Bitcoin Brief

Sup, freaks.

I have spent plenty of time expecting the bond market to force a reckoning sooner than it has. Today I want to look at the growth bet John and I discussed through a different lens, with help from Michael Every and Jordi Visser. There are a lot of moving parts, and I think it is worth stepping back and working through them.

LEAD STORY

The Growth Bet Has a Lot of Moving Parts

I am writing this as somebody who has spent a long time expecting the bond market to fail imminently. I have been one of those people. The debt is enormous, the interest burden keeps growing, and it is hard to look at the fiscal trajectory and see how it ends well.

But one thing I have learned following markets and bitcoin for well over a decade is that governments and central banks can kick the can down the road a lot longer than we like to think. Being right about the structural problem does not mean we have the timing right.

That is why I want to look at this through a different lens and think it through. Keep the debt problem in mind, try to read between the lines of what policymakers are doing, and be sober about what the data is telling us. I am not dismissing the people worried about a bond collapse. I share the concern. I just want to leave room for what could happen between here and there.

John and I spent a good chunk of this week's Ten31 Timestamp talking about the growth mindset in the G20 finance track's Asheville statement. Grow the economy, get energy costs down, make it easier to build, and get private balance sheets financing the investment. If productivity and the tax base expand fast enough, the debt becomes more manageable relative to the economy supporting it. That is the bet. I would much rather see people trying to produce their way out of this mess than taxing and regulating what remains of the productive economy into the ground.

There are a hell of a lot of things that have to go right, though.

Cheap energy has to show up as actual electricity and fuel. Permits have to become factories, transmission lines and data centers. Banks and investors have to keep financing the work. The AI companies have to turn all of that investment into products people will pay for, and the businesses using those products have to become meaningfully more productive. You cannot service a loan with a model benchmark. Somewhere along the way, somebody has to generate cash.

I asked Michael Every how narrow the needle is that Washington is trying to thread in our conversation last week. He sees a path through, but the financial, industrial, energy and geopolitical pieces have to work together. Dollar stablecoins could help pull demand into Treasury bills. Credit could be steered toward strategic industries. But a financing mechanism cannot manufacture copper, refine diesel or connect a factory to the grid. Even if you get the monetary plumbing right, the physical work still has to happen. And the geopolitical and political risks do not disappear while you are doing it.

Jordi Visser's video this week is a useful check on the tendency to jump straight from those risks to a bond crisis. He argues that elevated yields need to be read alongside bond volatility, credit spreads and inflation expectations. His assessment is that bond volatility is not displaying panic. I think that is worth weighing alongside the longer-term fiscal concerns.

Treasury's official 30-year par yield finished September 4 at 5.24%. Expensive financing, certainly. But the latest available high-yield credit spread was 2.65 percentage points on September 3, while the ten-year inflation breakeven was 2.35% on September 4. Corporate credit was still priced tightly, and market inflation compensation was in the low-to-mid 2% range. Breakevens include risk and liquidity effects, so they are not a pure inflation forecast. They also do not look like investors abandoning all confidence in price stability.

Jordi also makes an important point about corporate profit margins in his accompanying analysis. He describes revenues, earnings and margins rising globally, rather than flattening in the way he looks for ahead of a recession. That matters because a widening margin means a business is keeping more of each dollar of revenue. In his reading, corporate profitability is still showing strength, not the deterioration he would expect going into a downturn. A recession could still arrive. It is another piece of evidence I want to weigh before assuming one is already bearing down on us.

Markets can be complacent. They can change quickly. Every's warning is worth keeping beside Jordi's: calm financial prices cannot tell us that all the physical and geopolitical risks have been solved. But there is a meaningful difference between acknowledging those risks and claiming that a generalized funding panic is already here.

I can think the fiscal trajectory is a disaster and still acknowledge that markets are willing to finance the growth program. I want to keep reminding myself of that. The debt concerns are real, but so is the risk of becoming so attached to a thesis that I stop noticing evidence that complicates it.

