Below is Dolphin Research's summarized Trans of $SpaceX(SPCX.US) Q3 2026 earnings call.
I. Core takeaways
1. Outlook
a. Q1 2027 total revenue guided to $33–37bn. The range reflects higher visibility on compute contract pipeline over the next six months.
b. Total GPM in Q4 is expected to revert to roughly Q2 levels. Q3 benefited from one-offs and a favorable customer mix.
c. Cloud services ARR target of $100bn by year-end remains unchanged (annualized off Dec. revenue). The CEO said total company ARR could reach $500bn–$1tn by end-2027.
2. Capex, backlog and financing
a. Q3 capex was $19.6bn, mostly into SI compute, Starship infrastructure and satellite manufacturing. Given new capacity added this quarter, Q4 capex is expected to rise to $35–40bn.
b. Backlog was $63bn as of Sep 30. Of this, $17.5bn is in deferred revenue, and about $42bn is expected to be recognized within 12 months.
c. Maintain an investment-grade balance sheet. Investment remains demand-driven.
3. Key financials this quarter
a. Consolidated: revenue of $14.6bn (+161% YoY). Net profit was $4.4bn vs. a net loss of $1.1bn a year ago; Adj. EBITDA was $8.8bn (+306% YoY), outpacing revenue growth.
b. Space: revenue of $1.8bn (+84% QoQ, +6% YoY). The increase was driven by a favorable customer mix.
c. Connectivity: revenue of $5.2bn (+69% YoY, +22% QoQ). GP doubled, with GPM at 55%; about 300bps was one-off (mainly tariff refunds). Ex one-offs, Adj. EBITDA margin expanded by roughly 200bps YoY.
d. SI: revenue of $7.6bn, up $5.0bn QoQ and $6.7bn YoY. Adj. EBITDA was $5.0bn, up $3.8bn QoQ and $5.4bn YoY, mainly from high-margin cloud services agreements.
4. Accounting impacts from the Cursor acquisition
a. Cursor equity value was $60bn. Of this, $23bn is for retention of key employees and replacement of unvested equity (including executive re-vesting), to be recognized as SBC over roughly three years.
b. Q3 SI OP was $431mn. Excluding $1.2bn Cursor-related SBC, OP would have been $1.6bn.
II. Call details
2.1 Management highlights
1. Launch and Starship
a. 35 launches in Q3: 21 Falcon 9 Starlink missions, 2 Starship, and 12 major customer missions (including a Falcon Heavy launch of NASA's Roman Space Telescope). Six were national security missions.
b. Newly signed: three Dragon crewed missions (scheduled through 2028), an Artemis 3 Starlink communications contract, 18 national security missions, a Starship lunar payload contract, and three Starfall missions.
c. Flight 14 reached orbit for the first time and deployed V3 satellites. Roughly half-loaded, it delivered 26 Tbps of capacity, equivalent to 10 Falcon 9 launches.
d. Flight 15 is planned for next month, aiming to recover both booster and ship to the tower. FAA approval is expected to cover return and tower recovery; the CEO sees success as more likely than failure, which would mark the first fully reusable orbital launch system recovery.
e. Early 2027 will re-fly the second stage and attempt on-orbit propellant transfer. From 2027, launches of mobile V2 and Starmine satellites begin; the goal is weekly cadence in 1H 2027 and twice-weekly or more in 2H.
f. A Louisiana launch site has been announced, with a long-term plan for 20+ towers. The CEO said the company accounts for ~90% of global mass to orbit, China ~6%, and others ~4%.
2. Starlink broadband
a. Coverage now spans 174 markets, adding Vietnam and Uganda. Consumer users rose by 1.6mn in Q3 to 13.6mn, excluding hundreds of thousands of enterprise and aviation users.
b. Consumer ARPU rose 5% QoQ to $69/month, helped by mix. V5 terminals are ramping, inventory constraints have eased; retail channels are prioritized ahead of the holidays, and activations are expected to increase in Q4.
c. Enterprise and Gov. revenue rose 91% YoY and 18% QoQ, driven by Starshield and aviation. Management expects both Enterprise and Gov. to each surpass consumer connectivity revenue over time.
d. Aviation contracts cover ~12k aircraft, about one-quarter of the addressable fleet, including three of the four largest U.S. airlines. Aviation terminal capacity increased 4x this year and is set to double again by year-end.
e. V3 payload per Starship will ramp to more than 20x Falcon 9's current downlink capacity. The CEO expects constellation capacity to reach 100x today within ~3 years (about 100k satellites, per-satellite bandwidth >10x), and demand could reach 1,000x current levels in 3–5 years.
3. Starlink Mobile
a. MAUs exceeded 22mn, up 37% QoQ. In Q3, new carrier partnerships were added in Uganda, DRC, Kazakhstan and Costa Rica.
b. The FCC approved launching 15k mobile V2 satellites, and the company announced acquisition of low-band spectrum. Service is planned to start in early 2028, with the goal of becoming a major U.S. mobile operator.
