OpenAI's $100B backup server bet

Most AI labs are scraping together compute just to keep their current models online. OpenAI is preparing to spend twelve figures on servers that might sit idle.

According to a report from The Information, OpenAI is planning an extra $100 billion in spending over the next five years specifically for backup server rentals. That comes on top of the $350 billion the company already mapped out for server leases through 2030.

All told, OpenAI is looking at an average annual server rental bill of around $85 billion. To put that in perspective, that single annual line item is roughly half of what Amazon, Microsoft, Google, and Oracle earned combined in 2024.

Insurance against stalled launches

The reason for stockpiling spare compute comes down to a very real bottleneck inside the company.

Speaking at a Goldman Sachs conference, OpenAI CFO Sarah Friar acknowledged that the company has repeatedly had to delay product rollouts or withhold new features simply because it ran out of available chips.

The reserve servers are designed as a buffer against sudden usage spikes and to keep future training runs on schedule without choking consumer-facing apps. Executives also view the spare capacity as monetizable, meaning it could handle unexpected enterprise demand or research workloads rather than gathering dust.

CEO Sam Altman has been telling colleagues that compute is the ultimate differentiator in AI, reportedly suggesting that future requirements could eventually outstrip the capacity of the entire U.S. power grid.

The $115 billion burn

Committing this much cash makes OpenAI's balance sheet look downright terrifying.

Internal projections show OpenAI could burn through roughly $115 billion in cash through 2029, with positive cash flow not expected until 2030. At the same time, the company expects revenue to skyrocket from $13 billion this year to around $200 billion in 2030, driven largely by continued ChatGPT growth.

The massive spending plan is already rippling across the hardware ecosystem, fueling demand for cloud providers, chipmakers, and server vendors like Dell, Super Micro Computer, and HP Enterprise.

Why it matters

OpenAI is treating computing capacity not as an operational cost, but as a strategic moat to starve out the competition.

For regular users and developers, this should eventually mean fewer capacity errors, fewer delayed features, and far more reliable uptime when major model updates drop. But paying off an $85 billion annual server bill will almost certainly force OpenAI to squeeze every dollar out of subscriptions, enterprise contracts, and premium compute tiers.

For the broader AI industry, this level of spending effectively closes the door on independent labs. When a single company spends half the combined earnings of Big Tech on rented iron, competing at the frontier is no longer about clever architecture; it is an industrial infrastructure war.

We think the move is tactically smart for locking down limited cloud supply, but financially precarious. Banking on revenue surging from $13 billion to $200 billion while burning $115 billion leaves zero room for error if scaling laws plateau or enterprise adoption slows down.