CoreWeave's $2.58 Billion Quarter Proves AI Is Devouring Crypto's Money
What gets me about the latest CoreWeave earnings report isn't just the numbers. It's what those numbers signal about where capital is flowing when the AI infrastructure buildout hits velocity.
Second-quarter revenue came in at $2.58 billion, more than doubling year-over-year. The company raised its full-year guidance to between $12.4 billion and $13.2 billion. Third-quarter sales are forecast at $3.45 billion to $3.6 billion Source. The market rewarded the print with a 16% surge in pre-market trading, lifting IREN and Cipher Digital along for the ride.
But here is the thing most people miss. CoreWeave ended the quarter with $104 billion in contracted business and added another $25 billion in customer commitments after the period closed. Management noted recently signed deals carried margins five to ten percentage points above recent levels, reflecting scarce capacity and favorable pricing.
I genuinely don't know how to feel about a company sitting on a $104 billion backlog while simultaneously reporting a $626 million net loss. The paradox is the point. AI infrastructure is expensive. It eats capital like nobody else in tech history. Yet investors are piling into it anyway.
The broader market is revealing its preference plainly. Bitcoin trades below $64,000. Crypto ETF inflows cannot offset selling pressure. Meanwhile, neocloud stocks are popping. Investors favor computing infrastructure over decentralized assets right now. The divergence is stark and persistent.
What struck me was reading Morgan Stanley's earlier projection that 2027 AI capex would hit $1.4 trillion. Goldman Sachs estimated US data center power demand climbing from 41GW to 66GW next year, creating a 49GW gap by 2028. Power is now the binding constraint, not chips. CoreWeave's results confirm the thesis in real time.
The implications extend beyond stock tickers. A $104 billion backlog means hyperscalers, AI labs, and enterprises are locking in GPU access years ahead. Capacity scarcity drives pricing power. That pricing power funds further expansion. The cycle compounds.
I keep coming back to one question. When does the debt service hit? CoreWeave's net loss suggests leverage is funding growth. The question is whether contracted revenue converts to free cash flow fast enough to sustain the buildout. Five to ten percentage points of margin expansion on new deals helps. So does the $25 billion in post-quarter commitments. But the burn rate matters.
The AI infrastructure supercycle feels different from previous booms. This time, the demand side includes sovereign AI players and enterprises that previously could not afford GPU clusters. Supply is constrained by power availability, not just silicon. CoreWeave's print shows both dynamics playing out simultaneously.
Market participants who positioned early are collecting. Those timing the cycle from the outside should watch the backlog-to-revenue ratio and margin trajectory. They tell you whether the infrastructure boom is still accelerating or hitting diminishing returns.
One hundred four billion dollars in contracted business. That number is both impressive and unsettling. It proves demand is real. It also suggests we are still early in a capital-intensive phase that will test whether projected returns materialize.
For now, the money is flowing toward compute. Crypto is getting left behind. Whether that rotation persists depends on whether CoreWeave and its peers can convert contracts into cash flow before the capex treadmill swallows them whole.
I'll be watching the next quarterly print for exactly that signal.
Related
More from the blog
Anthropic's Building Its Own Chips. The Nvidia Dependency Problem Just Got Real.
Anthropic's hiring a custom silicon team, joining the growing list of AI companies building their own chips to reduce Nvidia dependence and optimize for their models.
Inertia Enterprises' Fusion Fuel Breakthrough: Weeks to Minutes
Inertia Enterprises cut fusion fuel pellet production time from weeks to minutes, solving a critical capex bottleneck for commercial fusion
Orbital Data Centers and the E-Waste Problem in Low Earth Orbit
Orbital data center plans ignore the looming e‑waste crisis in low Earth orbit, where defunct hardware becomes permanent hazardous debris.
Kog Bets the GPU Arms Race Is a Software Problem
French startup Kog claims 30x faster LLM inference through pure software optimization on standard GPUs, challenging the assumption that better hardware is the only path forward.