The Fed raised rates today, unanimously.
Not because Warsh described a weak economy.
He said inflation is “too high and has been for too long.” He also said economic activity is expanding at a solid pace.
That is the important split today.
The Fed tightened because inflation is still too high. At the same time, AI spending and AI valuations kept moving higher.
Those two stories did not settle anything today. They just sat next to each other.
The Fed Chose Inflation First
The FOMC raised the target range for the federal funds rate by a quarter of a percentage point to 3 and 3/4 to 4%.
Warsh said the vote was unanimous.
His reason was direct:
“inflation is too high and has been for too long.”
The inflation numbers were still not where the Fed wanted them.
The 12-month change in total PCE prices likely was around 3.6% in August.
Core PCE and CPI prices were running at about 3.2 and 2.4% respectively.
Too many categories were still rising above 3% on both the 6- and 12-month basis.
But Warsh was not describing an economy in trouble.
He said economic activity is expanding at a solid pace. Domestic spending has been resilient. Productivity growth has been strong. Capital investment has been robust.
He also said job gains have kept pace with the workforce, unemployment has changed little, and he does not believe the Fed needs to harm labor markets to achieve its objective.
Markets were already expecting the move. Kobeissi Letter said markets saw a 93% chance the Fed would announce its first rate hike since July 2023. It also said that, in data going back to 2008, the Fed has always raised rates when expectations of a hike were this high.
Political pressure was still there.
Trump said:
“What I’m saying, very simply, is that we should be paying the lowest interest rate in the world.”
He also posted:
“LOWER THE RATE OR I’LL STOP TRADING WITH COUNTRIES WITH WHICH WE HAVE A DEFICIT.”
Warsh’s answer was short:
“I’m not in the forward guidance business.”
And:
“Independence of the Federal Reserve is about staying in our lane.”
Long Yields Hit Their Highest Since Late 2023
The bond market did not relax.
The 10-year note rose 0.05 percentage points to just below 4.86 per cent, its highest level since late 2023.
That move came after Treasury planned to purchase $6bn in government debt in the first buyback operation since Scott Bessent unveiled an expanded scheme last month.
The problem was size.
The $6bn plan came in below the $8bn and $10bn estimates circulated by Wall Street analysts this week.
Mike O’Rourke at Jones Trading called the number a disappointment because investors expected Bessent to do more. Krishna Guha at Evercore ISI said markets appeared “underwhelmed.”
The reason this matters is simple.
Higher long-dated bond yields raise borrowing costs for US consumers, notably mortgages, and for companies seeking to raise cash.
Treasury will make the final announcement about eligible bonds on Thursday morning, conduct the operation from 1.40pm to 2pm Eastern time, and publish results shortly after 2pm. It will also hold a further auction for 30-year bonds on Thursday.
Meta Is Building Chips and Selling Agents
Meta had two AI stories today.
The first was custom silicon.
Bloomberg reported that Meta plans to begin deploying MTIA 450, code-named Arke, across data centers in the first half of 2027.
The next-generation MTIA 500, code-named Astrid, is expected to finish design work in roughly a month and enter data centers by the end of 2027.
Meta has committed to more than 1 gigawatt of capacity using its custom chips over a 12-month period.
It received its first 12 Arke chips from TSMC on September 1. Early testing showed performance within 2% to 3% of simulations.
Meta also canceled Olympus after determining that a training-plus-inference design could cost roughly 30% more.
That is not just a chip story. It is a cost story.
Meta is working with Broadcom on chip design and TSMC on manufacturing. The chips are mainly general-purpose inference “workhorses,” not the fastest latency-sensitive chips.
Meta says close coordination between AI teams and chip designers can deliver better performance per watt and per dollar for its own workloads than current NVIDIA systems.
The second story was Muse.
Meta introduced Muse on Tuesday. It lets people offload digital tasks like booking appointments, filling electronic forms, and monitoring home security camera feeds to AI-powered assistants.
Muse has a free tier and monthly plans of $20 or $100, depending on usage.
Meta said Muse runs in “its own isolated environment” and “never sees your actual passwords or payment details and asks before doing anything sensitive.”
Users must opt out if they do not want Meta using Muse interactions to train AI models.
Meta is under pressure from Wall Street to show returns on its AI investments and reduce reliance on digital ads.
Big AI Numbers, Contested Prices
Broadcom was the center of the AI valuation debate.
Leif said Broadcom is trading at its lowest valuation since December 2023.
He cited AI semiconductor revenue guidance of approximately $58B in FY26, up 186% year over year, approximately $115B in FY27, up 98%, and approximately $230B in FY28, up 100%.
Another author said Broadcom had Q3 AI semiconductor revenue of $16.7B, up 221% year over year and 54% sequentially.
Q4 guidance was $21.7B.
Quarterly free cash flow was $13.7B, equal to 46% of revenue.
That author bought Broadcom around $390, kept adding as it fell, and has an investment horizon of at least two years.
The same author modeled roughly $30 in adjusted EPS for FY2028. At a $390 purchase price, they said they paid about 13 times that estimate. They said a 22× multiple would put the shares around $660 in FY2028.
The downside cases were also spelled out.
The delay case was roughly $23.5 in EPS and an 18× multiple, for value around $424.
The more severe case was around $17.2 in EPS at 15×, or roughly $258.
The risks were clear too. The author said growing memory content in custom accelerators dilutes gross margins. They also said Broadcom’s involvement in customer financing can include residual-value guarantees that need scrutiny.
A separate valuation piece said $AVGO was at $339 and “appears to be a strong consideration for investment.”
It also called the stock “extremely volatile and not for the faint of heart,” with two drawdowns of about -30% and three of about -40% over the past decade.
Then there was OpenAI.
unusual_whales reported on X, citing The Wall Street Journal, that OpenAI held early investor discussions about a funding round that could value the company at more than $1.2 trillion.
The risk-watch claim pushed back on what that number means.
It said a $1.2 trillion private valuation is not a market price because nobody could actually sell at that valuation. It also said OpenAI still loses billions annually.
So the issue is whether that number is backed by earnings, or by belief that a later buyer will pay more.
Oracle was in the same AI infrastructure conversation. One X post said Oracle is down 60% from its highs. The same writer said Oracle’s backlog reached $664B, with approximately 50% expected to convert into revenue within 3 years.
The Fed tightened into an economy Warsh described as solid. Long-dated yields hit their highest level since late 2023, and higher long-dated yields raise borrowing costs for consumers and companies seeking to raise cash.
At the same time, Meta committed to more than 1 gigawatt of custom-chip capacity, Broadcom’s cited AI semiconductor revenue guidance roughly doubles for three straight years, and OpenAI held early investor discussions above a $1.2 trillion valuation.
Those stories do not resolve each other today. They just sit there, side by side.