#Gate广场四月发帖挑战 The ATM at the End of Computing Power: From Ethereum Mines to Wall Street's ComputeFi Surge

In New Jersey in 2017, a few big-shot commodity traders found after work was too boring, so they cobbled together a bunch of graphics cards in a utility closet and started secretly mining Ethereum. Nine years later, the company they founded, CoreWeave, carries a staggering $30 billion in Wall Street debt, and Mark Zuckerberg has just smilingly handed them another $21 billion in computing power contracts, pushing Meta’s total contributions to a suffocating $35 billion.
Don’t be fooled by Silicon Valley’s lofty narratives about general artificial intelligence. Beneath the glamorous veneer of AI’s rapid rise, what’s really inside isn’t the awakening of silicon-based life, but a hardcore, violent credit game on Wall Street. When a mining farm that once could have gone bankrupt at any moment due to a crypto crash successfully reverse-engineers the underlying logic of modern finance, what we see is no longer just a simple technological shift but the complete formation of a new paradigm: compute financialization (ComputeFi).

Wall Street Ghosts in the Graphics Card Graveyard

Everyone thought that the 2018 crypto crash and Ethereum’s switch to proof-of-stake (PoS) would sweep these GPU-based miners into the trash heap of history. Logically, GPUs that lost their mining value should be sold off as scrap metal. But these commodity trader founders had an extraordinary sense of smell—they didn’t liquidate, but instead keenly perceived a client base more hungry than crypto gamblers: machine learning researchers who couldn’t get a spot with traditional cloud providers.
The subsequent story is practically a walking textbook for nouveau riche. In 2025, CoreWeave’s revenue hit $513 million, a 168% year-over-year surge, and even more terrifying, they hold over $66 billion in backlog contracts. You might think they’re doing cloud computing, but they’re actually in the highest-end commercial real estate. Their business model isn’t to build buildings first and then rent them out; instead, they secure long-term lease contracts with top tenants like Meta and Microsoft, then storm into Wall Street conference rooms, slam the contracts on the table, and demand funding.
Here, GPUs are no longer hardware but income-generating assets. CoreWeave is essentially a “compute factory,” with all capital expenditures backed by long-term commitments from Silicon Valley giants before even laying a single cable. This approach of using property rights contracts to print money at banks is far more sophisticated than those Web3 projects that tried to fleece investors with a few whitepapers.

The Top-Level Power Play of Printing Money with Property Rights Contracts

If you want to understand the current AI infrastructure war, don’t look at the code—look at the balance sheet. After securing Meta’s astronomical order, CoreWeave immediately announced a $4.25 billion financing plan. The structure of this funding is extremely intriguing: $3 billion in convertible preferred notes due in 2032, with interest rates pushed down to a mere 1.5% to 2%, bundled with $125 million in high-yield bonds offering a 10% return.
This is the ultimate form of ComputeFi. Traditional venture capital firms have no chance to compete with the compute giants; equity financing alone can’t satisfy Nvidia’s appetite. Currently, CoreWeave’s total debt has ballooned to around $30 billion, three times what it was last year. They even completed an $8.5 billion debt financing in March this year, secured by new contracts. Wall Street elites are willing to buy these packaged debts into various financial derivatives because the underlying assets are extremely clear and robust.
This isn’t an isolated case; the entire AI infrastructure financing environment is now dominated by such massive debt structures. Look at SoftBank’s $40 billion bridge loan to support OpenAI’s Stargate project. The capital requirements for compute have become so enormous that new financial instruments, previously nonexistent two years ago, had to be invented. While the crypto industry once aimed to disrupt finance through token issuance, today’s compute giants are turning compute power itself into the most solid collateral, transforming Wall Street into their personal ATM.

Silicon Valley Giants’ Compute Stockholm Syndrome

At this point, you might ask: Meta’s own capital expenditure forecast for 2026 has already skyrocketed to $115–135 billion, and they’re even spending billions in Texas building data centers. Why are they still so willing to keep pouring $14.2 billion and $21 billion into CoreWeave’s mega deals? Is Zuckerberg’s money just blown by the wind?
The brutal truth is that Silicon Valley giants have developed compute-related Stockholm syndrome caused by anxiety over their capacity. Meta’s core problem is that hundreds of millions of daily active users are frantically calling on Llama models, requiring ultra-low latency computing that their own data centers can’t handle in the short term. By funding CoreWeave, they’re not just buying compute; they’re securing priority deployment rights for Nvidia’s latest Vera Rubin platform, gaining redundancy through multi-node distributed architectures, and ensuring they don’t fall behind in this large-model arms race.
For CoreWeave, this $21 billion bailout is a perfect life extension. Previously, one client—Microsoft—accounted for 62% of their 2024 revenue, a highly risky concentration. After fully tying Meta to their “battle chariot,” no single client will account for more than 35% of revenue. Using Silicon Valley’s fear of missing out on the AI era, CoreWeave has perfectly hedged against default risk from high leverage debt—this reverse harvesting is nothing short of performance art.

The Twilight of Mining and the Bretton Woods of New Energy

While CoreWeave is wildly leveraging on Wall Street, Bitcoin miners still committed solely to mining are undergoing a painful genetic reorganization. Despite Bitcoin’s total network hash rate surpassing an astonishing 1 Zetahash per second, the economics of mining are increasingly unprofitable. Leading miners like Riot Platforms and MARA sold over 19,000 Bitcoin in just one week—this isn’t just cashing out but a desperate act of survival.
A new monster is competing for the world’s scarcest resource: cheap electricity. Recently, Anthropic signed a chilling compute contract, directly locking in several gigawatts of next-generation TPU capacity with Google and Broadcom starting in 2027, while their annual revenue run rate by the end of 2025 has soared to $30 billion. Cambridge University data shows global Bitcoin mining’s continuous power consumption ranges between 13 and 25 gigawatts, yet a single AI company can devour several gigawatts of capacity.
This isn’t just a contest of compute; it’s a blow to energy monetization pathways. The same gigawatt of electricity, used for Bitcoin mining, faces volatile coin prices and rising difficulty, while renting that capacity to AI companies for large models guarantees a steady, monthly cash flow that can serve as the underlying asset for financial derivatives. Core Scientific, Iris Energy, and Hut 8 have long seen through this game—they’re rapidly shedding their pure mining labels and transforming into “leasing companies” that provide both power and data center infrastructure. In this new Bretton Woods system where compute equals power, whoever can turn the cheapest electricity into the most efficient floating-point operations will hold the future’s monetary minting rights.