Wall Street has spent the last six months questioning Mark Zuckerberg’s aggressive AI capital spending, with Meta Platforms (NASDAQ:META | META Price Prediction) raising its 2026 capex guidance to $125 to $145 billion after a Q1 in which revenue grew 33% year over year. While everyone debates whether Meta is overspending, a quieter filing from SpaceX just laid out a playbook that Zuckerberg should be studying tonight.
In its recently disclosed S-1, SpaceX revealed that in May 2026 it signed Cloud Services Agreements with Anthropic granting access to compute capacity across its COLOSSUS and COLOSSUS II supercomputers. The terms are striking. Anthropic has agreed to pay $1.25 billion per month through May 2029, with capacity ramping in May and June 2026 at a reduced fee, and either party may terminate on 90 days’ notice. SpaceX explicitly states that this structure monetizes unused compute capacity while still permitting reallocation of the capacity for its own internal initiatives if needed.
The AWS Precedent Meta Cannot Ignore
I have been holding Meta since December 2022, and I have read every transcript Zuckerberg has put out on infrastructure. The pattern is familiar. Amazon Web Services was built originally to handle Amazon’s own retail spikes. In 2006, Amazon (NASDAQ:AMZN) opened that excess capacity to outside developers. Two decades later, Andy Jassy is running AWS at a $150 billion annualized run rate, growing 28% year over year, with an AI-specific run rate over $15 billion just three years into the AI cycle.
Jassy said on the most recent call: “We have been through this cycle with the first big AWS growth wave, and we like the results. We expect to feel similarly about this next wave with much larger potential downstream revenue and free cash flow.” Hyperscale history rhymes here. Hyperscale infrastructure built for internal use tends to become a generational business when its owner decides to rent the excess.
Meta Has Every Ingredient Except the Decision
Look at what Meta is already building. Q1 2026 capex hit $19 billion, up 46.8% year over year. The 2026 guide nearly doubles 2025’s $72.22 billion actual spend. Zuckerberg told investors that Meta is “rolling out more than one gigawatt of our own custom silicon” developed with Broadcom alongside AMD and NVIDIA systems.
CFO Susan Li was even more explicit on scale: “These multiyear cloud deals and our infrastructure purchase agreements drove a $107 billion step up in our contractual commitments this quarter.” She also conceded the planning uncertainty: “If we end up not needing as much as we anticipate, we can choose to bring it online more slowly or reduce our spending in future years as we grow into the capacity that we are building now.”
Meta is building flexibility into its fleet by design. Meta is building flexibility into a fleet that will inevitably have periods of underutilization between training runs. SpaceX has now demonstrated exactly how to monetize those gaps: short-notice-terminable contracts with frontier labs that pay hyperscaler-grade economics for capacity you can reclaim whenever your own roadmap demands it.
The $100 Billion Optionality
Here is my framing, and I want to be clear this is a thesis on optionality. If Meta sold even a fraction of its excess capacity to one large frontier customer at terms similar to SpaceX’s Anthropic arrangement, that is a multi-billion-dollar annual revenue stream at near-pure incremental margin. Layer it onto a company already producing strong operating margins with a current market cap of roughly over a trillion dollars, and a re-rating in the neighborhood of $100 billion of additional market cap becomes a reasonable bull-case scenario.
Look at what Azure has done for Microsoft (NASDAQ:MSFT), where Commercial RPO reached $392 billion, up 51% year over year, with OpenAI alone committing an incremental $250 billion of Azure services.
The SpaceX filing tips its competitive intent helpfully, stating it intends to sell excess capacity to a limited number of third parties, potentially positioning it as a competitor to CoreWeave and Nebius as well as hyperscalers. The same filing names Meta directly as a foundational-model competitor. The race for compute customers has begun, and Meta is the lone top-tier player still on the sidelines.
The Risk Side, Honestly
Meta selling compute to Anthropic or any frontier lab would mean powering competitors. Zuckerberg has talked openly about needing to “fully optimize the stack” for Meta’s own agents. Year to date the stock is down 8%, and prediction markets are pricing only a 21.5% probability that the next Muse model variant ships by June 30. Investor patience is finite. But Meta’s family of apps still has 3.56 billion daily active people, and Reality Labs is still bleeding $4.03 billion a quarter. A high-margin compute revenue stream would help both stories.
If you believe Meta’s AI investment cycle will pay off in product, you should also want management to consider every adjacent monetization path the AWS history book opens up. SpaceX just publicly priced the option at $1.25 billion per month. The long memory of Wall Street says the operator who rents the excess capacity earliest wins the decade. Zuckerberg has the compute, the balance sheet, and the engineering culture to run that play. The only question left is whether he wants to.
Top Gaining Stocks
Top Losing Stocks