Yesterday morning's PCE report was the last major data point before Warsh speaks, and it wasn't the one the doves wanted. The Fed's preferred inflation gauge held at a 3.7% annual pace in July — unchanged from June and hotter than the 3.6% economists expected — rising 0.2% on the month against a 0.1% forecast. Core PCE came in as predicted: up 0.2% for the month, 3.3% over the year, exactly where it sat in June.

Read that carefully, because the nuance matters. This isn't a new inflation surge. It's a stall. Headline PCE is still 1.7 points above the Fed's 2% target and core is 1.3 points above it, and neither moved last month. After June's genuinely great numbers, July says the glide path we were celebrating has flattened out — which is precisely the ammunition the three dissenters from the July meeting have been waiting for.

Underneath the headline there's a more interesting story, and it's the one we'd point a client to. Goods prices actually fell 0.1%, with energy goods down 2.7% and furnishings off 0.9%. The pressure was entirely in services, up 0.3% — led by a 1.2% jump in financial services and insurance and a 0.3% gain in housing. And the Fed's own regional measures of underlying inflation look considerably calmer than the headline: the Cleveland Fed's median PCE ran 2.7% annually and the Dallas Fed's trimmed-mean measure has been tracking around 2.2–2.3%. Strip out the noisiest categories and the underlying trend is much closer to target than 3.7% suggests.

The consumer data deserves its own beat: real spending was flat after rising 0.4% in June, while after-tax income grew 0.4% — the strongest since January — and the savings rate ticked back up to 3.0% from a four-year low of 2.6%. Households earned more and chose to bank it. That's a cautious consumer, and cautious consumers slow economies. Combine it with July's −23,000 payrolls and you get the picture the doves will bring to the September meeting: demand is cooling even as the price data refuses to.

If you only read one section this week, make it this one. The 30-year Treasury closed at 5.31% on August 17 — its highest level since 2007 — and was still around 5.27% last week, roughly a 19-year high. The 10-year has climbed back to about 4.73%, matching its highest in more than a year. These are the yields your mortgage rate is built on, and they are going the wrong way while inflation data goes sideways and the job market weakens. That combination should tell you something important: this isn't really about the Fed anymore.

Here's what happened next, and it's remarkable. On August 19 — the same day the national debt formally crossed $40 trillion — Treasury Secretary Bessent announced without warning that the government would more than double its buybacks of long-term debt, raising the maximum per operation from $2 billion to at least $4 billion, aimed squarely at the 10-to-30-year part of the curve where buyers had gone on strike. The 30-year yield dropped 9 basis points to 5.19% on the news. By the next day it had given all of it back. Evercore ISI's Krishna Guha called the intervention "a weak form of Operation Twist." The bond market looked at the government's attempt to lower long-term borrowing costs and shrugged.

Also yesterday, while everyone watched PCE, Nvidia reported a quarter that reframes what this economy actually is: $96.2 billion in revenue, up 106% from a year ago, with data center revenue of $89.0 billion, up 117%. Jensen Huang's framing was blunt: "AI has reached its inflection point. It's doing useful work... Now, compute is revenue. And demand is accelerating."

That's not just a tech story, and here's why it belongs in a mortgage newsletter. Independent analysis from Epoch AI estimates that AI-related data center construction, compute hardware, and networking accounted for roughly 0.8% of U.S. GDP in the first quarter of 2026 — pushing computing infrastructure as a whole to about 1.5% of GDP, up from a 2015–2022 average near 0.7%. AI infrastructure is now the leading driver of growth in private investment in the United States. Warsh saw this coming: in his July testimony to Congress he called the AI buildout "the most striking feature" of the economy and predicted that "what is now called 'AI investment' will soon be called just 'investment'."

Now connect it to your rate, because the link is real and almost nobody explains it. An economy building hundreds of billions of dollars of capital infrastructure is an economy that doesn't need rate cuts — and one that competes for the same pool of savings the government is borrowing from. Massive private capital demand plus $40 trillion in public debt equals upward pressure on long-term yields. That's textbook. It's also, we'd argue, part of why the long end has decoupled from the weakening labor data: capital investment is holding growth up even as hiring falls. Two other threads worth watching: the buildout is electricity-hungry, and July's PCE showed housing and utilities among the services categories still pushing prices higher; and it's concentrated, so if capex plans get trimmed, the growth support underneath these yields thins out quickly.

