Connecting the thread This piece follows Why the AI Supercycle Keeps Getting Longer — IPO Funding and Memory/Storage, The SpaceX IPO and Korean Market Liquidity, The $5.3tn AI Data Center CapEx Era, and Goldman’s Token Demand vs JPMorgan’s Memory ASP. Where the earlier pieces framed “the IPO as the new relay for AI funding,” this one tests that hypothesis concretely against the actual SpaceX use-of-proceeds plan and today’s Chey-Musk meeting.
TL;DR
- The hypothesis: a successful SpaceX IPO accelerates AI infrastructure investment further. The reasoning is simple — the use-of-proceeds plan in the S-1 places AI compute infrastructure expansion, not launch vehicles, front and center.
- SpaceX’s biggest capex and loss center is not the rocket but xAI. xAI lost roughly $6bn in 2025 and burned about $2.5bn in Q1 2026 alone. In short, the ~$75bn raise is fuel poured into a cash-burning sink called AI compute.
- That capital flows into demand for GPUs, HBM, memory, and power. And today’s reported meeting between SK Chairman Chey Tae-won and Elon Musk (scheduled for late June) is the first link tying that flow to Korean memory — custom HBM for Tesla, HBM4 base die, memory for xAI data centers, storage for satellites, and even ESS.
- The conclusion is not a “SpaceX space stock” but “AI compute financing + a Korean HBM/memory/power option." That said, the meeting is still a sole-source report, and custom HBM is at an exploratory stage — both points must be kept separate.
1. Break the Conventional Wisdom: SpaceX Is Not a “Space Stock”
Viewing the SpaceX IPO solely as a “space-themed listing” misses the point. Where the earlier piece examined the liquidity angle, this one looks at where the money is actually spent. The answer is in the S-1 itself.
SpaceX filed a Form S-1 with the SEC on May 20, 2026, and an amendment on June 1. Selling roughly 555.6m shares at $135 per share, it aims to raise about $75bn at a target valuation of roughly $1.75tn — the largest IPO in U.S. history. (CNBC)
The crux is not the size but the use. According to multiple analyses, the S-1’s use-of-proceeds plan places “AI compute infrastructure expansion” ahead of the launch business, with most of the capital allocated to AI compute infrastructure, launch capacity, and Starlink build-out. (KraneShares)
2. Dissecting the Use of Proceeds
2.1 The Biggest Capex and Loss Center Is xAI
SpaceX’s ownership structure makes the picture clear. SpaceX owns xAI, and xAI owns X (formerly Twitter), with the AI data center and chip-related businesses sitting beneath. And the largest operating loss and capex sit precisely in this AI segment. xAI lost roughly $6bn in 2025 and burned about $2.5bn in Q1 2026 alone. (Via Satellite)
Capex itself has gone vertical.
SpaceX total capex trajectory
2023: ~$4.4bn
2025: ~$20.7bn
Q1 2026: ~$10.1bn (single quarter)
This is not the curve of a “rocket company.” It is the curve of a company building AI compute and a satellite network at the same time. (Morningstar)
2.2 “AI Compute” Is Not an Abstraction — It Is Physical Data Centers
SpaceX’s AI compute runs on two tracks.
- Ground data centers: GPU clusters for xAI’s training and inference. This is where enormous GPU, HBM, and power demand originates.
- Orbital data centers (AI1): SpaceX COO Gwynne Shotwell said the company plans to launch the first AI1 unit in late 2027, and before that will put compute onboard Starlink broadband and mobile satellites. (DCD)
So the proceeds are “money going to space” and at the same time “money going to AI compute." The two are not separate — Starlink’s launch capacity itself becomes the delivery vehicle for orbital compute infrastructure.
3. Why This Is “AI Infrastructure Acceleration”
This is where the hypothesis gets tested. The funding relay laid out in the AI supercycle extension piece — Big Tech cash → corporate bonds → public markets — sees its third stage actually open with the SpaceX IPO.
The core logic goes like this.
- xAI is burning cash fast right now ($6bn loss in 2025). The fuel for a loss-making AI compute business is, ultimately, financing.
- While at the private stage that fuel came from inside Musk’s ecosystem and private placements, the SpaceX IPO opens a new funding source: the public market.
- If part of the $75bn flows into AI compute, that translates directly into orders for GPUs, HBM, server memory, eSSD, and power equipment.
In other words, the more successful the SpaceX IPO, the more entrenched a structure in which AI compute capex is hard to stop. This is the mechanism of “successful IPO → AI infrastructure acceleration."
Let’s be honest about the distinction, though. The proceeds may create new capex, or they may merely backstop already-planned capex financially. The direction of “acceleration” is right, but its magnitude must be confirmed by actual execution after listing.