4. SI (Super Intelligence)
a. Nominal compute was 1.7 GW at Q3-end, near 2.2 GW by end-Oct, and tracking 2.5–3 GW by year-end. About 200k GB300 units are coming online each month, with Vera Rubin the main platform in 2027.
b. The 2027 goal is to add 10 GW, reaching ≥13 GW by year-end. This will be paired with an additional 20 GW of power and cooling capacity (see Q&A).
c. By end-Q2 2027, a training cluster of 440k Vera Rubin units will go live. It is intended to enhance Grok's competitiveness.
d. SI solutions and infrastructure revenue was $7.0bn, with Cursor (consolidated mid-quarter) contributing $608mn. Ex-Cursor, QoQ growth was +192%, mainly from activation of new cloud services agreements; ad revenue rose 19% YoY.
e. Compute unit pricing is slightly above the prior $30–50/W range's upper bound. New deployments have sub-1 year payback; deployment remains demand-led, with contracts allowing flexible allocation between internal training and external rentals.
f. Grok Bot launched in Aug; paying WAUs were 800k in early Oct, more than doubling over the past month. Per-user usage is rising; B2B is the fastest-growing, with full enterprise deployment opening in weeks.
g. Enterprise customers with ARR >$1mn grew 50% QoQ (incl. Cursor, pro forma). A bundled plan combining Grok, Grok Bot, Cursor and X Premium has been launched.
2.2 Q&A
Q: Has TerraFab placed EUV orders and when are deliveries? Where is the biggest room to improve the semi mfg. supply chain?
A: No EUV order confirmations and no specific improvement areas were disclosed; management only noted test chips by mid-2027 and volume production in Q3 2028. The first test chip (SRAM-type, for line validation) is planned to run in the Austin research lab around May–Jun 2027.
The Gibbons Creek main fab targets volume around Q3 2028, with wide uncertainty. Management said mass production in Q3 or Q4 is 'at least within the realm of possible outcomes' and thanked suppliers such as ASML for accelerating tool deliveries.
They cited major process breakthroughs but declined to share details on the call. The long-term goal is annual chip output corresponding to 1 TW of compute.
Q: How will you resolve power, cooling and infrastructure bottlenecks, and what underpins confidence in reaching 13 GW by end-2027?
A: Nearly 3 GW is locked in for this year, and 10 GW of additions in 2027 is highly certain, with the probability of surpassing 20 GW rising rapidly. About 220k GB300 units are energized each month, and the pace will accelerate next year. A plan is in place to add 20 GW of power and cooling by end-2027 (totaling ~23 GW), but specifics are proprietary; management also cautioned the path will not be perfectly smooth.
Public measures include using rocket engine casting capabilities to produce turbine blades and vanes for gas turbines. Suppliers indicated blades/vanes are delivered 12–18 months later than other turbine parts; in-house production can bring power online 12–18 months sooner.
They are also adding solar, wind and other sources, and exploring factory-prefab, modular inference clusters with embedded power and cooling for distributed deployments.
Q: How will you fund the compute scale required?
A: Prioritize operating cash flow, supplemented by multi-channel debt and ecosystem partnerships, while maintaining IG ratings. The company has an IG balance sheet and is in discussions with multiple lenders; compute payback is short, with capacity added by quarter-end expected to contribute cash from H1 2027. Financing can also be arranged with suppliers and on the customer side; management highlighted capital efficiency as rare in the industry.
Q: Pros, cons and hurdles of a merger with Tesla?
A: Personally, the CEO views a merger as a good idea, but the special committees at both companies have not agreed on any terms. At shareholder request, both formed special committees of independent directors; as a conflicted party, the CEO is recused, and management said the market overestimates the likelihood of a deal.
On the Tesla side, approval requires 50% of all shares outstanding (not just votes cast), a high bar. Synergies include super-intelligence software, chip design and TerraFab, robotics, Starmine constellation launch capacity and connectivity.
Q: How will the satellite layer and ground network split roles for Starlink Mobile? Which is the focus: retail wireless or physical AI?
A: S-band will carry most traffic, while low-band handles building penetration. Connectivity revenue could reach hundreds of billions per year. About 85%–93% of mobile traffic in practice rides on Wi‑Fi, and Starlink broadband already carries a large share of that.
The 800 MHz low band has limited capacity but is critical, ensuring indoor coverage without Wi‑Fi. Small cells will also be added to homes, enterprises and towers already equipped with Starlink, reusing existing sites; management said external observers have missed internal factors and a dedicated briefing will follow, without directly prioritizing retail vs. physical AI.
Q: How will you allocate compute between external rentals and internal (Grok, Grok Bot, Cursor), and how do you monetize over the next 12–18 months?
A: Roughly a 50/50 split, flexing around 60/40; rentals are short-term at spot, with a bet on rising $/W value. As models improve, useful intelligence per watt rises and compute value increases; rents for older H100s are already higher than at launch, and short-term renewals price higher per watt.
Hyperscalers and dozens of smaller firms continue to seek capacity, with demand multiples of supply. The company prefers to rent out inference to free training resources; GB300, GB200, TPU, Trainium and H100 are all sold out, and the CEO did not rule out payback shrinking to six months next year.
Q: When will users noticeably feel the capacity uplift from V3 satellites, and how many are needed?
A: It will take several hundred V3 satellites for users to feel a clear improvement. Management expects to reach that around mid-2027, noting the current satellite count is too low for consistent gains.
Q: Progress at the Louisiana Starbase site, and how likely is a first launch by 2029?
A: The internal target is a 2028 first launch. A 2029 first launch is highly likely, with scaled operations during 2029.
<End of text>
Risk disclosure and statement:Dolphin Research disclaimer and general disclosure
