Bitcoin Brief

Sup, freaks.

Private credit has spent years inside a very comfortable arrangement. The marks come quarterly. The valuations are friendly. Everybody gets to pretend the asset is fine because nobody has to sell it today.

Insurance supplies the money. Asset managers manufacture the products. A thin capital cushion sits between those assets and a mountain of promises made to policyholders decades into the future.

Now the public credit market is putting a real price on the arrangement.

Nick Nemeth has been warning us about this all year. The latest selloff in Sammons bonds falls well short of proving his thesis that the whole thing is about to come apart. It makes the risk much harder to wave away as an academic argument about private markets.

Let's get into it.

LEAD STORY

The Insurance-Private Credit Machine Is Finally Getting Marked

Nick Nemeth posted a five-minute video this morning walking through the ties between Guggenheim, Mark Walter, Sammons Financial Group, and a wider web of insurers and private assets.

Nick goes pretty far in the video. He thinks Guggenheim is going to fall apart. He believes Walter will have to sell more sports assets. He says the regional-banking crisis of insurance has arrived.

Those are big predictions. Time will tell if they prove to be facts. At the very least, the documents underneath his argument make the thesis impossible to wave away.

Federal prosecutors and the SEC are reportedly examining disclosure and related-party questions involving insurers linked to Walter. Delaware Life reclassified a large block of investments as affiliated. Fitch said the episode exposed transparency and governance concerns that could carry consequences across the life-insurance sector. Nothing in the material we reviewed reports charges against Walter, Guggenheim, Sammons, or their executives.

Sammons is a separate company from the insurers Walter controls. Its connection to Guggenheim runs through ownership, investment management, lending, fees, and governance.

Sammons' 2023 audited financials said Guggenheim Partners Investment Management managed 87% of its investment portfolio. Sammons paid GPIM $63 million in investment-management fees and another $11 million in commercial-mortgage fees that year.

After a 2024 restructuring, Midland National said Guggenheim was no longer considered an affiliate or related party. Yet current Guggenheim disclosures still describe Sammons as holding substantial economic and voting interests, supplying leverage to a private fund, and remaining a significant source of advisory fees. Sammons' June 30 statutory filings still name GPIM as an authorized investment manager and mark it unaffiliated.

Maybe the reclassification is perfectly proper. Fine. We should still ask the obvious question: how unaffiliated is a relationship when the money, fees, lending, and voting interests remain this tangled?

Now look at the capital underneath it.

Sammons' second-quarter supplement reports $152.3 billion of total assets, $126.5 billion of investments, and $21.0 billion of private credit at the consolidated group level. In a separate set of figures, it reports $7.6 billion of combined statutory capital and surplus for Midland National Life and North American Company, plus a 386% combined risk-based-capital ratio at year-end 2025. It lists stable A+ insurer financial-strength ratings from S&P, Fitch, and AM Best for those two insurers and says 96.3% of the consolidated private-credit portfolio carries NAIC 1 or 2 designations.

Those are strong reported metrics. They also show why the usual exposure-as-a-percentage-of-assets framing can lull people to sleep. When a forced mark arrives, it hits capital, surplus, and available liquidity.

Hunterbrook's investigation maps the relationship that remains and documents several investments that crossed the affiliation boundary. Hunterbrook Capital disclosed that it was short Sammons bonds when the report came out. Keep that conflict stapled to every claim sourced to Hunterbrook.

Then the public market weighed in. Bloomberg reported Sammons bonds had fallen to their lowest level since issuance after the scrutiny around its Guggenheim ties picked up.

That does not make Sammons insolvent. It tells us lenders want to be paid more to carry the uncertainty. Now we have a public market putting a harder price on the risk Nick and I have been talking about.

In TFTC #734 on April 8, we called private credit the fuse and insurance the bomb. The mechanism is pretty simple. An asset manager gets access to insurance liabilities, collects management fees, and directs policyholder capital into affiliated or privately valued assets that rarely face an outside mark.

In TFTC #763 on June 27, Nick explained how payment-in-kind interest, stale marks, layers of leverage, and captive insurance funding could push private-credit stress through insurer balance sheets and eventually into the rest of the financial system.

We have not reached contagion. We have reached the point where the market is putting a visible mark on the exact structure we were talking about.

Japan is flashing a different warning from another corner of the insurance market. Nikkei estimates that Japanese life insurers are sitting on close to $200 billion of unrealized losses in domestic bonds as Japanese government bond yields rise.

The Japanese insurers own public sovereign bonds. Guggenheim and Sammons are dealing with private assets and affiliated relationships. The accounting, liquidity, and governance issues are different. Higher rates also lower the present value of long-duration insurance liabilities and give insurers better reinvestment yields. An unrealized loss by itself does not prove insolvency.

The common problem is the balance sheet. Insurers make enormous promises with a much smaller pool of capital underneath them. The pain starts when rates, redemptions, or credit losses force a real mark against that pool.

UBS gave us a useful check on the thesis in an August 17 CIO note. The bank is neutral on direct lending and does not see the entire asset class as impaired. It does see liquidity restrictions and stress among smaller, junior, and overlevered borrowers. Stress starts at the weak links.

If this turns into a real scramble for liquidity, bitcoin will probably get sold with everything else in the opening wave. Liquid assets get sold when people need cash now.

My longer-term bitcoin thesis begins with what the policymakers do next. They will not sit back and let giant pools of insurance liabilities, private assets, and institutional credit clear honestly if the marks begin threatening the system. They will reach for guarantees, regulatory relief, emergency liquidity, lower rates, and a cheaper currency. They always do.

For years, opacity let this machine avoid the mark. The public market has started demanding one.

SIGNAL

Global Liquidity Is High. The Impulse Is Fading.

Michael Howell's August 18 Global Liquidity Watch puts nominal global liquidity at $194.4 trillion. The pool is enormous. The impulse behind it is fading. Three-month annualized growth has slowed to 2.9%, down 2.2 percentage points from his August 5 reading. Annual growth is 6.2%.

Howell's argument is straightforward. Markets are leaning on calm funding conditions and firm collateral instead of getting a fresh shove from central banks. Low bond volatility keeps collateral haircuts down and lets the machine run. If volatility at the long end jumps, that support can disappear quickly.

Monday's bitcoin ETF tape looked better with $297.5 million of net inflows. Zoom out one notch and the six sessions from August 10 through August 17 still show $87.7 million of net outflows on Farside's series. There is plenty of liquidity in the system. It is losing speed.

NVIDIA Is Using Its Balance Sheet to Make the AI Buildout Bankable

NVIDIA's August 17 8-K discloses residual-value guarantees tied to leases for approximately 4.25 gigawatts of IT load at SB Energy's Portsmouth, Ohio campus. OpenAI is the tenant.

NVIDIA's cumulative obligation for the initial commitment is capped at $105 billion. The guarantees are expected to begin when lease tranches start entering service in 2028, and specified OpenAI defaults can trigger them. OpenAI agreed to reimburse and indemnify NVIDIA for amounts paid.

Do not read $105 billion as cash NVIDIA has already spent, debt already outstanding, or the cost of the project. It is a contingent cap. Even with that caveat, this is a wild structure. Selling chips is no longer enough. NVIDIA is putting its own balance sheet behind the land, power, shells, and leases required to keep the AI demand machine humming.

If You Use a BitBox, Update the Firmware

BitBox released firmware 9.26.5 after disclosing two severe security flaws. One involved memory corruption in the Multi edition of BitBox02 and BitBox02 Nova devices running firmware through 9.26.4, when no wallet had been set up and the device was used with a malicious host. The Bitcoin-only edition was unaffected by that issue. The other flaw could lock funds sent through Silent Payments. It could affect BitBox02 and BitBox02 Nova devices running firmware 9.21.0 through 9.26.4 when used with a malicious host. BitBox reports no known exploitation or stolen funds.