4. Today’s Link: The Chey-Musk Meeting
The point where the hypothesis connects to Korea was reported today. Digital Daily reported exclusively that SK Chairman Chey Tae-won and Elon Musk are set to meet in the U.S. in late June. An SK Group official’s position is that this is “hard to confirm.” (Digital Daily)
The reported agenda is precisely the Korean end of the mechanism above.
| Cooperation Area | Details | Korean Memory Link |
|---|---|---|
| Semiconductors | Exploratory talks on custom HBM for Tesla’s in-house AI5/AI6 chips, HBM4 base die cooperation | SK hynix / Samsung Electronics HBM |
| AI Infrastructure | Memory chips needed for xAI data center expansion | Server DRAM / HBM / eSSD |
| Space & Comms | Memory and storage for SpaceX spacecraft and low-orbit satellites | High-reliability memory / storage |
| Energy | ESS (energy storage systems) for data centers and the power grid | Power / battery value chain |
One official assessed that “SK Group is broadening the scope of cooperation to seize leadership in next-generation AI infrastructure.” (Digital Daily)
The crux is this. The “public-market capital flowing to AI compute” created by the SpaceX IPO meets the “HBM and memory that AI compute requires” that SK holds — at the same table. The starting point of the capital (the IPO) and its destination (Korean memory) enter the same picture.
5. Korea Read-Through (Illustrative Watch Points)
What follows is not a stock recommendation but an illustration of where to look first when the mechanism above actually starts working.
- SK hynix — the most direct name in this link. The HBM leader and a core supply candidate for the meeting’s agenda (custom HBM for Tesla, memory for xAI data centers). That said, custom HBM is still at the “exploratory” stage, not an order.
- Samsung Electronics — a composite option of HBM4 base die, custom-chip foundry, and data center eSSD. If the chips in Musk’s ecosystem (AI5/AI6) use an external foundry, the option value rises.
- Power / ESS / cooling — AI compute is, in the end, a power problem. ESS being on the meeting agenda overlaps exactly with the “the bottleneck after GPUs is power” from the data center CapEx piece.
Common condition: for these names to become candidates, “a hot SpaceX IPO” alone is not enough. Actual use-of-proceeds execution, official confirmation of the meeting, and conversion of custom HBM into orders must all arrive together.
6. Red Team: When This Hypothesis Weakens
- Capital reallocation vs new capex: if the proceeds merely backfill already-planned investment financially, the incremental “acceleration” is small.
- xAI loss risk: AI compute is a massive cash-burning sink ($6bn loss in 2025). Even if listing proceeds cover it, late monetization raises the burden of follow-on financing.
- Unconfirmed meeting: the Chey-Musk meeting is a sole-source report, and SK’s position is “hard to confirm.” The custom HBM on the agenda is also exploratory. It is too early to price it as fact.
- Orbital data centers are long-dated: AI1 is slated to launch in late 2027. The Korean memory upside from “satellite compute” is an option, not near-term earnings.
- IPO funding-drain backlash: as seen in the earlier piece, if a large IPO becomes a funding source by selling existing AI positions, Samsung Electronics and SK hynix could become sell targets in the short term. The material for acceleration is also the material for short-term volatility.
7. Monitoring Checklist
| Checkpoint | Why It Matters | Read |
|---|---|---|
| SpaceX offer price / demand | Sets the scale of financing and the strength of risk-on | A hot deal strengthens the AI compute capex acceleration logic |
| Actual use-of-proceeds execution | Whether “AI compute first” shows up as orders, not just words | GPU / HBM / power order data |
| xAI capex / data center build-out | A leading signal of AI compute demand | Accelerating build-out widens memory demand |
| Chey-Musk meeting made official | Sole-source report → investable event | Official announcement / MOU |
| Custom HBM / base die orders | Exploratory → earnings conversion | SK hynix / Samsung Electronics disclosures |
| Starlink / AI1 satellite compute | Materialization of the orbital data center | 2027 launch schedule |
8. Conclusion
The hypothesis “successful IPO → AI infrastructure acceleration” is clearly supported in direction by the S-1’s use-of-proceeds plan (AI compute infrastructure front and center) and xAI’s capex/loss structure. But its magnitude and timing, and the transfer to Korean memory, must pass three gates: (1) actual use-of-proceeds execution, (2) official confirmation of the meeting, and (3) conversion of custom HBM into orders.
The practical stance is one sentence. See SpaceX not as a space stock but as “AI compute financing + a Korean HBM/memory/power option,” while keeping the sole-source report and the exploratory stage separate from earnings and confirming step by step.
The facts in this piece cite the reporting, disclosures, and analyses noted in the body. The SpaceX IPO is in progress; the Chey-Musk meeting is at the sole-source report stage (SK says “hard to confirm”) and custom HBM is at the exploratory stage. The stock names are not investment recommendations but examples illustrating the analytical flow, and actual investment decisions and responsibility rest with the investor.