Core Scientific Lines Up $3.3B Junk-Bond Raise to Swap Hashrate for AI Compute

Core Scientific is leaning hard into the AI buildout, planning $3.3 billion in speculative-grade notes due 2031 to convert a Bitcoin-mining footprint into high-density data centers. The Austin-based firm, valued around $6.55 billion, said part of the proceeds will refinance existing debt, while terms such as the coupon and offering date remain undisclosed.

The financing push follows up to $1 billion previously secured from Morgan Stanley and arrives alongside a 12-year agreement with CoreWeave that could generate roughly $10 billion in revenue. To meet that demand, Core Scientific is constructing six additional facilities as the AI scramble lifts power-dense colocation across the U.S. Investors have noticed: the stock climbed 4.5% to $20.77 on Tuesday and is up 42% year to date.

The company has been clear it will monetize Bitcoin holdings to fund the shift. Management indicated it will continue selling BTC; CFO Jim Nygaard recently estimated Core Scientific holds fewer than 1,000 Bitcoin after selling 1,900 BTC for $175 million in January. The firm ended 2025 with 2,537 BTC on its balance sheet, an amount worth about $192 million at recent prices near $75,800. Even so, Bitcoin remains the better earner today: in the fourth quarter, Core Scientific generated $41.1 million from self-mining versus $31.3 million from colocation.

What matters most here is the capital-structure bet. Issuing $3.3 billion of junk bonds to finance long-lived AI infrastructure is effectively an attempt to swap volatile, short-cycle mining cash flows for contracted, duration-heavy colocation revenue. If CoreWeave’s workloads scale as expected and the six-site build proceeds on schedule, the company can transition from a commodity-exposed miner to a capacity landlord with visibility. That would justify the re-rating we’re seeing in the equity.

But the spread between vision and cash conversion is where this deal gets tested. Unknown coupon terms will dictate how much free cash Core Scientific actually captures from new leases. Execution risk is real: converting or building for AI requires high power density, advanced cooling, and low-latency networking—disciplines adjacent to mining but stricter in uptime and service-level expectations. Any delay in energization, GPU availability, or interconnect readiness elongates payback while interest accrues.

Counterparty concentration also matters. A 12-year, roughly $10 billion pipeline with a single anchor customer simplifies underwriting but concentrates risk. Should AI economics normalize—say, from improved GPU supply, better model efficiency, or pricing pressure on inference—the revenue ramp can slip versus today’s assumptions. In that scenario, leverage magnifies downside. Conversely, if rates roll over and the AI build remains capacity-constrained, 2031 notes could be refinanced on better terms, pulling forward equity value.

There’s a signaling layer too. Selling down BTC reserves to fund capex may unsettle purists, but it reduces equity dilution and aligns the balance sheet with the new operating model. It also highlights the current crossover moment: despite branding around “digital infrastructure,” Bitcoin still delivers higher quarterly revenue than colocation. Core Scientific’s peers—Hut 8, TeraWulf, Riot Platforms, MARA Holdings, and Bitfarms—are pursuing similar arbitrage, repurposing power, land, and interconnect rights toward AI.

Context matters: the company emerged from Chapter 11 in 2024 after once ranking among North America’s largest miners, with its fleet producing 13,762 BTC the prior year—a haul worth roughly $1 billion at today’s prices. That history cuts both ways. The operational muscle to scale quickly is there; so is the reminder that commodity cycles punish over-leverage.

CEO Adam Sullivan said the new capital will accelerate project service timelines. If those timelines hold and contracts activate as planned, the debt-for-duration trade can work. The market is giving them the benefit of the doubt—for now.