TFTC

Bitcoin Brief

Sup, freaks.

Ten days ago, the U.S. government forced Anthropic to shut down its two most powerful AI models globally. Today, a lab in Tokyo dropped a model that claims to match them. Intelligence is becoming a commodity. The geopolitics of AI just got very real, and the implications for sovereignty, infrastructure, and bitcoin mining economics are the through-line connecting everything in today's Brief.

Lead Story

Artificial Intelligence

Japan's Sakana AI Drops a Frontier-Class Model Ten Days After the U.S. Shut Down Anthropic's

On June 12, the U.S. Department of Commerce forced Anthropic to suspend its Fable 5 and Mythos 5 models globally via a Bureau of Industry and Security "Is Informed" letter, citing national security concerns. A reported jailbreak vulnerability, flagged by Amazon (Anthropic's cloud partner), triggered the order. Anthropic disabled both models worldwide because it could not filter users by nationality in real time. As of June 20, both remain suspended. It was the first time the U.S. government had shut down a commercially deployed AI model under export control authority.

Ten days later, Tokyo-based Sakana AI released Fugu Ultra, a multi-agent orchestration system that claims to match Fable 5 and Mythos Preview on the industry's most rigorous engineering, science, and reasoning benchmarks. Sakana, founded by former Google Brain researchers, built Fugu as a single model API that dynamically delegates tasks to a pool of expert models, including instances of itself recursively. The pitch is explicit: "frontier capability without the risk of export controls."

The architecture is what makes Fugu interesting, not just the benchmarks. It is a language model trained to be an orchestrator. It decides when to solve a problem directly and when to assemble a team of specialist models. The agent pool is entirely swappable. If a single provider restricts access, Fugu routes around the disruption automatically. The approach builds on Sakana's peer-reviewed research from ICLR 2026, including their Trinity and Conductor papers on learned model orchestration. This is the first time a non-U.S. lab has credibly claimed frontier parity not by training a bigger monolith, but by learning how to coordinate existing intelligence more effectively.

The skeptic's view is fair: Fugu orchestrates other models, so it is only as good as its pool. Fable and Mythos were not in that pool because of the very export controls Sakana is marketing against. The benchmark comparison is not strictly apples-to-apples. And Sakana is a venture-backed company selling API access. They have every incentive to frame export controls as a selling point. But the structural point holds regardless. As we discussed in our recent conversation with Jamie McAvity about the AI data center gold rush, the infrastructure layer is where the durable value accrues. The models themselves are commoditizing.

The bigger question Fugu raises is what happens to the hyperscaler and closed-source frontier model valuations if orchestration-based alternatives can deliver comparable performance at a fraction of the cost. If open-source and multi-agent systems can reach parity while providing compute at a 90% discount to what Anthropic and OpenAI charge, the viability of closed-source pricing power becomes an open question. Does that change the valuation of the companies building and hosting GPU infrastructure, or does Jevons' Paradox come into play? If cheaper intelligence means dramatically more usage, especially once robotics enters the picture, total compute demand could dwarf what anyone is modeling today. The infrastructure buildout may not slow down. It may accelerate.

This is the AI version of what Bitcoin already proved: you cannot control information flows with national borders. The more governments try to restrict access to intelligence, the faster the ecosystem routes around them. As Brandon Bailey put it on today's TFTC episode: "Intelligence is now a commodity. You can't say enough about the importance of that for the world and for civilization."

Signal

Bitcoin Mining

TFTC 761: Brandon Bailey on Why Bitcoin Miners Are Sitting on the Most Undervalued Power Portfolios in Tech

Why it matters: Bitcoin miners are trading at a 5x discount to AI data center operators. The re-rating thesis is live.

Brandon Bailey (Director of Corporate Development at Nakamoto) joined today's episode to break down the calculus bitcoin miners are making as they transition to AI compute infrastructure. CoinShares' Q1 2026 report projects that listed miners could derive as much as 70% of their revenue from AI by year-end, up from roughly 30% today. "You've got Google, Amazon, Microsoft as your tenant. That lease is as good as gold," Bailey said.

The numbers on the mining side are stark. Network hashrate has fallen from an all-time high of 1.151 ZH/s in October to roughly 0.888 ZH/s. The June 13 difficulty adjustment of -10.09% was the 11th largest downward adjustment in Bitcoin history. Bitdeer has sold its entire bitcoin holdings. Cango sold $305 million in BTC to fund its AI pivot.

But Bailey's counter-thesis is the bullish one: as power capacity leaves the mining space for AI and gets locked into long-term leases that cannot easily pivot back, mining economics should improve materially if and when price recovers. The miners who stay are buying a golden era.

There is a second-order effect worth watching. As mega miners transition their facilities to AI compute, they are forced to sell their mining equipment. Those machines do not disappear. They flow to a more diverse set of buyers who plug them in at smaller, off-grid operations that are not competitive or viable for AI workloads. The net effect could be a more geographically distributed hashrate, spread across sites that AI hyperscalers would never touch. The network gets more decentralized precisely because the big players leave.

Artificial Intelligence

Jordi Visser: This Is Not a Bubble. It Is the Most Important Moment in Human History.

