Ten spokes. The bets that decide whether it all works.

Oct 26: easyJet said it will offer Wi-Fi for the first time, putting Starlink on its A320, A320neo and A321neo jets from the first half of 2027, finishing by early 2028; easyJet flies more than 350 aircraft and carried 100 million passengers in 2025. Pricing for passengers and the deal terms were not disclosed (easyJet). Its CEO had said in January 2026 that the economics did not justify Wi-Fi (Airliner World). Oct 27: a Falcon 9 launched 27 more Starlinks from California, with the booster landing on its 26th flight (SpaceX). Oct 25: Serhii Beskrestnov, a defense technology adviser to President Zelensky, disputed Musk's claim, writing that it is "PHYSICALLY IMPOSSIBLE" for Ukraine to use commercial Starlink for long-range strikes because it does not work over Russia and drones using it have been capped at 70 to 90 km/h for months (CNBC). Oct 24: Aer Lingus finished free Starlink Wi-Fi on all 12 A330s and 14 A321LR/XLR long-haul jets, ahead of its end-2026 target; short-haul A320s are next, with no timeline (One Mile at a Time). Oct 24: Musk posted that "Ukraine military violated Starlink terms of service and used it for offensive drone strikes to kill people. That loophole was closed last week," and said Starshield is controlled by the US government; President Trump confirmed he had spoken to Musk (CNBC). Oct 20: Bolivia began "Aula Conectada," a plan for Starlink in about 4,500 to 5,000 mostly rural public schools, with the government paying the first 24 months and towns paying after that (Bolivian government). Oct 24: Ukraine's Defense Ministry said Starshield, the military version of Starlink, held a stable connection above 400 km/h that morning, after weeks capped near 161 km/h; President Zelensky had said on Oct 23 that the limits were still in effect, and Senator Richard Blumenthal opened an inquiry the same day. Who set the cap is still not public. Oct 22: Ukraine signed a letter of intent with Nokia and ICEYE to explore its own government-run satellite network (not an order, ICEYE says; first satellite planned for 2028). Oct 23: a Falcon 9 launched 27 more Starlinks from California, the first Starlink launch in 11 days; Falcon 9 no longer flies Starlink missions from Florida, and the bigger V3 satellites need Starship (Space.com). Oct 22: Trump said Musk "put it back" after the Ukraine slowdown and declined to say whether they spoke; Ukraine said speeds were not fully restored and some restrictions remain (Reuters, BBC). Q3 (reported Oct 20): 13.6M consumer subscribers, up 1.6M in the quarter and ahead of the 12.19M analysts expected (Reuters), plus hundreds of thousands of enterprise subscribers. Connectivity revenue $5.25B, up 69% from a year earlier, with $2.37B of operating income, up 99%. Consumer revenue per subscriber rose 5% from Q2, to $69 a month from $66. Watch: ARPU trend. Consumer ARPU of $69 a month is down about 17% from $83 a year earlier (SpaceX Q3 results). Funding: Q4 capex is guided to $35B-$40B; CFO Bret Johnsen said SpaceX is funding the build-out from operating cash flow and debt while keeping an investment-grade balance sheet (Q3 call, Oct 20). Starlink is contracted on 12,000 aircraft and installed on more than 3,800 (Shotwell). Starlink is no longer SpaceX's largest segment by revenue (the AI segment brought in $7.56B), but it is still by far its biggest profit. The profit engine: the only profitable SpaceX segment in Q2, and still the biggest profit in Q3 ($2.37B of operating income vs $431M for the AI segment). AI compute rentals (Google pays about $920M a month from Oct 2026) pushed the AI segment past it on revenue in Q3, so watch margins. Q2 (reported Aug 4): 12.0M subscribers, double a year earlier. Connectivity revenue $4.3B, up 66%, with $1.7B operating income, the only SpaceX segment in profit. Average revenue per subscriber fell from $85 to $66 a month. Oct 18: on NBC's Meet the Press, Zelensky said he believes the US government was behind the Ukraine slowdown and that Starshield-linked vehicles and drones are still speed-restricted; Treasury Secretary Scott Bessent called the charge "very reckless" and said Trump and Rubio have "categorically" denied it. Oct 17: Zelensky said Starlink inside Ukraine still cuts out above 161 km/h while the same terminals kept working at 180 km/h in tests on EU roads, calling it "a planned operation"; Secretary of State Marco Rubio said "nobody ordered that Starlink speed to be slowed down. It sort of happened," and that service is "back to where it was before." Ukraine's chargé d'affaires in Washington told CBS that Kyiv "cannot confirm" a full restoration. Government pushback is now a pattern, not a series of one-offs: Canada's Prime Minister Mark Carney told the New York Times "Starlink, we will diversify away from, no question" (MobileSyrup, Sep 25); Poland is planning its own network (Oct 17, below); and the EU wants to keep two of the three 2 GHz phone-satellite blocks for government and European players, a fight Washington now threatens to escalate (see Bet 02). Oct 17: Poland's deputy defence minister Cezary Tomczyk said Poland has first offers and secured funding for a "Polish or European Starlink," military first, with a concrete initiative expected in the coming months. Oct 15: Southwest said every jet in its 800-plus Boeing 737 fleet will have Starlink by fall 2027 (300-plus by the end of 2026), free for Rapid Rewards members, which it says would make it the first major US airline with a fleetwide rollout. Oct 16: Starlink said it has built gateways and other infrastructure in India "with the requisite government" approvals and is ready to connect the country; Musk replied "Much appreciated" to Reliance's statement welcoming competition. Clearance still has not been granted. Oct 15-16: Reuters reported, citing two US officials and a Western official, that the order to slow StarShield service for Ukraine's military came from inside the White House; a senior US official said the president made no such request. Nine senators from both parties asked the Defense Department inspector general to investigate (Oct 16). A Ukrainian Defence Ministry source told Reuters terminals work in motion again, but only up to 100 mph (160 km/h); Musk said StarShield limits are set by the US military and ultimately the president. Oct 16: Reliance said it welcomes Starlink competition in India and that Jio's own satellite service also awaits clearance; the government says all three licensees are at broadly the same stage. Oct 15: Bloomberg reported the EU is seeking to fast-track antitrust approval for Project Bromo, an Airbus-Thales-Leonardo satellite venture meant to cut reliance on SpaceX; no filing yet. Oct 14: Starlink went live in Curaçao after the Oct 7 concession. Oct 14: Bharti's Sunil Mittal said OneWeb is still "significantly ahead" of Starlink in India's approval queue. Oct 12: Falcon 9 lofted 27 Starlinks from Vandenberg after a three-week gap; constellation above 11,100 active satellites. Airlines: Southwest (whole fleet by fall 2027), American (1,030+ jets from early 2027), KLM and Royal Air Maroc. India remains licensed but waiting on security clearance. Delta picked Amazon Leo for 500 planes from 2028.

