SpaceX locks in the missing layer. The most ambitious slide in the S-1 gets an application-layer proof point.

The deal. Signal 005 flagged the Cursor option; this week it became binding — US$60 billion all-stock, close targeted for the third quarter. All-stock keeps cash on the balance sheet for the build-out; it also means public shareholders absorb the dilution and the integration pitch becomes part of the believe-what-we-believe agenda.

Why now. The IPO roadshow still has to sell a segment climb the filing does not yet support on organic revenue alone. Starlink is the profitable engine; AI/xAI is the loss-making slice — US$3.2 billion in 2025 revenue, roughly US$6.4 billion operating loss, mostly disclosed compute leasing rather than application product. The gap is not only dollars; it is product shape.

What Cursor buys the stack. Cursor brings what SpaceX could not assemble on organic run-rate: enterprise coding product — the cleanest ROI field in the stack — plus a growing history of how large organisations actually write, review, and ship code, and customer relationships deep enough to land and expand. Inside one owner, compute, model, and application become a full-stack offer with consistency and operational history no integrated-stack rival can match on a shorter runway. Attached compute inside the stack is the hardware half; Cursor is the enterprise half. As data regulation moves into the buyer checklist, a single-vendor path for enterprise coding — audit trail, residency, one contract — starts to look like a feature, not overhead. None of that is in the filing yet.

What to watch. After close, the binding question is feed rate — whether attached compute compounds capability inside the stack. Segment reporting and Cursor usage become the visible proof points. Watch for deeper enterprise adoption, Composer economics, and GTM built to land and expand. The integration test is now on the public clock.

What's growing. Signal 002's capex magnitude met Signal 005's credit repricing; this week Nvidia's bond debut extended the pattern to the supplier layer. The chart maps where YoY pressure is rising.

Why. Capex is climbing on two tracks — more scope and a higher price per unit. Chips, power, and queue time all cost more; datacentre footprint is expanding anyway. Management frames the combination as demand-backed — customer-signed forward orders backing the spend. Credit markets ask how much is inflation on a build already in the plan, and what happens if those contracts are not large enough to cover it. Oracle was the live split last week: backlog and revenue held, capex stepped up, free cash flow went sharply negative — operations strong enough to print, cash weak enough to reprice.

What happens next. Debt, equity, guarantees, and inter-company spend all feed the same build-out, and the chart shows every layer accelerating. Cash goes out today; depreciation and debt service hit earnings around 2028. The bet is end-demand revenue — customer spend at the application layer, not circular hyperscaler flows — growing fast enough to pay for what was borrowed.

What to watch. Near term: the next CoreWeave–Oracle print — whether cash flow, not operations, drives the reprice. Through 2026–27: capex guidance, bond pace against the ~US$570 billion market sketch, and Nvidia's close. Into ~2028: depreciation hitting the P&L. The tell across all three clocks: does customer-layer revenue show up in segment reporting, or does inter-company spend keep funding the loop?

The Federal Energy Regulatory Commission (FERC) unanimously ordered grid operators to fast-track data-centre connections — data centres pay upgrade costs and accept limiting during grid stress. Interconnect queues have been the bind; a unanimous federal order on those terms is what that bind looks like when it stops being a forecast — the strongest read short of the grid actually failing. Power generation remains the industry bind — now with a federal can of Red Bull.

Other watch items. Water now runs paired with power in the public read — a UN-backed projection that 2030 data-centre water use could match the basic needs of 1.3 billion people; community acceptance and load limits bind as hard as queue time. Nuclear supply hardened in the same window: Oklo–Centrus uranium, Oklo building for Meta in Ohio, Amazon on Talen. Where the grid is still too slow, capital routes around it — Verse on off-grid battery and solar, gas plants fast-tracked off-grid, UK regulators weighing mandatory limiting at peak.

Two very different things are playing out at once — and most coverage collapses them into one. One is the narrative war: the fallen-out-of-love break with the US government tracks back through the autonomous-weapons line Anthropic refused — celebrated in much of the world, read as defiance in Washington. The other is frontier capability risk itself — perceived, but arguably externally validated when Project Glasswing put Fable and Mythos to work on live vulnerability fixes before Commerce switched the models off. Coverage merges them at its peril — neither thread is overstated. Distinct challenges, each demanding its own read.

The week after Signal 005's Commerce order, both threads accelerated. Cybersecurity executives pushed for relief; the trust breakdown with the White House became the narrative spine. Business press called it an AI Strait of Hormuz — a US chokepoint on frontier models that could block global access the way Hormuz blocks oil. AFR's Chanticleer column argued Anthropic's own safety warnings had boomeranged. Late in the week, the fight went global: Amodei's G7 appeal, allied pushback on unilateral US action, congressional demands on the legal basis, export rules even insiders struggle to parse. Washington acted on national-security grounds — the models reportedly surpass human experts at cybersecurity tasks, the strongest read yet on autonomous frontier capability. The restriction cuts both ways on the China gap: capability concentrates in US-controlled hands rather than diffusing globally, even as allies push for coordination.

The capital-markets mirror landed the same weekend. Anthropic's IPO path ran into the same political reality the export action created — disclosure and geopolitical risk are now part of the governance story, not footnotes.

China's self-reliance turn accelerated on two fronts: silicon and capital markets. ByteDance is in talks for 50,000 Iluvatar CoreX chips — a third domestic supplier. Zhipu's GLM-5.2 open-source release sent the stock up 48%.

Shanghai relaxed IPO rules to let unprofitable AI labs list on the mainland — DeepSeek, Zhipu, and MiniMax now have a domestic path. The price war among ByteDance, Tencent, DeepSeek, and Xiaomi deepens: capability and commoditisation moving together. The mirror to the export-control story above is clean — as the US restricts frontier models, China builds its own stack and its own funding rails. The gap narrows on cost and benchmarks even as export controls concentrate US capability at the frontier.

This week, the reckoning. Metered AI pricing is not new; the invoice landed this week. Enterprise finance is catching up — budgets blown out, caps imposed, named companies forced to show return under the new maths. Finance is still one stage behind the vendors: bundled seats are gone; seat plus metered API is what blew budgets now; next comes model tiers, tool calls, and connector traffic on top of the seat. Cost per completed task is the number finance will insist on, not list token price.

Two ROI patterns. RBC, Cisco, and Uber deployed hard on metered stacks — agentic coding, API-heavy workflows — without caps. Usage surged; mid-year invoices blew annual budgets; finance pushed back. Productivity may be real; on a metered stack ROI is usage × price, and both move fast. UnitedHealth is the counter-pattern: Bloomberg reports US$3 billion over two years to automate scheduling and claims — bots and workflow, not a token-invoice reset. Company states 2:1 return (Bloomberg; company figure, not verified here): ROI in cost cut and cycle time.

Where smart budgets go next. Caps and routing are table stakes. Survivors invest in internal structure — clean context, workflow prep, semantic layers — so fewer tokens buy a completed task. Margin defence is context discipline and auditable ROI tracks, not waiting for the next pricing letter. The borrow is uniform across the build-out; the payback is not, yet — at this layer, it has to prove out in pockets.