Here's the paradox worth sitting with: Treasury yields are at multi-year highs, and mortgage rates have quietly fallen three weeks running. Freddie Mac's survey went 6.69 → 6.67 → 6.65, with this week's reading due today, and the 15-year has slipped under 5.85% on some boards. That's lenders compressing their own margins, not the bond market handing anything over — which is exactly the "spread" story we walked through last week. It also means the gap between lenders right now is bigger than the gap between weeks. Freddie Mac's own research makes the point: in high-rate markets, shopping multiple lenders can save roughly $600 to $1,200 a year. That's real money available today, with no help from Wyoming.

Warsh delivers his first Jackson Hole keynote as Chair tomorrow morning at 10:00 ET, at a symposium running through Saturday and hosting roughly 120 officials and economists from more than 70 countries. He has abolished forward guidance, so a set-piece speech from this chair carries unusual weight — the tone is the guidance now. Here's our scorecard:

1. Does he mention the long end at all? This is the one that matters most for your rate. If Warsh talks about the government's cost of borrowing or what thirty-year lenders are demanding, it signals the selloff has his attention. If he stays entirely on short-term policy, he's treating the long end as the Treasury's problem, not the Fed's — and the buyers' strike likely continues.

2. How hard does he lean on the stall? He told Congress that five years of above-target inflation "cannot be cured in nine weeks — or by a single month of modest price decreases." Yesterday's PCE handed him the evidence. If he repeats that framing with force, September's hike odds — currently stuck near one-in-three — will climb, and mortgage rates with them.

3. Does he connect AI investment to policy? He's already called it the most striking feature of this economy. If he argues the buildout is genuinely raising productivity, that's a subtly dovish case — more supply means less inflation for a given amount of demand. If he frames it as a source of overheating and capital competition, that's hawkish. Almost nobody will be listening for this. It may be the most consequential thing he says.

Date	Event	Why It Matters
Tomorrow Fri Aug 28 · 10am ET	Warsh's first Jackson Hole keynote	With guidance abolished, the tone is the guidance · Flash edition this weekend
Fri Sep 4	August Jobs Report · BLS	Does July's −23K become a trend?
Thu Sep 10	August Existing-Home Sales · NAR	Does the fall market start with inventory or without it?
Fri Sep 11	August CPI · BLS-verified	Last inflation print before the decision · Flash edition
Sep 15–16	FOMC + fresh dot plot	First projections since June, when 9 of 18 penciled a hike by year-end

Closing in the next 45 days: lock today if you haven't. A first keynote from a chair who refuses to telegraph anything, delivered into a bond market at 19-year highs on the long end, is precisely the setup where a single paragraph moves your rate sheet before Monday. Pair it with a float-down and let tomorrow happen without you holding your breath.

Refinancing: the 15-year is where the value is — some boards have it under 5.85%, roughly 80 basis points below the 30-year. If you're eight or ten years into a loan and can carry the payment, that's a shorter, cheaper path than waiting for a 30-year rescue that this bond market isn't offering. Run it both ways before you decide.

Everyone: the highest-return move available to you this month still isn't a market call — it's collecting three or four Loan Estimates on the same day and comparing them line by line. With lenders compressing margins to compete while yields sit at highs, the spread between good and mediocre pricing is unusually wide. That's a gap you can close yourself, this week.

Three storylines converge on a podium in Wyoming tomorrow morning. Inflation stopped improving. The long end of the bond market is at levels it hasn't seen since 2007, and shrugged off the Treasury's attempt to fix it. And an AI buildout worth roughly 0.8% of GDP is holding growth up while hiring falls — breaking the old link between a weak labor market and falling mortgage rates.

Warsh has to address all three without the forward guidance his predecessors leaned on. That makes tomorrow genuinely unpredictable and genuinely important — the market has priced neutral, which means surprise cuts both ways. Meanwhile, mortgage rates have fallen three weeks running on lender competition alone, which is a quiet reminder that the most controllable variable in your rate has never been the Fed. It's been the shopping. We'll be back this weekend with what he actually said and what it does to your rate sheet.

Thanks for reading — and if you're closing soon, let's talk before that microphone turns on tomorrow. That's what I'm here for. — Ryan