Waiting for Compute Because They Have To Not Because OpenAI Wants to

by

$50B Per Year SpaceX Deal is $75B More than Waiting for Data Centers in 1 Year. There is an AI gold rush which is why SpaceX and its customers will win.

OpenAI signed the Oracle $300 billion five year deal because that is the plan B they can afford. Waiting for Compute Because They Have To Not Because OpenAI Wants to. Are you a loser bean counter or are you going to win the Gold Rush.

AI Demand is higher than available supply. Do you get capped on AI usage ? Yes you did.

Delayed capacity generates zero revenue from that capacity while customers can establish relationships elsewhere.

Under those conditions, the right strategy is to rent available capacity aggressively and build future capacity in parallel.

The early revenue also buys more than cash:

Customers integrate your models into their workflows.
Real usage improves products, distribution and operating experience.
Delivered growth strengthens the financing and IPO story.
Financing and retained cash support the next expansion.

In 1997, Amazon Bezos explicitly emphasized the window to establish customer relationships and prioritized long-term market leadership over short-term profitability in his 1997 shareholder letter.

Everyone is talking about Anthropic ARR. WOW it is so HUGE. Annual Recurring revenue. Without SpaceX lease Anthropic makes $30B less on ARR. IF 1 GW was available and usable in Q4 that is $100B more ARR. Oracle deal is ARR late in 2027.

How much business do you surrender while waiting for cheaper compute?

Does Oracle 4.8 GW AI data centers exist now? No. Open AI has a contract for $300 billion over about five years, starting 2027, not 4.8 GW of buildings with chips you can walk into in September 2026.

What exists today

Abilene / Stargate I: first halls live since late 2025. The 1.2 GW campus is still filling. Astra was trained on a large Blackwell slice there. That is hundreds of megawatts, not 4.8 GW. Maybe 5-10% which is already under contract. The long term contract was used to finance construction

New Mexico Project Jupiter (~2.45 GW) construction is underway, ~26% complete in September, customer delivery aimed at first half of 2027, permits still contested. Nvidia chips and micron likely NOT completely purchased 80%+ of the costs not committed.

Michigan “The Barn” (~1 GW) is groundbreaking 2026, not a finished factory.

Oracle uses the OpenAI paper to get the campus and the CHIPS financed. OpenAI’s five-year commitment, plus ~$75 billion of prepaid or customer-supplied GPUs inside Oracle’s backlog, is what lets bond buyers fund the build. If OpenAI will not sign five years, the 4.8 GW does not get poured. They are counterparties on the same project finance. Several years ago the Stargate plan was bigger then OpenAI flaked and it became 10% of the size. Now OpenAI is back..maybe. See what gets built.