SpaceX Signs $6.3 Billion AI Compute Lease With Reflection AI as Texas Court Clears Starbase Path

A $150-million-per-month lease of NVIDIA GB300 capacity to Reflection AI, a unanimous Texas Supreme Court win over Boca Chica beach closures, and United's first wide-body Starlink transatlantic flight together reframe SpaceX as a multi-market infrastructure company while Tesla fights to control the narrative on a fatal Katy crash.

The most important thing happening across Musk's companies right now is not a rocket or a robot—it's a balance sheet quietly turning into an operating weapon. SpaceX just agreed to rent the AI compute it built for itself to an outside customer, won the legal certainty it needs to fly Starship on cadence, and pushed Starlink onto the routes where connectivity is hardest to fake. Each item looks separate. Together they describe a company that no longer fits inside a single industry box, funded by a cash pile large enough to bankroll several frontier markets at once.

Key Takeaways

• SpaceX signed an AI compute lease with Reflection AI worth about $150 million per month starting July 1, 2026, totaling roughly $6.3 billion if it runs through 2029.
• Either party can terminate the Reflection contract with 90 days' notice after the first three months, meaning headline value is not guaranteed revenue.
• The Texas Supreme Court ruled unanimously for SpaceX and state officials, dismissing the Save RGV suit over Boca Chica beach closures with prejudice.
• United Airlines flew its first wide-body transatlantic Starlink flight on a Boeing 777-200 from Newark to London Heathrow, with nearly 60 wide-body installs planned this year.
• SpaceX reported $100.8 billion in cash as of June 19, 2026, disclosed in an SEC filing tied to a proposed senior unsecured notes offering.
• Tesla's Ashok Elluswamy said a Model 3 driver in Katy, Texas overrode self-driving by pressing the accelerator to 100%, reaching 73 mph before a crash that killed a 76-year-old woman.
• Beyond Reflection, SpaceX has had compute deals with Anthropic, Google, and Cursor, pointing to a repeatable leasing business.
• Drone imagery of Giga Texas showed Tesla's planned Optimus factory rising with at least four floors, targeting substantive completion in late 2027.

---

The Real Story Is the Compute, Not the Customer

The Reflection AI deal gets read as a startup landing GPUs. That misses the point. The story is that SpaceX built Colossus-class infrastructure to train Grok, and it now has enough scarce capacity to rent to strangers at $150 million a month. When a company can lease its internal advantage to the open market, that capacity has become a balance-sheet asset, not just an expense line.

The market obsesses over model leaderboards. But the durable winners in AI are the ones sitting on hard-to-replicate data and infrastructure. GB300 chips at scale, wired into a company that already runs its own power, cooling, and integration muscle, is precisely the kind of moat that leaderboards can't measure. This doesn't make SpaceX an AI lab. It makes SpaceX impossible to price as a rocket company.

Why the Termination Clause Matters More Than the Headline

Six point three billion dollars is the number that travels. It is also the number to distrust. Either side can walk with 90 days' notice after the first three months, which means the guaranteed portion of this contract is a small fraction of the sticker. Treating the full figure as locked revenue is exactly the kind of mistake that gets corrected painfully later.

There's a second tension buried in outside leasing: every hour of GB300 time sold to Reflection is an hour not spent on SpaceX's own AI ambitions. The interesting disclosures ahead are how much capacity gets allocated externally, whether the lease survives past the early-exit window, and whether SpaceX signs more long-term commitments that convert optional revenue into structural revenue.

Boca Chica Becomes Boring, Which Is the Point

The Texas Supreme Court dismissed the environmental groups' suit with prejudice, holding that a 2009 constitutional amendment preserves public beach access but doesn't hand private groups the right to sue to enforce it. That's a narrow, procedural win—the court did not say beach access lacks protection, only that these plaintiffs couldn't enforce it through this vehicle. Different plaintiffs, new legislation, or fresh regulatory fights all remain live.

But narrow legal certainty is exactly what a cadence business needs. Starship is shifting from isolated test events toward a rhythm, and launch timing depends on local road and beach closure logistics as much as hardware and federal sign-off. Removing a recurring legal overhang at Starbase is unglamorous. Boring constraints are also the ones that quietly throttle how fast the hardware learns.

Wide-Body Starlink Is the Aviation Proving Ground

United's first wide-body transatlantic Starlink flight—a 777-200 from Newark to Heathrow—matters more than any regional-jet install because long-haul is where connectivity gets tested honestly. Oceanic routes break traditional systems, and passengers on an eight-hour flight notice bad Wi-Fi instantly. United says it will equip nearly 60 wide-body jets this year and its entire wide-body fleet by summer 2027, on top of 400-plus aircraft already carrying Starlink, heading toward roughly 1,000 by the end of 2026.

If the 777 experience holds up, the framing shifts from amenity to infrastructure. An airline can monetize reliability, and fleet-wide installs become easy to justify once a flagship carrier proves the demanding case. The execution risk is real—type approvals, downtime windows, and installation capacity all constrain rollout—but a visible reference customer on high-value routes is worth more than any spec sheet.

A $100 Billion Cash Balance Changes the Question

SpaceX disclosed $100.8 billion in cash as of June 19, 2026, in a filing tied to a proposed senior unsecured notes offering intended partly to repay a bridge loan. That figure has to be read carefully: it's a point-in-time liquidity number in a debt-raising context, and the filing itself warns the balance could look materially different by June 30. It says nothing about burn rate, profitability, or project-level allocation.

Still, the strategic implication is hard to ignore. The market instinctively separates Starship, Starlink, and AI infrastructure into distinct valuation buckets. A balance sheet at this scale lets one private company fund several enormous addressable markets simultaneously—which is precisely why SpaceX is so consistently mispriced. The skeptics arguing that burn rate matters more than cash on hand are right to push. The point is that the cash gives SpaceX the option to keep answering that question on its own terms.

Tesla's Problem Is Speed, Not Just Data

Tesla's account of the Katy crash is that the Model 3 driver manually overrode the driving system, flooring the accelerator to 100% and reaching 73 mph, with the pedal still pressed after impact. If confirmed by investigators, that materially changes a story that early coverage had pinned on autonomy software—and the distinction is central to Tesla's liability and reputation. Investigators could still clarify the system's status from the vehicle logs.

The deeper issue isn't the safety data; it's the gap between how fast a narrative forms and how slowly the technical record catches up. A fatality involving a 76-year-old woman inside her home generates instant framing. Tesla having the exculpatory data days later is not the same as Tesla shaping the story in real time. Until the facts arrive early enough to be impossible to ignore, the company keeps losing the first, loudest round.

The Smaller Signals Worth Tracking

Three quieter items round out the picture. Starlink now reportedly sits aboard the interim Boeing 747-8 serving as Air Force One, a marquee government installation in a mission-critical connectivity market. At Giga Texas, drone imagery shows the planned Optimus factory rising with at least four visible floors, on track for substantive completion in late 2027—the first physical evidence that construction is starting to match the robot-production rhetoric. And Tesla is positioning Vision as a safety layer beyond driver assistance, claiming its cameras can flag unavoidable collisions and support airbag deployment up to 70 milliseconds earlier, with impact sensors still making the final call.

The through-line across all of it: SpaceX is compounding advantages that don't map to any one industry, while Tesla's autonomy story keeps living or dying on whether it can anchor public debate in verifiable data before the narrative hardens. Watch the AI infrastructure economics, Starbase cadence, Starlink aviation execution, and Tesla's incident-data discipline—those four are where the next quarter gets decided.