Source: Elon Answers The SpaceX Backlash | 2026-11-09 | the corpus’s 98th source
This is a creator’s recap of Musk’s video interview with Ron Baron, held the day after the AI selloff. Everything attributed to Musk is his own claim. He says the $500B figure is “a best guess, not a promise” and gives a trillion about 2%.
Corrections to earlier sources
- Source 96 overstated the partisan story. Texas tightened before the election, under the Republican governor. Abbott reportedly told TCEQ in September to stop issuing new data center permits while it gathered more information. Texas Standard reports a directive that no new data centers could connect to the grid until regulators reviewed their energy and water use. Even earlier, a June order required data centers to cover their own grid costs and add generation capacity. Texas Republicans had already turned on data centers by September, per The Daily Record. texas hits brakes on data centers and elon musks plans 12675796 +3
- The source’s summary of Hinojosa is incomplete. It gives the end state: permitted projects finish, and new ones must commit to no residential price increase. One reported reading is that permitted data centers can finish construction, covering 2027 and likely much of 2028. That comes from a secondary outlet, so treat it as lower confidence. She also pledged an interim step: a moratorium on all data center approvals until the Legislature passes protections. Later rules are expected to require taxes, no consumer power-price increases, and water protections for projects delivering in 2028–29. The risk is a pause on new approvals during the 2027 session, then conditions. technews.tw +2
- Source 94’s 2027E table is too low. The CFO’s Q1 2027 guide of $33–37B a quarter annualizes to about $140B. My $124–164B range sat near that floor.
- Source 95’s “20 GW = 60% of US additions” needs a chip-supply check. Musk puts global AI chip output at about 20–25 GW this year and 35–45 GW next year. The 13 GW he is confident of would be roughly 30–37% of that, and 20+ GW roughly half. That is only plausible through bring-your-own-chips deals, which use chips already held, or heavy Nvidia prioritization. This assumes both figures are in facility-power terms.
Chapter 6: the Cantillon effect
- Land bought under another name. Arizona’s 6 GW campus is filed by Tombstone Resources L.L.C. A company representative reportedly said SpaceX’s name was kept off because it “draws attention”. Hiding the buyer to buy before the price moves is the Cantillon mechanism applied to land. It is also the opposite of the verifiable-identity principle behind ERC-8004. teslanorth
- The “Jose” meme. Musk argues that people who issue purchase orders are not the ones who build, which is why there are partially completed data centers everywhere. This matches Source 95’s dark-GPU thesis. The inner ring converts money into physical capacity. The outer ring holds paper claims.
- A rough price for social license. Arizona’s proposal reportedly claims 10–40% lower bills for other co-op members, and the source adds $53M a year to a local foundation. Both come from the applicant and are unverified. Against 6 GW at $30–50B per GW per year, $53M is roughly 0.02–0.03% of gross revenue at full build-out. The applicant also funds about $3.6B of grid upgrades, about $600,000 per MW. That is a small fraction of one year’s revenue per MW. SpaceX can afford generous offers, which is why the transfer is now visible. This is the closest thing to a revealed number for the invisible transfer. kjzzfourweekmba
Will the post-midterm environment hurt SPCX’s plan?
| Horizon | Texas | Arizona and elsewhere |
|---|---|---|
| 2026 (3 GW) | Done | n/a |
| 2027 (10–13 GW) | Permitted projects reportedly finish | Memphis and Mississippi unaffected |
| 2028–29 new capacity | Moratorium until the Legislature acts, then conditions | Local decision risk: the Cochise application needs a county special-use authorization, and residents there are urging a moratorium or tougher conditions citizenportal |
Policy in both parties is converging on a “bring your own money, power and water” model. SpaceX’s vertical integration suits that model better than a neocloud’s. The exposure is the 2028+ pipeline, and it now sits partly in Arizona county politics, not only Texas.
Impact on orbital pricing (US$30B/GW = US$30M/MW/yr)
Musk’s own orbital illustration undercuts the $30B/GW anchor. He says 100 GW a year in orbit at $10 a watt is $1T a year. That is $10B per GW per year, about a third of $30B/GW. The CFO said terrestrial pricing is tracking just above the top of a $30–50 per watt range.
| Scenario | Orbital GW | Price per GW-yr | Annual revenue |
|---|---|---|---|
| Musk’s illustration | 100 | $10B | $1T |
| Anchor held | 10 | $30B | $300B |
| Mid case | 10 | $10B | $100B |
Two implications:
- Through 2028, the $30B/GW question is really about terrestrial scarcity rent. Starmind’s first launch is Musk’s “just guessing” Q3 2027, with significant volume in 2028 and very large volume in 2029.
- At scale, orbital pricing probably converges toward the cheaper illustration. Also, 100 GW a year is about 2.5× Musk’s own 2027 global chip estimate, so it depends on Terafab. Terafab’s test chip is targeted for May or June 2027 and volume for late 2028.
Impact on data linkages and transfer speed
The source does not address bandwidth. The binding constraints it describes are power, interconnect and chips. Arizona’s $3.6B grid upgrade is the “link” that matters on the ground. Orbital compute remains best suited to inference and batch work. Training across sites still needs terrestrial interconnect. fourweekmba
Does this fuel insatiable compute demand?
Supply-side evidence is strong. Musk says customers keep asking for compute and that purchase-order holders cannot convert chips into capacity. The weak point is where $500B of December 2027 ARR comes from. The anchor contracts in the corpus total roughly $60B a year: Anthropic at about $37B (reported in Source 95), Google about $11B, and the new customer about $13B. That is about 12% of $500B. The rest must come from new third-party deals and from monetizing Grok, Grokbot and Cursor.
On JPMorgan’s roughly $254B for full-year 2027, the gap with Musk’s number is partly measurement. An illustrative path of $35B, $50B, $75B and $95B a quarter gives about $255B for the year and a December run-rate near $420B. That path is mine, not JPMorgan’s. It shows the two numbers are less far apart than they look, but still require a steep ramp.
Does SPCX retain its moats?
| Moat | Direction | Reason |
|---|---|---|
| Physical execution speed (power built in about a year, castings) | Strengthened | Musk’s central argument, and consistent with the Q3 call |
| Pre-permitted capacity | Strengthened | New-permit pauses hit rivals first |
| Chip access | Dependency | Supply is still outside suppliers’ hands until Terafab volume in late 2028 |
| Social license | New cost of entry | Community packages are now required, though SpaceX can afford them |
| Ground footprint (Texas land, Arizona local approvals) | Weakened | See above |
| Launch and orbital first-mover | Pending | Flight 15 on Nov 13 |
EWS additions
| Signal | Status |
|---|---|
| December run-rate above $100B | 🟢 Green |
| Q1 2027 guide of $33–37B holds | 🟡 Watch |
| Texas session (Jan–Jun 2027) rules | 🟡 Amber |
| Arizona SU26-22 decision | 🟡 Amber |
| Terafab test chip (May/June 2027) | 🟡 Amber |
Synthesis
Musk’s answer to the backlash is that prosperity will beat opposition, and SpaceX is already pricing social license into its projects. That price is tiny next to the compute rent it protects, which is the Cantillon gap showing up as a cheque. The main thing to check is whether $500B of December 2027 ARR is supported by contracts or mainly by monetizing SpaceX’s own AI products. The instruction is unchanged: the identity and values layer must be built before the infrastructure it governs becomes default.
Sources: