TFTC

Bitcoin Brief

Tuesday, June 2, 2026 • Block 952,098 • BTC $68,412 • 1,462 sats/$

Sup, freaks. Google just did something most of the market has not processed yet. Both Gemini 3 and Claude Opus 4.5 were trained on Google's TPUs, not NVIDIA hardware. Anthropic is building more than a gigawatt of TPU capacity. Meta reportedly wants TPUs for training, not just inference. The AI chip landscape that everyone assumed was a permanent NVIDIA monopoly is fragmenting in real time. We dig into what this means for the $90 trillion buildout thesis, the compute bottleneck, and why Bitcoin miners should be paying attention to what happens next in the chip wars.

Lead Story

Google Trained the Two Best AI Models on Its Own Chips. The AI Chip Landscape Is Fragmenting.

Google's Tensor Processing Units have gone from an internal research tool to the training hardware behind two of the most capable frontier models on the planet. Gemini 3 and Anthropic's Claude Opus 4.5 were both trained on TPUv7 (Ironwood), not on NVIDIA GPUs. SemiAnalysis called it "the 900lb gorilla in the room." Citrini Research, which called the Google trade back in August, says the market's perception has dramatically reversed from "AI loser" to "stalking horse undercutting the consensus winners."

The adoption is broadening fast. Anthropic has committed to a buildout exceeding 1 GW of TPU capacity. Meta reportedly wants TPUs for training workloads, not just inference. Google Cloud estimates expanded TPU adoption could capture up to 10% of NVIDIA's annual datacenter revenue. Given NVIDIA's $51 billion quarterly datacenter haul, even a modest shift represents billions in displacement. Google's own DeepMind researchers are now queuing for TPU compute alongside paying cloud customers, a sign of just how constrained supply is.

This matters because the entire AI infrastructure thesis has been built on NVIDIA dominance. Jordi Visser's $90 trillion buildout framework, John Tinsman's compute leasing math ($3-4B to build, $15B/year to lease), the bottleneck cascade in copper, silver, and power. TPUs have been on the radar for years, but the market had largely treated NVIDIA as the default. If the compute layer is fragmenting, the bottleneck math changes. More competition means lower margins, which means cheaper compute, which means faster AI deployment, which means more energy demand, not less. The physical constraints Visser flagged do not go away. They accelerate.

Citrini Research raises two frameworks for thinking about this. First: switching costs are high. Labs locked into NVIDIA will rush to buy next-gen chips to compete with Gemini 3, which is broadly bullish for compute demand. Second: NVIDIA's margins are so high that any competitive threat must be aggressively priced in even at higher volumes. In other words, a breakthrough model is bullish for the buildout even if it is bearish for NVIDIA's margin structure.

The bigger picture: more competition in AI chips is a net positive. More suppliers mean more efficiency, lower costs, and a faster path to making this technology available to everyone at reasonable prices. The NVIDIA monopoly narrative was always going to break eventually. The question was when, not if. What matters now is whether the buildout accelerates as a result. For Bitcoin: the energy demand story does not change. It intensifies. More competition in chips means compute gets cheaper per unit, which means more of it gets deployed, which means more power consumption. The binding constraint was never the chip. It was the grid, the copper, the cooling, the physical infrastructure. Google training frontier models on TPUs does not solve the energy problem. It makes it bigger. Visser's scarcity portfolio thesis (silver, copper, Bitcoin) holds regardless of who makes the chip.

Sources: SemiAnalysis, Citrini Research, VentureBeat

Signal

Lightning

LDK Server Ships. It Already Powers 25% of Lightning Volume.

Spiral released a production-grade Lightning daemon built on LDK at Bitcoin 2026. Cash App and Square already run on it.

LDK Server is a new high-performance Lightning daemon built on the Lightning Dev Kit, the framework that already powers an estimated 25% of Lightning Network volume through integrators like Cash App and Square. Released by Spiral at Bitcoin 2026, it brings splicing, BOLT 12, asynchronous payments, zero-fee commitments, and full Lightning Service Provider support. This makes it significantly easier for developers to run their own Lightning infrastructure and become LSPs. A preview is available on GitHub. If you care about Lightning decentralization, this is infrastructure worth watching.

Bitcoin Protocol

Brink Funds Post-Quantum Bitcoin Research. The Clock Is Running.

Quantum-resistant cryptography for Bitcoin is no longer theoretical. Brink is now funding a dedicated researcher.