The financing is where John's argument gets especially interesting. AI investment is supposed to help generate the productivity that makes the debt burden easier to carry. But the buildout itself needs financing, including more outside borrowing. Nvidia's proposed financing platforms with six major financial firms aim to mobilize more than $500 billion over time, subject to final agreements. The announcement describes a financing target, not completed investment. Treasury needs buyers for its debt. The companies building the infrastructure need investors willing to lend for years against expected future cash flows. Higher long-term rates make that arithmetic harder for both of them.

That does not mean every data center is uneconomic or every AI loan will blow up. It means the timing matters enormously. Can the projects get power, come online and earn money before financing costs or refinancing needs overwhelm them? Can cheaper and more capable models create enough new demand to make up for falling prices? Can the government lower the cost of building without spending away the gains? Those are the questions I want to work through as we watch this unfold.

Michael Howell adds another useful angle in his latest Capital Wars piece: government spending and the way Treasury finances it can support activity even without a new round of Fed bond purchases. He calls his fiscal-duration framework Treasury QE. The name can confuse things. Spending provides the fiscal support; issuing shorter-term debt changes the mix of liabilities investors hold. It is not the same operation as the Fed buying bonds. The practical point is that watching the Fed's balance sheet alone can leave you with an incomplete view of what is supporting the economy.

Put all of this together and you can see why policymakers would want to keep borrowing costs from choking off the very investment they are counting on. That gives them a reason to intervene if funding markets become disorderly. It does not give them perfect control over the long end, and it certainly does not guarantee that every intervention will work. Cheaper financing, stable prices, abundant energy, real productivity gains and fiscal sustainability are a lot of plates to keep spinning at once.

I hope the productive part works. I want abundant energy, better tools and businesses that can build without asking permission from a dozen bureaucracies. Owning bitcoin does not require rooting for everybody else to get poorer.

It does mean I would rather hold money whose supply cannot be adjusted to rescue this experiment. If the economy becomes more productive, scarce money is a good way to save the fruits of that productivity. If the debt wins and policymakers fall back on debasement, the reason to own it becomes even more obvious. Neither path promises a smooth bitcoin price. Both leave me with the same preference for money I can hold outside somebody else's balance sheet.

SIGNAL

Liquid Gets 3,400 BTC Back, but the Network Is Still Paused

The bitcoin started coming back after yesterday's Brief. On September 7, the party holding Liquid's funds returned 3,400 BTC to the federation's return address. The return covers about 85% of the original 3,996.01834922 BTC payout. Roughly 598.5 BTC remained at the holder's address this morning, although that balance also includes funds from outside the exploit.

Blockstream says updated software has been deployed and federation members are preparing a coordinated restart. Its status page still lists a major outage for public bridge nodes. The network has not reopened.

I am glad most of the bitcoin came back. I am not ready to call the retained amount an agreed bounty or the people holding it whitehats. Holders still need an accounting of the remaining gap, a full explanation of the failure, and confirmation that they can redeem their L-BTC.

Nick Shirley Challenges California's Immigration-Provider Privacy Law

Nick Shirley sued California officials on September 4, arguing that AB 2624 threatens his reporting on immigration-services organizations and public spending. The law becomes operative October 1, 2027. He wants an injunction before enforcement begins.

The written-demand provision lets certified program participants demand that their personal information stop being posted online. But the prohibition after a demand retains a specific-intent requirement: inciting likely imminent great bodily harm, or threatening someone in a way that creates objectively reasonable fear for personal safety. It is not a blanket power to erase any unfavorable video.

Shirley argues the broad definitions and threat of litigation will still chill lawful investigations. That deserves a hearing. Protecting people from threats should not become an excuse to make reporters afraid to follow taxpayer money.

Google's TPU Gets an External Inference Scorecard

SemiAnalysis's new Ironwood preview gives outside customers something more useful than another chip-launch promise: an initial comparison of serving economics. At 100 output tokens per second per user, its model puts Ironwood at $0.181 per million total tokens, compared with $0.222 for Nvidia B200 and $0.276 for B300.

