SpaceX is currently locking in $1.11 billion in monthly revenue from a single, undisclosed customer, a move that pushes the company’s annualized run rate for AI compute hosting to approximately $41 billion. This latest contract, confirmed by CFO Bret Johnsen at the Goldman Sachs Communacopia conference on September 10, 2026, places SpaceX firmly at the center of the compute landlord thesis – but with a critical blind spot at the core of its newest revenue stream.
During the event, Johnsen stated: “Earlier this month, we closed another hosting deal… that translates into about $1.11 billion a month starting December 1st of this year, which is another roughly $13 billion of ARR that starts December 1st of this year.”
This new agreement is the fourth pillar in a massive infrastructure portfolio. The current breakdown includes Anthropic, contributing roughly $1.25 billion per month at the Colossus Memphis facility; Google, which begins a $920 million monthly commitment in October; and Reflection AI, adding $150 million per month. Combined, these four deals represent a staggering $3.43 billion in monthly revenue, or roughly $41 billion in annualized run rate. This trajectory supports Elon Musk’s stated goal of reaching $100 billion in ARR by the end of 2026, a target he described by noting, “$100 billion ARR in December is not a question mark. That’s what we would achieve if we basically did nothing.” According to CNBC’s reporting on the Q2 earnings call, this projection assumes closure of the $60 billion Cursor acquisition.
The identity of the latest $13.3 billion-a-year customer remains a mystery, though industry observers are focused on three primary scenarios. The first is OpenAI, which is currently building the Jalapeno chip and maintains a complex relationship with SpaceX following the Cursor deal – including OpenAI’s August 28 notice to wind down its Cursor model-supply contract by November 12. The second is Microsoft, which previously evaluated a bid for Cursor but ultimately declined. The third possibility is a new entrant seeking to secure massive-scale agentic capacity. Each scenario carries distinct implications for the compute landlord thesis, ranging from further vertical integration to a broader commoditization of high-end silicon.
Despite the aggressive revenue growth, the operational reality remains volatile. SpaceX’s AI unit reported a $1.26 billion operating loss in Q2 2026 on $2.56 billion in revenue, following a $2.47 billion loss in Q1. The company faces significant structural risks. Starting in 2027, all hosting contracts will include 90-day termination clauses, allowing SpaceX to reclaim capacity for its own models – a flexibility that Musk has suggested may be necessary given the uncertainty surrounding the long-term duration of existing deals like the one with Anthropic. These risks are compounded by the impending wind-down of the OpenAI model-supply contract and a $354 million accrual for probable litigation losses related to Memphis pollution lawsuits.
The compute landlord thesis is evolving from a simple infrastructure play into a high-stakes game of capacity management and model-supply dependency. As SpaceX balances its $18.4 billion in quarterly capital expenditures – more than 80% of which is directed toward AI – against the need for consistent, long-term revenue, the market will be watching whether these short-term hosting deals can provide the stability required to sustain a $100 billion enterprise. The immediate focus for investors is whether the undisclosed customer represents a strategic partner or merely a temporary tenant in an increasingly crowded and expensive chip factory ecosystem.