Brink announced it is supporting "conduition," a cryptographic engineer focused on post-quantum research for Bitcoin. The work includes driving the SHRINCS protocol toward a draft BIP, exploring commit/reveal protocols, and advancing isogeny-based cryptography as a potential long-term replacement for Bitcoin's current elliptic curve system. Separately, Paradigm researcher Dan Robinson proposed PACTs (Provable Address-Control Timestamps) in May, a scheme allowing Bitcoin holders to prove address control via a blockchain timestamp without moving assets, as preparation against quantum threats. Glassnode recently estimated 6 million BTC (30% of circulating supply) is vulnerable to quantum-exposed addresses. The research is catching up to the risk.

Source: Brink

Mining

Open Hardware Mining Is Growing Faster Than Its Infrastructure. PlebBase Says That Is the Next Problem.

Bitaxe, NerdQaxe, NerdQX, and Ocho are scaling fast but documentation, safety standards, and comparability are lagging.

PlebBase published a piece arguing the open hardware home mining movement has hit an inflection point. Single-chip miners have evolved into multi-ASIC boards delivering hashrates that would have been absurd for home mining a year ago. The NerdQaxe++ has been overclocked to 10 TH/s. The Bitaxe GT is shipping. But the infrastructure around it, documentation, safety standards, thermal management best practices, production quality benchmarks, is not keeping pace. The next hurdle is not a faster ASIC. It is structure. The community solves technical problems faster than it builds the frameworks to make those solutions safe and reproducible.

Source: PlebBase

Mining / Energy

4CP Season Is Here. Texas Miners Are Managing the Grid.

Bitcoin miners in ERCOT territory are now active demand-response participants during peak load season.

Luxor flagged that 4 Coincident Peak season has begun in Texas. 4CP determines transmission costs for large power users based on their consumption during the four highest system-wide demand intervals of the year. Bitcoin miners who curtail during those peaks can save millions in annual transmission charges. Intelligent mining software automates the process, monitoring grid conditions and shutting down hashrate before peaks hit. This is the practical version of the "Bitcoin miners stabilize the grid" thesis playing out in real time during the highest-demand months of the year.

Source: Luxor

Macro / Liquidity

Howell: Liquidity Edges Higher on Narrow Drivers. Headwinds Persist.

Michael Howell's Global Liquidity Watch is the macro framework Marty references regularly. Fresh update today.

Michael Howell's Capital Wars published his weekly Global Liquidity Watch this morning. The headline: "Liquidity Edges Higher on Narrow Drivers: Headwinds Persist." Global liquidity is inching up but the improvement is narrow and fragile. This matters for Bitcoin because Howell's research shows BTC price correlates with global liquidity more tightly than any other macro variable. Narrow, headwind-prone liquidity improvement suggests the path higher for Bitcoin is grinding, not impulsive. Consistent with the on-chain picture showing stabilization without conviction.

Source: Capital Wars (paywalled)

On-Chain

On-Chain: LTH-SOPR Back to 0.82. Net Realized Losses at $735 Million. STH $7,400 Underwater.

The weekend stabilization reversed. Macro pressure from Iran is showing up in the chain data.

LTH-SOPR pulled back to 0.82 from 0.967. Net realized P/L at -$735 million vs -$31 million two days ago. MVRV Z-Score at 0.59, fair value range. STH realized price at $76,958 with BTC at $68,412. NUPL at 0.243. The chain data is reflecting the macro: Iran cutting communications, oil surging, the BOJ hike two weeks out. The question remains whether LTH cost basis at $48,716 holds if a carry trade event materializes on June 16.

⚡ Freedom Tech Corner

Update to Sparrow Wallet 2.5.2. Silent Payments Are Here.

Sparrow Wallet released version 2.5.2 on May 31. The bigger feature landed in 2.5.0: Silent Payments (BIP 352) receiving wallets. Silent Payments let you publish a single static address that generates unique on-chain addresses for every sender, eliminating address reuse without requiring interaction. This is a meaningful privacy upgrade for anyone receiving bitcoin regularly. Download the update at sparrowwallet.com and verify your binary using the manifest signature.

Source: Sparrow Wallet

Data Snapshot

If this landed, forward it to someone who could use more signal and less noise. The Bitcoin Brief is free, always will be.

See you tomorrow,

Marty Bent

Follow: @MartyBent · @TFTC21

Nostr: primal.net/marty

YouTube: TFTC · Podcast: tftc.io/podcast