Why it matters: The former CIO of Weiss Multi-Strategy Advisers is rotating from early-cycle AI winners to late-cycle physical bottlenecks.

Jordi Visser's latest "Signal Over Noise" video from 22V Research pushes back hard on the AI bubble narrative. His argument: AI capex is being funded by cash-rich companies with real earnings growth, not dot-com speculation. The professional consensus keeps calling it a bubble and has been wrong consistently. He is rotating out of early-cycle winners like memory and chip stocks and into late-cycle physical bottlenecks: power, energy infrastructure, chemicals, silver, and lithium.

On Bitcoin, Visser is long-term bullish. AI needs verifiable authenticity and ownership tracking, which plays directly to Bitcoin's strengths. But he warns that BTC faces short-term capital competition from high-flying AI equities. His key line: "The risk isn't that AI disappoints. It's that it keeps improving while everyone treats it as a novelty."

On-Chain

Bitcoin Network Activity Hits 2024 Highs. 80% of Transactions Are Dust.

Why it matters: The busiest Bitcoin network in two years, but the economic content has changed fundamentally.

Daily Bitcoin transactions have topped 800,000, the highest since late 2024, with the network activity index sitting just 7% below its all-time high. But according to CryptoQuant, sub-0.01 BTC and sub-0.001 BTC transfers now account for roughly 80% of all daily transactions, up from around 44% in 2023. The driver: Ordinals, Runes, BRC-20 tokens, and OP_RETURN data stamping, which surged after the OP_RETURN byte limit was removed last year.

The mempool has expanded to approximately 128,000 pending transactions, the highest level since February 2025. CryptoQuant warns that "sustained expansion could drive fee increases for time-sensitive economic transactions." The divergence is stark: the network has never been busier, but the price sits at $64,600, down roughly 50% from the October 2025 high of $126,080. High transaction counts correlated with rising prices in prior cycles. This time, the volume reflects protocol use, not a surge in financial demand.

Macro

Howell: Let the Markets Tighten. Warsh Will Welcome a Stronger Dollar.

Why it matters: The Capital Wars thesis says the new Fed chair will not hike rates. He will let the market do it for him.

Michael Howell's latest Capital Wars note argues that new Fed Chair Warsh will not sanction rapid policy rate hikes, but he is willing to see U.S. monetary conditions tighten via higher bond yields and a firmer dollar. The logic: AI capex spending is largely insensitive to rate hikes. The Treasury's enormous interest bill means higher rates actually boost government income transfers to the private sector. Housing is weak but Wall Street is buoyant. A stronger dollar, Howell writes, "might be a neat solution."

The bigger framework is a currency story. China has engineered roughly 30% real depreciation since 2014 through domestic deflation rather than nominal RMB devaluation. Howell's "Shanghai Accord II" thesis: Asian currencies are now constrained within a managed dollar-linked corridor. If the dollar rises, they move with it. The adjustment burden falls elsewhere, likely towards European currencies. Weaker euro is the call. For anyone watching the Warsh Fed's hawkish debut last week, this framework fills in the mechanism.

Market Structure

ETF Outflow Pace Drops 87% as Bitfinex Calls the Market "Under-Positioned"

Why it matters: Six weeks of bleeding, but the wound is closing. The question is whether spot demand fills the gap.

U.S. spot bitcoin ETFs logged a sixth consecutive week of net outflows: $226.8 million in the week ending June 18, bringing the six-week total to $5.94 billion and the 30-day record to $6.35 billion, per Galaxy Research. But the pace dropped sharply, from $1.72 billion in the first week of June to just $227 million last week. Bitfinex analysts note that funding rates remain subdued and leverage has not expanded. "Investors remain cautious given the shift in the macro regime, while institutional and treasury-style buyers continue to provide the marginal bid," the firm wrote. "That combination points to an under-positioned market rather than an overheated one." Margin longs are building in a range 10-25% off recent lows, a positioning pattern that has historically preceded medium-term bottoms.

AI Hardware

Nvidia's Rubin Ultra Needs Four Dies Stitched Together. The Packaging Bill Is the New Bottleneck.

Why it matters: Advanced chip packaging is becoming the dominant cost variable in AI hardware. This is the physical scarcity Visser is pointing to.

Nvidia's next-generation Rubin Ultra reportedly requires four reticle-sized compute dies stitched together into a single package. The reticle limit (roughly 858 mm²) means you cannot make a single die any bigger. You have to connect them. A deep dive from Chip Strategy breaks down the two dominant approaches: TSMC's CoWoS (Chip-on-Wafer-on-Substrate) and Intel's EMIB (Embedded Multi-die Interconnect Bridge). There are rumors of warpage problems on the 4-die package, with TSMC potentially needing panel-level packaging or a 2+2 configuration as a fallback. As AI accelerators keep getting bigger, the packaging itself is becoming a larger share of the bill of materials than the silicon. This is exactly the late-cycle physical bottleneck rotation that Visser describes above.

Sponsored

Sponsored

⚡ Freedom Tech Corner

Data Snapshot

If you found this useful, forward it to someone who needs it. If someone forwarded it to you, subscribe here so you do not miss the next one.

See you tomorrow,