OpenAI Raises Infrastructure Spend to $750B as Losses Hit $38.5B
OpenAI raised its projected infrastructure spend to $750B through 2030, up from $600B, while posting a $38.5B loss on $13B in revenue. The commitment-to-revenue gap is 57:1. Its own CFO has raised concerns. Bitcoin miners are in the direct path.
The AI capex cycle just re-escalated. The company's losses tell the real story.
Key takeaways
- OpenAI raised its projected infrastructure spending to approximately $750 billion through 2030, up from $600 billion earlier this year, first reported by The Wall Street Journal on July 22, 2026.
- The company posted a net loss in 2025 on $13.07 billion in revenue. CFO Sarah Friar has privately flagged that OpenAI may not be able to honor future computing contracts if revenue growth stalls, leaving a 57:1 commitment-to-revenue gap.
- OpenAI's Georgia Power deal alone locks up 3.2 gigawatts of grid capacity between 2028 and 2032, a direct energy-cost pressure point for U.S. Bitcoin miners operating in the same grids.
OpenAI has raised its projected infrastructure spending to approximately $750 billion through 2030, up from $600 billion set earlier in 2026, first reported by The Wall Street Journal. The revision is driven by new deals with Oracle (roughly 6 gigawatts of data-center capacity) and AWS ($138 billion over eight years, including 2 gigawatts of Amazon Trainium chip capacity), and it arrives while the company is hemorrhaging cash at a rate no private revenue trajectory has yet justified.
The spending figure has followed a whiplash trajectory: a Stargate Project announcement of $500 billion in January 2025 with SoftBank, Oracle, and MGX; a Sam Altman boast of $1.4 trillion; a walk-back to $600 billion in February 2026; and now a re-escalation to $750 billion. The underlying financials make each revision harder to defend. OpenAI's 2025 results, reported by The Wall Street Journal, show a net loss of $38.5 billion on $13.07 billion in revenue, with $34 billion in total costs and expenses. CFO Sarah Friar has privately raised concerns, per The Wall Street Journal, that the company may not be able to honor future computing contracts if revenue growth does not keep pace.
The Build Continues Anyway
OpenAI announced Project Camellia, a $20 billion data center campus in Effingham County, Georgia, described as the company's first facility where it serves as principal designer and builder. Sachin Katti, OpenAI VP of Compute Strategy and GPT-Infra, said the company has contracted with utility Georgia Power to receive 3.2 gigawatts of power between 2028 and 2032. OpenAI hired Brent Mayo, who previously served as general manager of xAI's Colossus data center operation in Memphis, to oversee construction.
The company's compute footprint has scaled from roughly 200 megawatts in 2023 to 600 megawatts in 2024 to 1.9 gigawatts by end of 2025, according to reporting attributed to statements by CFO Friar. Greg Brockman testified in the OpenAI v. Musk lawsuit that the company projects $50 billion in computing expenditure for 2026 alone.
OpenAI's target revenue for 2030 is $280 billion or more.
What This Means for the Grid and for Bitcoin Miners
The energy competition is direct. The Georgia Power contract for a single campus represents 3.2 gigawatts locked into a long-term utility agreement. That capacity does not flex. Every gigawatt committed to a hyperscale AI facility on a long-dated power purchase agreement (PPA) compresses the available supply for other industrial users, including Bitcoin miners, and puts upward pressure on PPA pricing across ERCOT, PJM, and the Southeast.
The AI capex and power demand story has been building for two years; OpenAI's Georgia deal is the clearest single-site example of what grid lock-up looks like in practice.
Bitcoin miners competing for grid access in data-center-dense regions face a structurally repriced power market, not a temporary spike. The DOE's Texas grid loan and the broader AI eminent domain push are downstream effects of the same force: AI capex is reshaping energy infrastructure in ways that will outlast any individual company's balance sheet.
The sound-money framing is worth stating plainly. A $750 billion infrastructure commitment from a company losing $38.5 billion annually on $13 billion in revenue exists because credit is cheap and money is abundant. The downside, if it arrives, gets distributed across Oracle's balance sheet, AWS's eight-year contract exposure, utility ratepayers holding long-dated power agreements, and the broader credit markets financing the stack.
OpenAI's CFO flagging this risk privately is an acknowledgment that the math has decoupled from the business.
The falsifiable version: if OpenAI reaches $280 billion in annual revenue by 2030 with GAAP profitability in sight, and grid capacity scales without structural energy shortages or rate spikes, the capex commitment is defensible. The trigger to watch is OpenAI's quarterly revenue run-rate against its spending pace, and wholesale power prices in ERCOT and the Southeast. Neither trend currently supports the optimistic read.
What to Watch
The next pressure point is whether OpenAI's 2026 revenue trajectory closes any meaningful portion of the 57:1 commitment-to-revenue gap. Watch PPA pricing in Georgia and Texas through 2027 as the Camellia buildout moves from announcement to construction. If Friar's private concern becomes a public contract restructuring, Oracle and AWS absorb the first hit, and the ripple into grid investment timelines affects every industrial power user behind them, including miners building or expanding sovereign mining operations in those markets.
Sources
- OpenAI Stargate Project announcement
- First reported by The Wall Street Journal, July 22, 2026
Frequently Asked Questions
The effect runs through two channels: energy and capital. On energy, OpenAI's long-term utility contracts lock up grid capacity that would otherwise remain available for flexible industrial users. The Georgia Power deal alone is 3.2 GW for one campus.
As AI hyperscalers sign multi-year PPAs across ERCOT, PJM, and the Southeast, the floor price for power purchase agreements rises across those markets, compressing miner margins directly. On capital, the credit facilities financing this buildout are part of the same pools that could otherwise flow toward Bitcoin mining infrastructure. A credit crunch in the AI complex does not automatically redirect capital to miners, but it signals the broader fiat credit cycle is extended.
No, not at current scale. OpenAI posted $13.07 billion in revenue in 2025 against a reported net loss of $38.5 billion. Its projected infrastructure commitment through 2030 is $750 billion, a 57:1 ratio of commitment to last reported annual revenue.
The company's own CFO has privately raised concerns that it may not be able to honor future computing contracts if revenue growth does not keep pace. Reaching the $280 billion revenue target by 2030 would require roughly 21x growth in five years from a base that is already loss-making at scale.
Project Camellia is a $20 billion data center campus OpenAI announced in Effingham County, Georgia, its first facility where it acts as principal designer and builder. OpenAI contracted with Georgia Power for 3.2 gigawatts of electricity delivery between 2028 and 2032. To put that in context, 3.2 GW from a single utility for a single campus represents a substantial claim on regional grid capacity. The deal is a concrete example of how AI infrastructure commitments translate into long-term, inflexible grid load that affects power pricing and availability for every other industrial user in the region.