Good sign

Q3 subscriber adds above Q2's 1.7M, and India clearing security review.

Bad sign

Growth only from price cuts, a formal India block, or more big airlines choosing Amazon.

Oct 26: Starlink said Starlink Mobile is now live across Brazil with carrier Vivo, letting 50M Vivo customers use apps and messaging in cellular dead zones, with 7M phones already connected; it called this the first nationwide service to cover an entire country. No pricing or revenue was given. Oct 26: on Verizon's Q3 call, CEO Dan Schulman said SpaceX would need 65,000 to 80,000 cell sites and $300B to $500B to compete in US cities and suburbs, taking "a minimum of four to six years" to build and costing $300B to $500B "over the next seven or eight years," plus more time for stores and customer service; he said he assumes "everything SpaceX says they will do, they will do," and that Verizon sees "absolutely no benefit whatsoever" in letting a satellite provider resell its network. Musk replied on X the same day that it "will take 2 or 3 years." So the two sides are about two to four years apart on timing, and Brazil shows a third path: working through a carrier like Vivo instead of building cell sites first. Oct 29: the FCC votes on a proposal (the first step, which opens public comments) to open another 482 MHz of airwaves for satellite-to-phone service, raising the eligible bands from five to 13. CTIA, the trade group for AT&T, T-Mobile and Verizon, asked the FCC to drop the 2.5 GHz and 3.45 GHz bands, 290 MHz of the total, citing interference with 5G (PCMag, Oct 23). Reported Oct 22: Bloomberg, via Reuters, said SpaceX is in talks with tower companies including Crown Castle for access to at least 80,000 US cell sites for a nationwide network; Crown Castle's CEO said it is talking to "all parties" without confirming a SpaceX deal. SpaceX has not commented, and terms and cost are unknown. Dated targets (Shotwell, Q3 call, Oct 20): FCC approval for 15,000 Mobile V2 satellites (granted Oct 6); an agreement to acquire 800 MHz low-band spectrum (signed Oct 8, closing expected in the first half of 2027); US service in early 2028 (management target). Oct 20 (Q3 call): Starlink Mobile passed 22 million monthly users, up 37% in the quarter, after launches with carriers in Uganda, the Democratic Republic of Congo, Kazakhstan and Costa Rica. Shotwell said the near-term priority is flying the Mobile V2 constellation through 2027 and using the newly bought spectrum to turn on service in early 2028; the 800 MHz license deal, signed Oct 8, is expected to close in the first half of 2027 pending regulators. Still unknown: pricing, and whether SpaceX sells service directly or through carriers. Oct 19: Deutsche Telekom, Orange, Telefónica and Vodafone signed a letter of intent to explore a joint bid for Europe's mobile-satellite spectrum, the 2 GHz band used for direct-to-device service. They would build the ground network and run coordination and monitoring in Europe, with satellites from a partner still to be selected, and said the bid does not replace existing satellite partnerships. The current EU rights expire in May 2027 (European Commission). Oct 18: US Ambassador to the EU Andrew Puzder told the Financial Times that European companies "should be wary" of retaliation that could be "very devastating" if Brussels reserves most of the 2 GHz satellite-to-phone band for European operators. The European Commission's May 27 proposal splits the band into three 10 MHz blocks for 20 years when current licenses expire in 2027 (extendable by up to two years): one reserved for secure government communications including the EU's IRIS² constellation, one for European startups, and only the third open to non-European operators such as SpaceX and Amazon (Politico); the European Parliament and EU governments still have to agree. Starlink Mobile's direct-to-cell service is built around that band. Oct 16: Reuters reported that TD Cowen estimates building a US mobile network would cost more than $80B, on top of Starship, AI and Terafab spending; MoffettNathanson's Julie Zhu said "it's just hard to imagine that this is all going to get done within a decade." SpaceX's regulatory filing lays out a hybrid plan: satellites talking to phones plus antennas on towers and rooftops. T-Mobile's CTO said the spectrum would not match established networks. Oct 12 (reported Oct 15): Deutsche Telekom CEO Timotheus Höttges said SpaceX has to build the equivalent of T-Mobile's 85,000 US antennas first and warned carriers against wholesale deals that help it; DT still plans Starlink phone service in Europe in 2028. Oct 8: SpaceX agreed to buy Grain Management's nationwide 800 MHz spectrum (about $8B cash, per WSJ) and said it wants Starlink Mobile to be a major US carrier. Oct 6: the FCC approved up to 15,000 next-gen direct-to-cell satellites. The $17B EchoStar spectrum is approved, but commercial use waits for final closing, targeted Nov 30, 2027. Today US phone service runs only through T-Mobile. Abroad, partner launches keep coming: Banglalink in Bangladesh (Oct 7, first in South Asia), Beeline in Kazakhstan and CNT in Ecuador (late September). AT&T, Verizon and T-Mobile just formed a joint venture to answer it.

Good sign

FCC approval of the Grain deal and a first own-brand plan.

Bad sign

Spectrum sitting unused until 2028, or carriers refusing wholesale deals.

Oct 26 (Anthropic IPO filing): Anthropic's S-1 says that between May and October 2026 it signed and amended several deals with SpaceXAI, SpaceX's AI subsidiary, for compute on Nvidia GPUs. It agreed to spend up to $120.5B over terms running into 2029, about $3.1B a month in total once all capacity is delivered, which Anthropic expects in December 2026, with different fees while capacity ramps up. The first non-cancelable periods end in January 2027 and March 2027 for the different deals; after that, either side can cancel each deal on 90 days' notice. What it means: the two-way exit only starts after that early-2027 lock-in, and the deal is now about 2.5 times the $1.25B a month first reported. Oct 26: SpaceX filed the details in an 8-K (a report of a major event to the SEC). On Oct 24 it signed a binding commitment for a master finance lease with an Apollo-managed company, which will buy GPU pods (racks of AI chips, including Nvidia's next-generation Vera Rubin) and lease them to SpaceX. Principal payments should total about $40B; each delivery gets its own lease of about five to five and a quarter years, with rent paid quarterly. The fixed interest rate is set at pricing, before signing, so it is still unknown. Closing is expected before year-end and leases start in the first half of 2027 as chips arrive. Apollo and Goldman Sachs lead, with BlackRock, Goldman Sachs Asset Management and KKR-advised accounts as anchors. CFO Bret Johnsen said SpaceX sees "less than one year payback periods" on its Super Intelligence investments; that is the company's claim, not an audited figure. So it is a lease, not a classic loan, and it is a commitment that can still fall through. Oct 25: Musk posted that "SpaceX closed $40B of debt for GPUs, but we're investment grade and have Starlink & launch," calling it "primarily callable debt" (debt SpaceX may repay early) for chips arriving in Q1 and Q2 2027 and "not a major factor this quarter." He also said debt markets have almost shut off money for companies without revenue outside AI data centers, and that "SpaceX is not buying any neoclouds" (companies that only rent AI computing). Reuters said it was not clear what "closed" meant or whether pricing was agreed; the interest rate is not public. Reported Oct 23: Bloomberg said the $40B chip financing hit hurdles as credit markets wobbled, and pricing talks reopened. The debt would sit in a special purpose vehicle (a separate company that holds the chips and the loan) to keep it off SpaceX's books and protect its investment-grade rating; Goldman Sachs leads talks with Apollo-led investors, and Pimco is weighing joining. Earlier talk put the riskiest $25B-$30B slice at a yield in the mid-8% range. Oct 22 (10-Q (quarterly SEC report)): one unnamed Super Intelligence customer was 34.5% of SpaceX's $14.58B of Q3 revenue (24.2% for the nine months), and another customer across all segments was 16.6%. Backlog (orders signed but not yet delivered) was $62.96B at Sep 30. SpaceX held $51.3B of cash and $38.6B of marketable securities against $38.0B of total debt; the $20B bridge loan (a short-term loan, repaid once longer-term money comes in) was repaid in full in June. The 10-Q now names the segment Super Intelligence (SI), formerly AI. Reported Oct 22: Semafor said SpaceX's $40B financing is expected to price next week at about 8.3% for its lowest-rated part and about 7.75% for the chip-backed part, led by Apollo and Goldman Sachs, above the June notes' 5.35% and up; final pricing can change. Q3 (reported Oct 20): the AI segment (Super Intelligence) became SpaceX's largest by revenue at $7.56B, up from $2.56B in Q2, mostly from renting compute under cloud deals, plus $608M from Cursor after the Aug 14 acquisition. It turned an operating profit of $431M after $1.2B of Cursor-related stock pay (Q2: a $1.26B loss), with $5.0B of adjusted EBITDA. Nameplate compute reached 1.7 GW at Sep 30; CFO Bret Johnsen said nearly 2.2 GW by the end of October and 2.5 to 3 GW by year-end. The cost: AI capex was $16.4B of the $19.6B total in Q3, and Q4 capex is guided to $35B-$40B. Backlog is $63B, about $42B of it due within a year; Q1 2027 revenue guidance is $33B-$37B. Musk said SpaceX expects to face a long compute shortage. Memphis (Oct 20): the council's one-year data-center moratorium failed 6-6; a substitute ordinance passed that CBS describes as a 135-day halt on new data-center construction and WSMV describes as new data-center rules (closed-loop cooling, limits on local water use). One profitable quarter moves this bet to Scaling; it is not yet a trend. Oct 18: Musk wrote that a meme about renting compute "describes the conversation SpaceX has had when we have tried to rent inference compute from AWS, GCP or anyone," so it could shift load off its own training clusters. Scarce capacity supports rental prices, but financing is tightening: CoreWeave bonds sold in June at a 9.675% yield traded near 15% in the week of Oct 12, which 9fin's Max Frumes called "the canary in the coal mine" for AI credit (Reuters, Oct 16). SpaceX's $20B bridge loan, once due September 2027, was repaid in full in June from the $25B notes (10-Q, Oct 22). Contracts: CFO Bret Johnsen said Sep 11 that a fourth hosting deal, about $1.11B a month from an unnamed AI customer, starts billing Dec 1, taking contracted compute income to $41.1B a year alongside Anthropic, Google and Reflection AI (about $150M a month) (DCD, Sep 11). The biggest customer so far is Anthropic. Under a May 3 agreement it pays $1.25B a month through May 2029 for Colossus 1 and 2, at a reduced fee while capacity ramped in May and June. After the first three months, either side can exit on 90 days' notice (SpaceX prospectus, reported May 20). Update Oct 26: Anthropic's S-1 now puts all its SpaceX deals at up to $120.5B, about $3.1B a month from December, with no cancelling before January or March 2027 (see above). After that the exit works both ways: if compute prices rise, SpaceX can reprice the capacity or take it back for other customers. Neither the prospectus nor the Q3 10-Q supports an earlier claim of a 180-day term. Q2 compute-rental revenue was $1.6B, all from Anthropic (Deutsche Bank, Aug 10), so Q3 is the first full quarter at the full rate. On the Q2 call, CFO Bret Johnsen said SpaceX had contracted another $6.7B of cloud revenue over six months, ramping from October. Q3 consensus (TipRanks, Oct 12): AI segment revenue about $6.4B, above connectivity at about $4.8B. Google pays SpaceX $920M a month for about 110,000 GPUs from October 2026 through June 2029 (SpaceX SEC filing, signed Jun 5). That is about $2.8B a quarter from Google alone, starting this month, from the Colossus data centers. Oct 16: SpaceX confirmed it reports Q3 after market close on Oct 20, with an audio webcast at 3:30 PM CT. The Google rental starts in October, so it lands in Q4 numbers, not Q3. Oct 16: Evercore ISI's preview expects about 2.3 GW of compute by end-2026 (above consensus), an annualized revenue run rate near $104B exiting 2026 driven mostly by compute deals, and new compute deal announcements around the report; it models 2027 pricing near $30 per watt and 2027 capital spending near $151B. Oct 15: Nvidia CEO Jensen Huang said on the BG2 podcast that Nvidia now counts SpaceX alongside its cloud-provider customers because it trains, deploys and rents compute; to a clip saying Musk wants 10 GW of compute by end-2027, Musk replied "Maybe more than 10GW." Oct 13–14: Morgan Stanley's Adam Jonas called SpaceX's Distributed Inference Cloud the most ignored growth story, tying terrestrial Colossus rentals to a longer orbital/edge compute path. To buy more chips, SpaceX is reportedly in early talks to borrow about $40B for Nvidia chips (FT and Bloomberg, Oct 7); SpaceX has not confirmed and the talks could end without a deal. The open question is profit. In Q2, Starlink made $4.3B revenue and $1.7B operating income. The AI segment made $2.56B revenue and lost money. Starmind is the possible orbital version of this business later. Still unknown: is SpaceX a landlord or a capacity-short tenant? It rents Colossus capacity to Anthropic (about $3.1B a month from December, per Anthropic's S-1) and Google's rental starts this month, while Musk says (Oct 18) it cannot rent inference compute from AWS, GCP "or anyone" to take load off its training clusters. Do the rentals squeeze Grok?