BitBox calls the bugs severe, not critical. Update before you use the device again, but do not mistake this disclosure for an active drain.

Install the update through the official BitBoxApp you already use. Ignore firmware links sent through email, direct messages, or replies. The phishers always show up when a security disclosure lands.

The Stablecoin Law Passed. Now the Definitions Decide Who Gets Trapped Inside It.

Treasury issued a proposed rule to implement section 3 of the GENIUS Act. The proposal defines when a payment stablecoin is issued, offered, or sold in the United States and starts drawing the licensing and foreign-issuer boundaries Congress left behind.

The rule is open for public comment and remains a proposal. It has no final legal force yet.

The broad promise of clarity turns into a jurisdictional trap during rulemaking. The definitions will determine which issuers need licenses, which intermediaries can serve U.S. customers, and which foreign stablecoins U.S. platforms can offer.

Stablecoins make dollars easier to move. They can also carry the dollar's surveillance, seizure, and compliance perimeter straight into software that people casually call permissionless. Congress passed the statute. The implementation language will tell us how far the trap extends.

Nostr Key Recovery Got Better Without Handing the Keys to a Custodian

Sidecar 1.9.0 adds printable and scannable Nostr-key backups generated inside the browser, including password-encrypted ncryptsec. The release also brings consent-first wallet reads, relay controls for each account, encrypted drafts, and a Forget all sites control.

The project says it finished remediation from an external security audit by tightening browser-extension message boundaries, restricting callbacks, keeping raw Nostr Wallet Connect strings inside the worker, and rejecting crafted high-work-factor inputs. It does not link the audit report. For now, remediation remains the project's claim rather than an independently reviewed result.

Printing a secret introduces a new pile of ways to screw this up. Printers, PDF queues, cloud drives, and camera rolls are all terrible places to leak identity keys. Still, the direction is right. Give people better recovery tools without making them surrender a Nostr identity to a custodian.

⚡ FREEDOM TECH CORNER

The Model Is Becoming a Commodity. The Harness Is Becoming the Product.

I have been watching this shift happen in real time. The model is becoming one piece of a much larger product. The agent built around it is where the real value starts to accumulate.

Nous Research introduced Bot Mode for Hermes Desktop. An agent profile becomes a named bot with its own role, model, memory, skills, and profile picture. You can use any model, have the bots talk to each other, and build them once for jobs you need done again. Hermes Agent is open source and model agnostic. That means the user can run the harness and decide which intelligence goes inside it.

xAI launched Grok Bot in early beta. Its pitch is simple: give the bot a job, let it sign into your tools, and get completed work back. The beta is available on desktop and iOS to SuperGrok Heavy, Cursor Ultra, and Cursor Teams Premium subscribers.

Nous is taking the open, self-hostable route. xAI is selling the managed, polished route. I expect both approaches to find a market.

The part I care about is what happens after you pick the model. Can the agent remember? Can it use tools? Can it work with other agents, respect permissions, recover from a mistake, and finish the damn job? A smart model with no memory or execution wakes up in a blank room every morning. Give it durable context and real tools, and it becomes an operator.

That makes ownership the fight to watch. An agent that understands your business, holds credentials, and accumulates operating history becomes critical infrastructure. If you cannot move the memory, skills, approvals, and operating context somewhere else, you do not own the relationship. You are renting it.

Open runtimes and portable memory matter for exactly this reason. The models will get cheaper. The context an agent earns while working with you will get more valuable.

DATA SNAPSHOT

As of August 18, 2026, approximately 10:09 a.m. ET

Sources: Kraken for spot price; mempool.space for block, fee, hashrate, and difficulty data; Farside Investors for August 17 ETF flows; TFTC Bitcoin ETF Flows for ETF assets through August 14.

See you tomorrow. Nothing here is investment advice. Do your own research.

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

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