said that if the data continue to show inflation improving, he would prefer to hold in September. Williams at the New York Fed also did not support an immediate increase. Two published vote reconstructions are similar: about six leaning toward holding, five leaning toward hiking, and Powell unclear.

September 4 changed part of that. Unemployment held at 4.1% while participation rose from 61.4% to 61.6%, meaning more people entered the labor force and employers absorbed them. Labor market weakness is no longer a strong argument for waiting.

Powell is likely the deciding vote.

Tech and Gold Are Trading Different Variables

Equity markets have begun distinguishing between companies that fund capex from cash flow and companies that fund it through credit markets.

Nvidia reported second-quarter revenue of 96.2 billion dollars, up 106% year over year, with data center revenue of 89 billion, up 117%, and guided to 108 billion for the following quarter. The stock rose 8.7% the next day, adding 442 billion dollars in market value.

The broader semiconductor sector did not break out, and Nvidia remains below its May high. On September 1, with the 10-year at 4.79%, the Nasdaq declined twice as much as the S&P.

Companies with stable cash flow, low net debt and identifiable AI revenue face a valuation issue. Companies that depend on debt issuance, project companies, vendor financing or assumptions about GPU residual values face a funding issue, and rate increases affect that directly.

Gold is currently trading on the real yield, the same series shown on the TIPS chart.

Gold rose about 10% in August, its strongest month since January 2026, with spot reaching 4,651 dollars intraday on August 24 and futures briefly above 4,700. On August 28, following Warsh’s speech, it fell 3.16% to 4,455. By September 2 it had fallen below 4,300 to a four-week low. On September 3, after Waller’s comments, it closed up 1.93% at 4,472.67. On September 4, after the payrolls report, it declined again.

Throughout that period, ten-year inflation expectations remained near 2.35%. Gold also remains roughly 20% below its January high.

An asset still recovering from a 20% drawdown moved 7% lower and 3% higher within two weeks in response to two Fed officials’ comments, while inflation expectations did not change.

The variable driving gold’s price has shifted from inflation expectations to real yields. One group of buyers is not responding to the Fed at all. The World Gold Council surveyed 76 institutions in 2026: 89% of reserve managers expect global central bank gold holdings to increase over the next twelve months, a record 45% intend to add to their own holdings, and 1% intend to reduce. These institutions allocate over multi-year horizons, are not price-sensitive on entry, and do not use leverage.

A hawkish Fed therefore produces price declines in gold without changing the reserve diversification trend.

What to Watch After September 11

August CPI is released on September 11. The Fed decides on September 16.

• Core CPI at 0.2% month over month or lower would likely reduce hike odds to around 50%.
• 0.3% would likely raise them to around 80%.
• 0.4% or higher would likely push them above 90%.

I put September hike odds at 70%, above the market’s 58% to 62%. I put the probability that rates are higher than current levels by December at about 91%, against roughly 88% priced.

After the meeting, the first item to check is whether “we have work to do” appears in the statement. If it does, the market will begin pricing further increases. If it does not, the market will treat the move as a single precautionary adjustment.

Winter is the larger risk window. If energy prices continue rising, wage agreements could change in early 2027, transmitting energy costs into wages and services prices.

Conclusion

The United States does not have a problem that rate cuts would solve.

AI capex and fiscal expansion have raised real yields. The war has removed the room to tolerate additional inflation. Treasury buybacks cannot set the price of long-duration debt. And rate increases cannot return the real cost of capital to the level of the past decade.

What Warsh can defend is the market’s expectation of inflation, which is why he restated that the 2% PCE target has not changed.

Real yields are up 57 basis points over the past year. Inflation expectations are down 6. The first is outside the Fed’s control. The second is not.

The September 16 vote determines whether the Fed remains willing to defend that expectation. The main risk for retail investors is allowing one variable, that Warsh was nominated by Trump, to override the rest of the evidence.