Good sign

Q4 shows rental revenue near Starlink's, with positive margins, and more outside customers.

Bad sign

Chip spending and debt outrun rental income.

Oct 26-27: Ship 42, the upper stage for Flight 15, moved to the Massey's test site on Oct 26 (NASASpaceflight, via The Space Tap), and tracker Next Spaceflight reports it completed a single-engine static fire on Oct 27 that demonstrated a deorbit burn; SpaceX had not confirmed it when we checked. Ship 43, for Flight 16, finished its cold-fluid testing and went back to the production site. Booster 22 returned to Mega Bay 1 after its Oct 25 test. Still to come: the rest of Ship 42's testing and the FAA license. Oct 25: SpaceX finished a "full duration 33-engine static fire" of the Super Heavy booster for Flight 15 (Booster 22, Pad 2, ignition about 8:22 AM CT per Spaceflight News); after the last two campaigns, launch followed in about two to four weeks. Ship 42's static fire and the FAA license are still to come. Oct 24: Booster 22, the first stage for Flight 15, rolled to Launch Pad 2 (NASASpaceflight), and the City of Starbase closed the beach and Highway 4 on Oct 25 from 6 AM to 6 PM CT (backup Oct 26), likely for a static fire, where the engines light while the booster is held down. Ship 42 still needs its own static fire before Flight 15 (no earlier than Nov 8). Oct 23 (NASA Artemis update): the Starship lander will now meet Orion in Earth orbit and then push both to an orbit around the Moon, which NASA says will significantly lower the propellant Starship must gather in space. NASA showed the first photos of Cabin 1, the crew cabin SpaceX is building, and SpaceX added two cabin demos to its contract. A SpaceX milestone chart shown with the update lists a ship-to-ship propellant transfer demo in Q1 2027 and an uncrewed Moon landing late in 2027 (TeslaNorth). Artemis III (2027) is a docking test in Earth orbit with a test Starship launched from Florida; the crew will not go inside, and NASA will pick the first 2028 landing lander by readiness. Oct 22: Politico reported that SpaceX declined to bid on three National Security Space Launch missions, two for the Space Development Agency, and no other company bid; a Space Force official said the service is "working through" how to support them, and SpaceX did not comment. Starship is built for flat-stacked satellites, while most military satellites need a larger bay. The 10-Q shows 33 Falcon launches in Q3 (21 internal) vs 43 a year earlier, and 2 Starship launches. Oct 22: NASA's Aerospace Safety Advisory Panel said it was told about 22 launches would be needed to fuel the Artemis IV lander and will recommend NASA re-examine the mission; Musk said the number "will be more like 6 to 8." NASA said Oct 20 that SpaceX plans an in-space cryogenic transfer demo in early 2027 and an uncrewed Moon landing test in late 2027. Oct 20 (Q3 call): Musk said Flight 15 is expected next month and will try to catch both the booster and the ship at the tower, calling success "more likely than not" but not guaranteed, and SpaceX said it expects the FAA license to cover both catches. SpaceX's plan: refly a ship by early 2027, attempt on-orbit refueling early next year, fly weekly at some point in the first half of 2027 and twice weekly or more by the end of 2027. Flight 14 was SpaceX's first revenue-generating Starship mission, adding 26 terabits per second of Starlink V3 capacity. A Louisiana spaceport is planned, with a first launch targeted for 2029. These are targets, not results. Oct 18: NASASpaceflight reported that SpaceX's latest schedule puts Flight 15 no earlier than Nov 8, and Musk has confirmed SpaceX will try to catch both the Super Heavy booster and the Ship. Ship 42 had its engines removed and went back for another cryo test before its static fire, possibly tied to the Raptor Vacuum software glitch Musk described after Flight 14; Booster 22 awaits its static fire. Flight 16 (likely Booster 23 and Ship 43) is aimed at December and could be the first Starship launch from Florida's LC-39A, which would make five Starship launches in 2026. FAA public scoping meetings on the Starbase expansion run Nov 17–19. Oct 16: Texas Monthly reported, citing two sources, that the Texas Parks and Wildlife Commission had been discussing selling the 1,300-acre Boca Chica State Park, which surrounds the Starbase launch site, with SpaceX as the prospective buyer; a draft agenda had it for a closed session on Nov 5. After Texas Monthly asked, the agency said the commission "is not considering the sale" at its November meeting; sources said it could come up later. Oct 14: the FAA published a Notice of Intent to prepare a Programmatic EIS for Starbase Starship operations, evaluating up to 456 launches and 912 landings a year versus today's 25/50 authorization; comments due Nov 30 — review, not approval. Sep 28 (Flight 14): first orbit, all 26 Starlink V3 satellites released, controlled splashdown despite an early engine shutdown. Oct 10: Musk said Flight 15 is about 4 to 6 weeks out and will try to catch the ship with the tower. The FAA planning schedule lists Flight 15 no earlier than Nov 8. Ship 42 for Flight 15 went back for more cryogenic testing, and Ship 40, recovered after Flight 13, is back in Texas for inspection (Oct 7-9). Morgan Stanley's Adam Jonas says Flight 16 is targeted before year-end (note shared Oct 11). Every other SpaceX bet, from V3 Starlink to Starmind, waits on this.

Good sign

A ship catch on Flight 15, then a reflight.

Bad sign

Long pauses after every problem.

Oct 26: another rival appeared. Cowboy Space (formerly Aetherflux, founded by Robinhood co-founder Baiju Bhatt) unveiled Big Iron, an 86-meter reusable stainless-steel rocket whose upper stage stays in orbit as a megawatt-class data center carrying Nvidia Vera Rubin GPU racks. First flight is planned for 2028; it is a plan, with no hardware flown (SpaceNews, Axios). It shows more money chasing orbital compute, and that rivals also think they need their own big, cheap rocket to do it. Oct 20 (Q3 call): SpaceX said it will begin flying Starmind satellites in 2027, and the call was hosted from the Starlink factory in Bastrop, Texas, described as the future Starmind production site. No satellite count, launch date or customer was given, so the status stays Behind. Oct 15: Musk wrote that the Star Mind computer SpaceX is co-engineering with Nvidia "is far better than almost anyone expects and will reach volume production next year," replying to Jensen Huang, who said the two teams are building a new server architecture "that might be super good in space and on Earth." First Starmind satellites are targeted for Q4 2027 (Musk, August); volume production of the computer is not the same as satellites in orbit. Oct 13–14: Morgan Stanley's Adam Jonas called SpaceX's Distributed Inference Cloud — terrestrial data centers plus orbital/edge nodes — the growth story investors are largely ignoring; Starmind is the orbital piece. SpaceX's Starmind design: up to 250 kW peak per satellite, with production of thousands as soon as late 2027. A target, not hardware in orbit. Google is already flying: on Oct 1 its Project Suncatcher prototype put four TPUs in orbit on a SpaceX rideshare, built with Planet, and Google confirmed it is operating as expected. Two more Google satellites with laser links come in 2027. The prototype runs its chips about 15 minutes at a time before cooling, and Google's research chief compared the timeline to 15 years of research (Fortune, Oct 10). Blue Origin has filed for a 51,600-satellite network, Project Sunrise. Google's own paper says orbit beats the ground only once launch costs fall under $200/kg, around the mid-2030s. That cost is what Starship is meant to deliver.

Good sign

Google publishing good chip data (a validator), or booking Starship rides (a customer).

Bad sign

Google proving the idea first and keeping the hardware in-house.

