When Mining Hardware Became a $41.9 Million Lesson: Block's Proto Chip and the Death of Crypto's Industrial Dream

We didn’t see it coming. Not because the signs weren’t there, but because we were all too busy worshipping the narrative of the next great hardware revolution. I remember sitting in a meetup in Sydney last year, a founder of a new ASIC startup glowing with excitement about 3nm chips and the promise of bringing competition to Bitmain’s stranglehold. I wanted to believe it too. But deep down, I knew the truth: hardware is unforgiving. You can’t ship a whitepaper and call it a product. You can’t fork a codebase and pretend you’ve built a better mousetrap. The ledger doesn’t lie, and neither do customers when they’d rather pay $41.9 million to walk away than use your chips. That’s the story of Block’s Proto mining chip. Core Scientific, once its biggest and only public customer, terminated its contract for Block’s 3nm ASIC miners and took a $41.9 million pre-tax impairment charge. Yes, they paid forty-two million dollars to say “no thanks.” Why would a company do that? Because the alternative—actually deploying those miners—would have cost them far more in lost opportunity, diminished hashpower competitiveness, and stranded resources. This isn’t a story about a contract dispute. It’s a story about the fundamental misalignment between crypto idealism and the brutal physics of chip design. Let’s rewind. Block, Jack Dorsey’s company, decided to enter the mining ASIC market in 2021, announcing a custom chip development project. By 2023, they had a 3nm design and a manufacturing agreement with a foundry. Core Scientific, one of the largest public mining companies, signed on as a customer. The narrative was perfect: a beloved tech icon taking on the duopoly of Bitmain and MicroBT, bringing decentralization to the hardware layer. We cheered. We retweeted. We bought the vision. But hardware doesn’t care about visions. The first signal something was wrong came when Block revealed the chip’s performance metrics—or rather, what they didn’t reveal. In a 2024 investor presentation, they boasted about reaching 15 exahash (EH/s) of total hashrate capacity, but omitted the most critical metric: energy efficiency in joules per terahash (J/TH). For context, Bitmain’s S21 series operates around 17.5 J/TH. MicroBT’s M50 series is in the same range. If Block’s chip was competitive, they would have shouted that number from the rooftops. Their silence was deafening. Truth in blockchain isn’t written in code; it’s written in the quiet decisions of those who hold the purse strings. Core Scientific’s decision to walk away tells us everything. They didn’t just cancel an order; they paid a penalty that was cheaper than the losses they would have incurred running those miners. That means the chip’s real-world efficiency was probably well above 30 J/TH, maybe even higher. In the current mining environment, where post-halving margins are razor-thin and power costs dictate survival, a 30+ J/TH miner is a liability, not an asset. But the story doesn’t end at the chip’s specs. It extends to the entire ecosystem of crypto hardware. Core Scientific wasn’t just canceling a contract; they were signaling a strategic pivot that should terrify anyone who believes Bitcoin mining will remain the default use case for large-scale power infrastructure. The company announced a 15-year, $14 billion revenue contract with AMD to provide AI data center services. They are repurposing their Texas facilities from Bitcoin mining to high-performance computing for AI workloads. This is the real news. Not the $41.9 million write-off, but the fact that one of the largest publicly traded mining companies has chosen AI over Bitcoin. Core Scientific’s management effectively said: “We can make more money, more reliably, by renting out our buildings and power to AMD than by running ASICs.” They didn’t just walk away from Block’s chips; they walked away from the very premise that mining is the highest and best use of their assets. Let me be clear: this isn’t about Block failing. It’s about the end of an era. The “industrial dream” of crypto—building giant warehouses filled with custom chips to secure a decentralized network—is losing the battle for capital allocation to the AI revolution. And that battle is not ideological; it’s computational. Both industries need electricity, cooling, and real estate. But AI pays better, offers more predictable contracts, and doesn’t require hodling a volatile asset to realize returns. We’ve seen this pattern before. In the early 2010s, GPU miners for Bitcoin were quickly replaced by ASICs. Those GPUs didn’t die; they found new homes in AI research and gaming. Now, the same thing is happening at the industrial scale. ASICs are specialized for SHA-256, but the real estate and power infrastructure they sit on is generic. And generic assets always flow to the highest-paying tenant. The contrarian view here is that this is actually healthy for Bitcoin. The argument goes: if mining becomes less profitable, the difficulty adjusts downward, making it cheaper for remaining miners. But that’s a dangerous oversimplification. The security of Bitcoin depends on a large, distributed, and economically motivated network of miners. If the most efficient miners start defecting to AI, the concentration of hashpower among remaining players could increase, or worse, the network’s total hashrate could decline, making it more vulnerable to attacks. We’re not there yet, but the trend is clear: the mining industry’s best talent and capital are being siphoned off. I’ve been in this space long enough to remember the ICO boom, where every project claimed to be “building the infrastructure for Web3.” Most of them failed. But hardware is even less forgiving. A failed smart contract costs you gas fees; a failed chip costs you millions in fabrication, lost deposits, and, in Block’s case, a damaged brand. Based on my observations of multiple mining hardware launches over the years, I can tell you that the gap between a tape-out and a profitable production chip is wider than most founders imagine. Block’s attempt was laudable, but their exit from this market is inevitable. What does this mean for the average crypto participant who holds Bitcoin or mining stocks? Three things. First, if you own shares of mining companies, check whether they have AI contracts. The market is beginning to price these transformations, and the pure-play miners are trading at a discount. Second, understand that the narrative of “digital gold” doesn’t pay for electricity. The security of Bitcoin depends on the profitability of mining, and that profitability is now competing with a global AI boom that shows no signs of slowing. Third, recognize that the hardware supply chain for mining is consolidating further. Bitmain and MicroBT will only grow stronger as new entrants like Block exit or shrink. That’s not a healthy sign for decentralization. We are witnessing a monumental shift. The same trucks that carried ASICs to Texas are now carrying racks of AMD MI300 GPUs. The same electricians wiring up immersion cooling for miners are now wiring up liquid loops for HPC clusters. The same mining CIOs who optimized hashprice are now optimizing return on capital for AI workloads. This isn’t a fad; it’s a structural reallocation of resources. I’ll leave you with a question: when the next bear market hit and the hashprice collapsed, where will the miners go? If they have AI contracts, they’ll stay alive. If they don’t, they’ll sell their assets—and those assets will likely be bought by AI companies. The very machines that secure Bitcoin may one day be repurposed to train large language models. The poetry of that transition is not lost on me. It’s a reminder that technology is a series of shifting abstractions, each layer cannibalizing the one below. Block’s $41.9 million lesson is a microcosm of a macro reality: in the long run, capital finds its most efficient expression. For now, that expression is not Bitcoin mining hardware. It’s AI infrastructure. And the sooner we accept that, the better we can navigate the future that’s already being built—one data center at a time.