Those results cover Qwen3.5-397B-A17B in FP8, with 8,192 input tokens, 1,024 output tokens, aggregated serving and single-token prediction. The dollars come from modeled external ownership costs, not a Google Cloud rental quote. Change the latency target or serving setup and the advantage changes.

I want more competition here. Getting open models running economically outside Nvidia's stack is worth watching. But SemiAnalysis says the tested native TorchTPU path remains private beta, and its public InferenceX project does not yet provide an equivalent reproducible TPU run. Nvidia still has a software advantage. I want to see how quickly that gap closes.

IREN's 2GW Sweetwater Hub Clears a Conditional Hurdle

IREN says ERCOT conditionally included Sweetwater 1 and Sweetwater 2 as base load in Batch Zero. Together they represent 2GW of planned gross power capacity. That does not mean 2GW is operating or cleared to switch on. ERCOT's public notice confirms provisional classifications went out, but does not name the projects.

Both sites already have grid-connection agreements. Sweetwater 1's high-voltage substation connected in May, with power delivery ramping as facilities are built and commissioned. IREN says 300MW gross is under construction for targeted delivery in Q4 2027. Only 15% of the planned hub is in that construction phase.

I keep paying attention to miners' power portfolios for exactly this reason. Land, substations and grid access take real work to assemble. IREN is carrying that advantage into AI infrastructure. Final approvals, construction and the permitted load ramp still have to follow.

BitcoinPIR Adds a Cashu-Paid Cashier for Private Queries

BitcoinPIR's cashier repository is an interesting attempt to make private wallet lookups pay their own way. A client hands the cashier Cashu ecash and receives a signed session grant. BitcoinPIR servers can then verify and meter that grant offline using the cashier's public key, without holding its signing seed or wallet seed.

The payment does not create the query privacy. BitcoinPIR supplies the private-information-retrieval machinery; Cashu pays for access. The cashier swaps the token at its accepted mint before issuing the grant. The mint interaction still happens; the private queries are the BitcoinPIR lookups.

I like the separation. A service needs a way to charge without making every wallet lookup part of a payment trail. The implementation is early, though. I would want an independent privacy audit before trusting a new service with my wallet activity.

ETF FLOW CHECK

The TFTC ETF tracker remains current through September 4, with approximately $1.007 billion of net inflows across three consecutive positive sessions. There is no new completed U.S. session in the tracker after the Labor Day holiday. James Check's latest High & Tight update adds useful context: in his reading, profit-taking is being absorbed without an aggressive futures-leverage buildup, but bitcoin still needs to clear the $80,000 to $83,000 resistance area. Encouraging demand, unfinished breakout.

⚡ FREEDOM TECH CORNER

Ride The Lightning v0.15.12-beta Closes an Unauthenticated Rebalance Route

Ride The Lightning v0.15.12-beta fixes an unauthenticated circular-rebalance route on Eclair nodes and tightens the login and application-settings boundaries. Failed-login lockouts now expire properly, and clients cannot simply rotate an X-Forwarded-For header to get a fresh guessing budget. The settings API also stops authenticated callers from changing sensitive credential paths, server URLs and the 2FA seed.

If you run RTL's own login behind a reverse proxy, read the upgrade notes. Set trustedProxies, or TRUSTED_PROXIES, to the proxy's exact address. A broad range that includes clients can reopen the bypass. Without the setting, clients behind the proxy share a lockout counter. The release says SSO deployments such as BTCPay Server are unaffected by that proxy-login change. Worth handling carefully. A Lightning dashboard has enough authority over your node that its login cannot be an afterthought.

DATA SNAPSHOT

As of September 8, 2026, approximately 9:38 a.m. ET

Sources: Coinbase for spot price; mempool.space for block, fee, hashrate, and difficulty data.

Stay humble. Stack sats.

YouTube: https://www.youtube.com/@TFTC

podcast: https://www.tftc.io/tag/podcasts/