Oct 25: Musk posted that Grok Bot "can now call someone or a business for you, make a request on your behalf, take notes on the response and send you a note summarizing the conversation"; the official changelog did not list it yet on Oct 26, and no user or revenue numbers came with it. Oct 23: Grok Bot v0.75.5 added a Vault that saves logins, payment cards and contact details and fills them in to sign in or check out. Oct 24: Musk wrote that "Agentic demand is growing faster than I've ever seen anything grow"; no new user or revenue numbers. Oct 22: Grok Bot can now use Calendar, Reminders, Contacts and other phone apps on iOS and Android. SpaceXSI launched its Super Intelligence Partner Network with more than 130 companies in early access (including Nvidia, Apollo, Snowflake and Databricks), and Grok for Startups offers credits across Grok, Grok Build, Grok Bot and the API; no revenue figures yet. Oct 22 (10-Q): Virginia's attorney general sent SpaceX, SpaceXAI and X Corp. a notice of violation on Oct 19, saying it is contemplating a suit over allegations that they misrepresented and omitted Grok's propensity to generate nonconsensual explicit images. Oct 20 (Q3 call): SpaceX said Grok Bot reached 800,000 paid weekly users earlier this month, more than double the month before, and Musk said enterprise Grok Bot should be in wide deployment within a few weeks. X advertising revenue was $538M in Q3, up 19% from a year earlier (Q2: $367M), after a new ad platform launched in Q2. Customers paying more than $1M a year in run-rate rose 50% in the quarter on a pro forma basis. Still unknown: X subscriber numbers and Grok subscription revenue on its own. Oct 19: AWS made Grok 4.7 available on Amazon Bedrock in AWS GovCloud (US), Amazon's cloud for government workloads. That evening Musk said X switched on a new AI ranking model that predicts time spent and a retrieval model that surfaces posts people read all the way through, and that user active seconds are "hitting all-time highs almost every day"; X has not published the figures. Reported Oct 18: Musk wrote "X is seeing all-time high usage," quoting a fan account that cited Similarweb data of 13.43B website visits in Q3, up 6% from 12.67B in Q2. Similarweb has not published that quarterly figure itself, website visits leave out app use, and X has not released new user, subscriber or ad numbers. Oct 16: Grok Bot's account said computer tasks now finish 3x sooner on average and other replies arrive in less than half the time; Musk said SpaceX is putting "a lot of effort" into a business Hagglebot he thinks could improve enterprise cost-efficiency by 5% or more. Oct 16: Musk said all X ads will face "a reasonable esthetic standard" and advertisers whose ads are "ugly, shrill, strident or overstate claims" will be barred; no criteria or start date yet. Oct 15: Nvidia's Jensen Huang said Grok Bot consumption "is really skyrocketing," and Musk said he has never seen anything grow as fast, with no free tier. Amazon is blocking Grok Bot orders as an "unauthorized AI agent"; Musk told users to shop elsewhere, floated an Amazon competitor, and said SpaceX will make users whole for Grok Bot purchase mistakes (a post on X, not a change to published terms). Oct 14: SpaceX Super Intelligence launched Grok Voice 3.0 — #1 on Artificial Analysis's Speech-to-Speech Index at 86.9. Starlink uses it for ~20,000 weekday support/sales calls (~70% resolved without a human; ~1 in 4 English sales calls convert). Tesla cars run millions of Grok Voice conversations a day; API price stays $0.08/min. Oct 12: The X Subscription bundles Grok, Grok Bot, Cursor and X Premium (Free through Ultra at $200/mo). Conversion and churn numbers are not public yet. Earlier: the AI division's X account became SpaceX Super Intelligence (@SpaceXSI) on Oct 10. X ads were $1.8B in 2025, below Twitter's roughly $4B in 2021, and $367M in Q2 2026. X and Grok had 6.3M paying subscribers (May). Oct 10-11: Tesla switched Grok Bot on in every Cybercab and Model Y robotaxi. X Money is live for US Premium subscribers, reported in 41 states (since Jul 27).

Good sign

Paid subscribers and ad revenue both rising after the unified plan, with clear conversion from legacy stacks.

Bad sign

Flat ads, with subscriptions and X Money too small to matter, or users sticking on old plans.

Oct 23: Rep. Michael McCaul said after meeting Grimes County Judge Joe Fauth that the first phase of the planned Terafab plant there is expected to bring more than $16B of investment and about 3,000 jobs, in a planned 100-million-square-foot building; these are his figures, not a company filing. Oct 22 (10-Qs): SpaceX's board formed a special committee (independent directors who review a deal so insiders don't decide alone) to evaluate a potential combination with Tesla, and Tesla's board formed one of disinterested directors to explore deepening the relationship, which could include a business combination. Neither has made a determination; a deal would need SpaceX's committee, board and controlling shareholder, and Tesla's committee recommendation and any shareholder approval. Who pays for Terafab is part of that open relationship. Oct 21 (Tesla Q3 call): Musk repeated the targets, commissioning machinery at the Austin research fab in the first half of 2027, a first test chip around May or June, and "some kind of useful chip at volume" around Q3 2028. He said Terafab will need "tens of billions" in equipment through early 2028 and probably more than $100B long term, and that the split of who pays, Tesla or SpaceX, is still being worked out. CFO Vaibhav Taneja said most of the first equipment orders for the research fab are placed. Still nothing produced. Oct 20 (Q3 call): Musk gave SpaceX's internal chip timeline, with what he called big error bars: a first functional test chip (a simple memory chip to test the line) around May or June 2027 at the research fab in Austin, and volume production around Q3 2028 at the main Terafab site at Gibbons Creek, Texas. He called the dates "very aggressive" and thanked ASML and other suppliers for speeding up equipment deliveries. Nothing produced yet. Oct 19: Tesla's proxy statement lists expanding into semiconductor fabrication, "beginning with our Austin Fab," and the chair's letter says chip investments including the Terafab initiative will help secure supply of critical components. No timeline or output was given. Oct 7: Musk said Tesla and SpaceX will build and run Terafab themselves, with no operating role for TSMC. Intel remains the named process partner. Oct 10: Musk said logic, memory, packaging and masks will sit in one building at Giga Texas, aiming for about 10 chip designs a year. Oct 15: TSMC CEO C.C. Wei said TSMC is talking with Terafab about using its capacity, but "the fab got to be run by TSMC"; there is no conclusion yet. That differs from Musk's Oct 7 plan for Tesla and SpaceX to run it themselves. Nothing produced yet.

Good sign

A signed partner deal and a first chip out of the building.

Bad sign

Years of plans with no output.

Oct 26: Tesla's Robotaxi account said a "bigger service area" is "now live" in Houston. Trackers put it at about 254 square miles, roughly 10 times the April launch zone, now reaching downtown, the Heights, Montrose and Bellaire, but not the airports (Houston Public Media, Not a Tesla App). The fleet is small for that area: Robotaxi Tracker, a crowd-sourced count, lists about 34 driverless Model Ys in Houston and no Cybercabs yet (Electrek). Oct 27: Musk posted "A Storm of Cybercabs" over a photo of Cybercabs parked in Dallas, where Cybercab service is due in November. Oct 26: Tesla Europe said its supervised driving software (called Tesla Assisted Driving in Europe) is "Now approved in Latvia," the 10th EU country, after Latvia's CSDD recognized the Dutch approval; rollout "begins soon" (BB.LV). Oct 25: Cybercabs carried US Grand Prix fans to Circuit of the Americas; reported: a fan tracker logged a record 1,250 Austin Cybercab sightings on Oct 24, 1,159 with no one in the driver's seat (EV). Tesla has not shared race ride numbers, and the Texas list is unchanged at 970. Oct 25: Texas DMV lists now show 970 vehicles under Tesla Robotaxi LLC, 463 Cybercabs and 507 Model Ys, up from 927 on Oct 23 (Robotaxi Pulse); still a registration list, not cars in service. Oct 22: CHP said a customer Tesla its driver said was in FSD (Supervised) struck and killed a 71-year-old pedestrian on I-80 in Vallejo, Calif., at about 70 mph; it was not a Robotaxi, and CHP is investigating (CBS San Francisco, Los Angeles Times). Reported Oct 22-23: rider app screenshots show Austin Robotaxi taking requests until 1 AM (from midnight), with late rides limited to nearby drop-offs, and the service area now reaching Circuit of the Americas for the US Grand Prix; Tesla has not announced either change. Oct 23: Texas DMV records list 927 vehicles under Tesla's Robotaxi business, 420 Cybercabs and 507 Model Ys, up from 803 on Oct 16 (Cybercab Collective); that is a registration list, not cars in service. Oct 21: Tesla's first Robotaxi Safety Report says the fleet drove 3.1 million miles with no onboard safety monitor from Jan 22 to Oct 18 in Austin, Dallas, Houston, Orlando, Tampa and Miami, with zero injuries, airbag deployments or pedestrian and cyclist strikes by Tesla's definitions, and lists Las Vegas, Jacksonville, Phoenix and San Antonio as coming soon (no dates). The figure is above the nearly 2 million unsupervised miles in the Q3 update for Sep 30; Tesla has not reconciled the two. Oct 21 (Q3 update and call): Tesla reported nearly 2 million cumulative unsupervised Robotaxi miles at Sep 30, up eightfold in the quarter, and Musk said the Austin Cybercab fleet has grown 10x since its Sep 3 launch. AI software head Ashok Elluswamy said Cybercab launches in Dallas next month, service in Austin runs 6 AM to midnight and will "pretty soon" run to 2 AM, 24/7 operation is targeted by year-end, and customer-owned cars could start joining the Robotaxi fleet around Q2 2027. Tesla says it now runs Robotaxi in six cities. Active FSD subscriptions reached 1.74M, and over 75% of North American deliveries signed up for paid FSD. CFO Taneja said Tesla is not yet at its $0.30-a-mile cost goal; Musk said Robotaxi revenue should become noticeable around Q2 2027. Tesla still does not publish ride counts, and its sworn Cybercab answers to NHTSA are due Oct 30. Oct 19: Tesla's proxy statement says FSD (Supervised) has passed 15 billion cumulative miles, and the annual report lists "Vision-only Robotaxi" launched in the US; neither gives ride counts or a Dallas date. Before the Oct 21 call, two federal reviews are open: NHTSA petition review DP26010, opened Oct 9, into claims that about 1,000 Model Y and Cybercab robotaxis exceed posted limits in active school zones (four incidents, no crashes or injuries; a petition review, not a defect finding), and Tesla's Cybercab safety-standards answers, due Oct 30. Robotaxi is moving, but supervised FSD drew new safety scrutiny. Oct 16: the California Highway Patrol said a 22-year-old driver told officers he was asleep with FSD (Supervised) engaged when his Tesla hit a parked patrol car on I-80 in Fairfield late on Oct 13, seriously injuring two officers, who were released the next day; NHTSA said it is investigating. CHP stressed that FSD is a Level 2 system that needs constant supervision. Tesla did not comment. Oct 16: Slovakia's transport ministry recognized the Dutch approval of FSD (Supervised), the ninth EU country by TeslaNorth's count; Tesla Europe said rollout begins soon, with no date. Reported Oct 16: Austin riders got an app alert that Robotaxi rides now run until midnight, up from 11 PM; Tesla has not announced it. Oct 15: Electrek, reading NHTSA crash filings, counted five Robotaxi crashes where a remote operator was driving (two new, from August in Dallas and Austin), all low-speed and property-damage-only; Tesla has 36 Robotaxi incident reports on file and does not disclose how often remote operators take over. Oct 14–15: Texas DMV records show 349 Cybercabs registered (+30 overnight from 319), with Model Y robotaxis still at 420, for 769 autonomous vehicles on the Texas list (Texas AV Tracker / TxDMV). Tesla said Oct 10 that 300+ Cybercabs serve Austin (8x in 30 days) and AI software head Ashok Elluswamy wrote "Dallas soon." Oct 12: a production Cybercab from the US fleet went on display at the Paris Motor Show with Optimus and free FSD (Supervised) rides; no European Cybercab service is planned yet. Oct 12–13: FSD v14.3.12 began a wide release; notes say the model is now unified across Actually Smart Summon, FSD and Robotaxi. Oct 13 evening: Tesla said FSD Supervised v14 Lite is rolling out to Intel Model S and Model X. Europe: the supervised software, renamed Tesla Assisted Driving, faces an EU vote in December at the earliest. Ride counts and a Dallas date are still unpublished.

