Authored by Ed Dowd via 'Beyond the Narrative' substack,
On August 19th the Treasury announced it would increase the size of its nominal long-end liquidity support buybacks beginning September 9. The long end yields declined on the headline. Cue the usual chorus of X hot takes: stealth QE, yield-curve control lite, money printing etc.
The reality is much less bombastic...it is mostly a jawboning exercise.
The Treasury Is Not the Fed...The Treasury Cannot Create MoneyBuybacks of this type are a recycling operation. You issue more bills and notes on the front end and you take some longer paper off the street. You can tidy up liquidity in off-the-run issues. You can send a signal that you would prefer the 30-year not print a fresh multi-decade highs on a Tuesday. What you cannot do is print reserves, expand the monetary base, or run a proper balance sheet policy the way the Federal Reserve can. Confusing the two is how people talk themselves into thinking a few billion of "liquidity support" is 2020 all over again.
The size tells the story. Coupon supply at the long end is still large. Doubling a buyback program that was already small relative to annual issuance is, at best, a band-aid. Markets gave it a day. Then they remembered the calendar. The signal from Bessent is not nothing but it is not as big as it seems in the broader picture.
Who is actually in charge of the long end of the yield curve? It is not the Fed. It is not Scott Bessent's operations desk. It is priced by growth expectations and inflation expectations or said differently the boom/bust cycle.
Who is in charge of that? The laws of nature and God.
Bessent Will Get Lower Yields...He Won't Like WhyBessent will get lower long end yields eventually, however he won't like the reasons why. That is not a shot at the man. It is a description of the cycle. You can rearrange the maturity mix. You can jawbone fiscal consolidation.
You can tell reporters that yields do not reflect fundamentals. None of that overrides a growth scare once the growth scare arrives. That reality is not what Bessent or Trump want to manifest especially before the midterm elections.
Look at China if you want the preview: bond yields collapsing because the economy is in a disinflationary grind, not because Beijing discovered a clever buyback program.
Three Pillars of Risk: Growth Scare AheadAt Phinance Technologies we put our US economic outlook on paper in January. An Emerging slowdown with yields set to drop starting in 2026. A deflationary scare is on the horizon. The risks outlined below are not exotic.
They are the white swans sitting on the lawn.
Housing: Roughly 20% of GDP. Forty percent of CPI when you let the shelter component speak. Home prices still too high...call it 30% on our work. New home data has been ugly for months. Builders talking about persistent headwinds with high rates, affordability and cautious buyers. The border closing removed a bid that was quietly holding up rents and prices in a lot of metros. That floor is unwinding slowly, which is how housing always dies...not a bang...a rollover. Southeast first, then the map fills in. A frozen housing market is a frozen chunk of the real economy whether the S&P is making a high or not.
The AI bubble peaking: In my post on July 23rd I outlined that the AI Capex party was approaching closing time. First the private credit market is undergoing flow issues and credit stress making financing more expensive. Since that post Nvidia confirmed those issues on their recent earnings call by disclosing that their balance sheet exploded with extra commitments to suppliers and sweeter payment terms to their customers. They want to become a bank to their customers much like Lucent did in the dotcom days, which did not end well for Lucent. Second Enterprise demand is cracking with ROI skepticism and token pricing backlash. Third power constraints are hitting hard with the grid needing massive additional supply that won't be ready in time for the proposed amount of datacenter projects announced. Finally there is open-source pricing pressure as many users are embracing cheaper models. They call them capex cycles for a reason. The order book always gets inflated near the top, credit is always the disciplinarian.
China entering acute phase of crisis: Factory of the world with fixed-asset investment falling, construction in contraction, real estate still working off a multi-year start collapse, and demographics that do not bottom until 2032. Contagion does not need a press conference. It eventually shows up in Asian supply chains, commodity demand, and the global credit impulse decelerating.
Put those three looming risks on the table at the same time and Bessent will get lower long-end US yields. This is currently not consensus thinking but as the risks manifest themselves and the business cycle exerts its natural downturn the narrative will quickly change. The US long bond is the scoreboard and we believe soon it will begin to respond to these headwinds as we roll through the rest of the year and into the next. In hindsight the current Bessent intervention will be seen as ironic.
The Treasury is not the Fed. The Fed is not the long end. The long end is the cycle.
The signs are not hiding. They are just inconvenient for the people who need the narratives to keep the party going.
Tyler Durden Thu, 09/03/2026 - 09:45Nvidia has agreed to buy AI startup Hugging Face for roughly $12.93 billion, putting the world's most important open-model hub under the same roof as the most dominant AI chip company on the planet, according to The Wall Street Journal.
Image via FTCEO Jensen Huang announced the deal on Thursday, a day after NVIDIA filed that it had signed a definitive agreement. The transaction includes about $11.9 billion for Hugging Face stockholders and an equity retention package of up to $1 billion for employees who join NVIDIA. Closing is expected in the first half of 2027, pending regulatory approvals.
Huang said NVIDIA will “scale Hugging Face’s platform, strengthen its infrastructure and expand access to AI for developers and institutions worldwide.” Hugging Face, he added, “will remain an open platform for the entire AI ecosystem,” and NVIDIA compute will not be required to build or deploy on it.
Hugging Face - where people can find, publish, fine-tune, and deploy open-weight models, has over 18 million developers, researchers, and creators who have shared over 3 million models, 500,000 datasets, and one million applications. Over 200,000 companies use the platform to discover, evaluate, customize, and deploy AI, according to AIM.
It’s great to see Nvidia supporting open-source AI in a big way. Keeping innovation decentralized and accessible is the key to avoiding an unsafe and dystopian future where advanced AI capabilities are centralized and controlled by only a few hands. https://t.co/TjddRfr377
— David Sacks (@DavidSacks) September 3, 2026Founded in 2016 in New York by French co-founders Clément Delangue, Julien Chaumond, and Thomas Wolf, it last raised a $235 million Series D in 2023 at a $4.5 billion valuation. NVIDIA was already an investor, alongside Google, Amazon, Salesforce, AMD, Intel, IBM, and Qualcomm.
Why NVIDIA wanted itThe deal is about the distribution layer. NVIDIA already sells the GPUs most labs use to train and run models. Hugging Face is where those models live, get versioned, and get downloaded. Whoever owns that junction influences which models get used and, downstream, which hardware they run on.
That matters more as OpenAI, Anthropic, Google, and others design their own accelerators to reduce reliance on NVIDIA. Open models still need someone else’s chips. Hugging Face is where a large share of that demand originates.
Hugging Face had previously kept NVIDIA at arm’s length. Late last year it rejected a $500 million NVIDIA investment that would have valued the company at $7 billion, saying it did not want a single dominant investor that could sway decisions. Delangue has argued that concentration of power is the biggest risk in AI. This time he framed NVIDIA as a partner that would keep the platform “open, independent and compute-agnostic,” with a goal of making open source “the default way to build AI” and empowering “100 million AI builders to own their intelligence rather than rent it.”
Tyler Durden Thu, 09/03/2026 - 09:15