Bitcoin Brief

Sup, freaks.

Let's get into it.

LEAD STORY

Texas Just Put 474 Gigawatts of Data Center Requests on Trial

The AI boom has finally reached the point where the spreadsheet has to answer to the power grid.

Governor Greg Abbott ordered Texas regulators and ERCOT to audit every datacenter moving through the interconnection process before those projects can advance. The queue contains more than 474 gigawatts of requested load. The state says roughly 90% of it comes from datacenters.

Texas' record peak demand is less than one-fifth of that queue.

The 474-gigawatt figure does not mean Texas is about to build 474 gigawatts of data centers. It is a queue of requested load totaling more than five times ERCOT's record peak demand.

Abbott directed regulators to examine each project's financial incentives, dependence on the grid, plans for on-site generation, water requirements, community effects, and ownership. In effect, Texas is asking whether each project's financing, power, water, and operating plans can survive scrutiny.

The timing makes sense. J.P. Morgan forecasts calendar-year capital spending by Amazon, Microsoft, Google, Meta, Oracle, CoreWeave, and SpaceX rising from $443 billion in 2025 to $901 billion in 2026 and $1.489 trillion in 2027. The bank expects debt and equity issuance to help finance the rollout as free cash flow comes under pressure. Its second-quarter table also adds Google backlog of $514 billion, AWS backlog of $496 billion, and Microsoft commercial remaining performance obligations of $678 billion to reach $1.688 trillion. Those measures are not fully interchangeable, and the report cautions that the capital spending may not all support public-cloud revenue.

SpaceX's second-quarter earnings release, furnished as Exhibit 99.1 to its August 4 Form 8-K, reports 1.4 gigawatts of nameplate compute, $2.561 billion of quarterly AI revenue, $15.828 billion of quarterly AI capital spending, and $14.1 billion of contracted sales under Cloud Services Agreements. Quarterly AI capital spending was 6.18 times quarterly AI revenue, a derived comparison rather than a company-reported unit-economic ratio.

On the company's August 4 earnings call, Elon Musk told J.P. Morgan analyst Douglas Anmuth that his estimate for Rubin monetization was $30 to $50 per watt and repeatedly characterized it as a guess. If he meant annual revenue, the arithmetic scales to $30 billion to $50 billion per gigawatt per year. The operative word is guess. The filing does not validate that revenue density.

Musk also identified memory as the current limiting factor and estimated that supply is increasing roughly 20% per year while demand is increasing 200% or more. Again, those are management estimates. They are directionally consistent with the banking reports, which show high-bandwidth memory remaining tight while chip designers begin reducing memory content in some configurations to control cost and secure supply.

Matt Dratch's useful synthesis is that deliverable power and memory should capture more of the AI stack's economics as compute demand accelerates. His specific Anthropic and OpenAI revenue paths are scenarios, not company guidance. The thesis does not need those numbers. The capital spending, contracted demand, Texas queue, and fight for power already make the case.

Bitcoin miners enter the picture here.

Miners were forced to become power-market operators before AI labs discovered the grid. They learned how to secure land, negotiate interconnections, build substations, source generation, curtail during grid stress, and monetize electrons every ten minutes. The market spent years dismissing that expertise as a side effect of mining. AI has revealed that the power stack may be the asset.

That does not make every miner an AI winner. An operating site with energized capacity is different from an application in a queue. A signed interconnection agreement is different from a presentation claiming future megawatts. A customer willing to fund construction is different from a management team adding "HPC" to an investor deck.

Texas' audit may increase the value of sites that can prove their power, generation, capital, and customers. It may also expose speculative miner-to-AI conversions that were built on the assumption that a queue position was the same thing as an energized datacenter.

I've been saying for years that bitcoin miners are energy companies. The AI buildout is forcing the rest of the market to understand why. Chips get the headlines. Power determines whether the machine turns on.

SIGNAL

A Fake Wasabi App Weaponized the COLDCARD Panic

Nick Neuman is warning that scammers are exploiting the COLDCARD panic after a user reported losing 6.06 BTC through a fake Wasabi Wallet app downloaded from Apple's App Store.

The official Wasabi website says the wallet is unavailable on mobile. Apple's listing presented an iPhone app under the developer name "zkSNACKs," while Apple's lookup data named the seller as VFS d.o.o. and linked to wasabimobile.com.

When downloading wallet software, always ensure you are on the project's official site. Find the project's official X account and other independently verified channels, confirm they point to the same domain, then cross-check the domain, supported platforms, publisher, release page, and signatures or hashes. Use multiple independent confirmations. Never treat an app-store result as authentication. A test transaction cannot make unauthenticated software safe if it has captured your recovery words.

Apple says every app is reviewed by experts and scanned for malware, and its rules prohibit misleading or copycat names and metadata. If Apple sells curation as a trust product, it should explain how this apparent impersonator passed review. This is completely unacceptable.