Good sign

Dallas Cybercab launch in November, 24/7 service, published ride numbers.

Bad sign

Safety monitors back in, or expansion stalls in Austin.

Oct 21 (Q3 update and call): Tesla said Optimus production began on its high-volume line in Fremont and these are its first production-quality bots, adding the process is "suitable for a commercial product." Musk said most of a 1 million-unit-a-year line is installed but reaching 20,000 a week will take "at least a year, maybe a little longer," deliveries to outside customers are likely around the end of 2027, and the hand is still the hardest part. He also targeted the AI5 chip in Optimus around April 2027 and a non-scripted demo video in a month or two. A larger Optimus factory is under construction in Austin. Status moves from Behind to Watch: production has started, but no output numbers or outside customers yet. Oct 19: Tesla's 2026 annual report, filed with the proxy, says Tesla "Readied V3 Optimus for mass-production," and the chair's letter says it is building "the first production lines for Optimus." No production date, unit target or output numbers were given, so the status stays Behind. Oct 18: Musk wrote "Optimus and Cybercab will be the vast majority of Tesla's value," adding that only super intelligence, robots, solar power and mass to space scale to incredible size. A statement of intent, not new production data. Oct 12: Optimus Gen 2 is on the Tesla stand at the Paris Motor Show in a glass case, next to a production Cybercab. Output reportedly rose to a few hundred a week, but the hand is still largely hand-assembled and Musk called it "shockingly difficult" (Oct 10). Reported Oct 14: All Things Elon says Optimus units are doing live Model Y line tasks at Giga Texas (sorting cells, moving parts); not confirmed in a Tesla IR release. Still not robots on paid shifts outside Tesla.

Good sign

Optimus doing real shifts, a hand built on a production line.

Bad sign

Optimus stuck in demos.

Oct 23: Tesla said Giga Berlin built 7,000 Model Ys in one week, a new high, on the way to a 7,500-a-week target set in June (roughly 364,000 a year if it holds). Oct 23: Tesla cancelled Cybertruck reservations in Australia and New Zealand, ending right-hand-drive plans, and is refunding deposits (Electrek). Oct 22 (10-Q): Tesla recognized $221M of Q3 revenue ($626M year to date) from SpaceX buying Megapacks, and SpaceX owed it $729M at Sep 30, including capacity reservations for undelivered Megapacks. Musk said Oct 22 Tesla is "increasing production of Megapacks as much as possible." Oct 21 (Q3 results): revenue $28.05B, flat from a year earlier and just under the $28.13B Bloomberg consensus; non-GAAP EPS $0.48 vs about $0.44 expected (Yahoo Finance). GAAP net income was $787M, down from $1.37B, and operating income fell to $718M as spending on AI, chips and new products rose. The cash test passed: free cash flow was $1.53B vs a consensus of about negative $4.9B, operating cash flow $6.26B, and cash and investments rose to $45.4B. Tesla cut 2026 capex guidance to about $20B from more than $25B and plans to fund it mainly from operating cash flow; Musk said capex rises next year and reaches "tens of billions" in 2028. Tesla has $30B of new credit lines and plans no draws this year. Vehicles: 486,532 deliveries (down 2% from a year earlier), the biggest order backlog in four years, and battery supply still the limit on output; auto gross margin excluding credits rose to 18.1% from 16.3%. Musk targets about 5 million vehicles a year by around 2030. Semi volume production began in Nevada (25,000-a-year line). Energy: 13.7 GWh deployed with gross margin up to 33.1% from 20.4%, helped by tariff refunds and a supplier warranty reimbursement, so some of that is one-time. Megafactory Texas started production (planned 50 GWh), the first Megapacks with Tesla's own Nevada LFP cells shipped, and Musk said Megapacks are "almost sold out" for 2027 on data-center demand. Earlier: Oct 15 the new Roadster was revealed (production in 2028; NHTSA said Oct 16 it will keep watching the thruster design). The annual meeting is Dec 3 at 3:00 PM CT.

Good sign

Free cash flow stays positive while capex climbs, and storage margins hold without one-time items.

Bad sign

Capex outruns cash flow and pulls in the $30B credit lines.

Achieved

A bet that hit its goal, with the date and source.

None yet.

Retired

A bet dropped or failed, with the reason.

None yet.