Alex Thorn Maps the Waves and the Recovery Path

In my latest episode with Galaxy Research's Alex Thorn, we map a high-confidence set of 1,596 BTC from roughly 7,300 addresses and a broader pattern-matched set exceeding 2,000 BTC that still includes unconfirmed activity.

Wave 1 was a mechanical 41-minute sweep. Wave 2 used a similar funnel structure, and Galaxy has medium-to-high confidence that the same operator may control both. Wave 3 split 293 transaction chains through 293 staging addresses into separate P2WSH holding addresses with no common collector. The major-wave balances were inert at recording time, not frozen or recovered. Smaller attackers have already moved funds through THORChain and an offshore casino.

Victims should secure any remaining funds, file reports with local police and the FBI's IC3, and send drained addresses and attacker transaction IDs to Alex. Galaxy can return a forensic report for authorities and platforms. Keep the physical COLDCARD as potential ownership evidence. Reporting cannot guarantee recovery, but it creates the paper trail recovery may require.

Capital Was Already Winning Before AI Hit Scale

J.P. Morgan says labor received 53.7 cents of every dollar of nonfarm-business value added in the first quarter of 2026, an all-time low and roughly three cents below the pre-pandemic level.

The bank's private-sector shift-share comparison of 2018-19 with 2023-24 finds that 74% of the decline occurred within industries rather than because output shifted toward sectors with lower labor shares. Automation, market concentration, globalization, and offshoring all belong in the explanation. The report does not establish that AI caused the current low.

AI is arriving after the nonfarm-business labor share has already reached a record low. J.P. Morgan sketches a scenario in which labor's share falls below 50% by the end of the decade. The bank does not present it as a forecast. The warning is clear anyway. If productivity gains accrue to owners while wages and hiring absorb the adjustment, the political reaction will be severe.

The Liquidity Tailwind Is Getting Thinner

Michael Howell's latest model puts global liquidity at $193.6 trillion, but three-month annualized growth has slowed to 5.1%. His shadow monetary base fell to $109.5 trillion and contracted at a 5.4% annualized rate over the same window. Howell says low bond volatility, stable collateral, and a softer dollar are offsetting weaker central-bank liquidity.

Citi separately expects the Treasury's cash rebuild and heavier August bill issuance to drain reserves and add repo collateral. The bank thinks that could push the Secured Overnight Financing Rate toward or above the interest rate the Fed pays on reserve balances. Its forecast is modest tightening, not a replay of last autumn's funding stress, with some easing expected in September.

These are model outputs and bank forecasts, not mechanical laws. The useful distinction is between bitcoin's structural monetary case and the cyclical environment around it. Governments can keep debasing while the marginal liquidity impulse weakens and dollar funding becomes more expensive.

The Memory Shortage Is Already Redesigning the Machine

J.P. Morgan expects memory supply to remain below demand for two to three years. Goldman forecasts SK Hynix's blended high-bandwidth-memory selling price near $2.90 per gigabit in 2027, approximately double its 2026 estimate.

The buyers are already adapting. J.P. Morgan says Nvidia and AMD are planning lower-HBM-density accelerators and reduced SoCAMM content to mitigate tight DRAM supply and higher bill-of-material costs. These are product-roadmap observations, not proof of final shipped configurations.

Scarcity raises prices. Then it forces redesign. Memory manufacturers can capture extraordinary economics while the shortage lasts, but sufficiently high prices create an elasticity ceiling. Chip designers are planning to change the machine before accepting every quoted price forever. The buildout remains real, but the configurations cited by the bank have not yet been confirmed as shipped products.

⚡ FREEDOM TECH CORNER

Bitcoin's Red Team Needs More Than More Agents

Rob Hamilton said the Bitcoin Red Team had scanned more than 300 repositories and spent nearly $40,000. Calle's latest campaign update says the effort has grown to 16 globally distributed people working around the clock.

At 27.5 hours, Calle reported 4,962 findings across 390 projects, including 85 classified critical and 635 high severity, at 2.31 high-or-critical findings per person-hour. The attached situation report says 21.4% of findings had proof-of-concept evidence, 91% arrived through automated scan intake, and 19 projects had findings disclosed upstream.

Those are campaign classifications, not 720 independently confirmed exploitable vulnerabilities. The funnel still needs humans to reproduce findings, understand code in context, judge severity, work with maintainers, and coordinate disclosure without handing attackers a roadmap. The fact that only 19 projects had received upstream disclosures shows how far validation and communication lag raw discovery.

That gap is now the story. AI can saturate the discovery funnel faster than maintainers can absorb it. Bitcoin needs shared triage, reproducible evidence, careful disclosure, and a feedback loop that teaches future reviews which signals matter. More agents can generate more reports. Human judgment determines whether those reports make bitcoin safer or merely bury developers.

DATA SNAPSHOT

As of August 5, 2026, 11:01 a.m. ET

See you tomorrow. This is not investment advice. Do your own research.

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

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