Google has agreed to pay SpaceX roughly $920 million a month to rent artificial-intelligence compute capacity housed at xAI data centers, a commitment that runs from October 2026 through June 2029 and totals about $29 billion over its 32-month term (TechCrunch, June 5, 2026; CNBC, June 5, 2026). Bloomberg framed the same agreement as a "$30 billion computing power deal" once ramp-up periods are included (Bloomberg, June 5, 2026). The arrangement covers approximately 110,000 Nvidia GPUs plus "CPUs, memory, and other related components," with access ramping up through September 2026 at a reduced fee before the full rate begins (TechCrunch).
The headline is a U.S. infrastructure story. The read-through that matters for Seoul sits one layer down in the bill of materials.
Why a U.S. compute lease lands in Korea
Every one of those ~110,000 Nvidia accelerators ships with stacks of high-bandwidth memory (HBM), and that market is overwhelmingly Korean. SK hynix (000660.KS), the world's leading HBM maker and Nvidia's primary HBM supplier, held about 62% of HBM shipments in the second quarter of 2025 and roughly 57% in the third quarter, and is expected to hold a majority through 2026 (Counterpoint/Astute Group data, 2026). Samsung Electronics (005930.KS), Korea's largest company and one of the world's biggest DRAM makers, accounted for an estimated 17% of HBM shipments in the second quarter of 2025 and 22% in the third quarter (Counterpoint via Astute Group, 2026), with analyst estimates of 25–30% allocation for Nvidia's HBM4-based Vera Rubin platform — leaving the two Korean firms with the dominant share of every HBM stack inside an Nvidia data-center GPU. SK hynix has reportedly secured roughly 70% of Nvidia's initial HBM4 orders for the Vera Rubin platform (Seeking Alpha citing industry reports, January 2026).
The Google-SpaceX deal does not name a memory supplier, and neither company has disclosed which GPU generation fills the contract, so the precise HBM content cannot be itemized from public documents. What is verifiable is the direction: a fresh, multi-year commitment for six-figure GPU volumes adds to an order book that is already, by suppliers' own accounts, sold out. Samsung and SK hynix have said HBM capacity is reserved through the end of 2026, with customers booking supply years ahead (Tom's Hardware, 2026).
Sizing it against what's already booked
This is the second giant compute lease SpaceX has signed since folding in Elon Musk's xAI in a February 2026 merger that valued the combined entity at about $1.25 trillion (CNBC). Last month, Anthropic agreed to pay SpaceX roughly $1.25 billion a month through 2029 for all the compute at the Colossus 1 data center near Memphis, Tennessee (CNBC). Google's monthly commitment is roughly three-quarters that size (TechCrunch). Stacked on top of OpenAI's Stargate program — which involves a reported memory commitment of up to 900,000 DRAM wafer starts per month from Samsung and SK hynix (Tom's Hardware, October 2025; TrendForce, October 2025) — the pattern is a series of multi-year, take-or-pay-style compute contracts that convert into durable, pre-booked memory demand rather than spot orders.
That distinction is the substance here. Korean memory makers spent the 2022–2023 DRAM downturn exposed to volatile commodity pricing; the 2026 HBM cycle is being underwritten by named, multi-year hyperscaler and neocloud contracts. SK hynix reported a record first-quarter profit in 2026 as memory prices climbed (CNBC, April 23, 2026) — context for how far the demand signal has already been absorbed into Korean chip valuations.
The open question
The deal's structure leaves room to watch rather than conclude. SpaceX must deliver committed GPU capacity by September 30, 2026, or Google can terminate after a grace period or accept reduced capacity (TechCrunch) — meaning the hardware, and its memory content, has to physically arrive on that timeline. The cleaner confirmation point for the Korean read-through is the supplier cadence: Nvidia's Vera Rubin platform entered full production with Samsung, SK hynix, and Micron named as HBM4 suppliers (TechTimes, June 2, 2026), and Nvidia has asked for 16-layer HBM delivery by the fourth quarter of 2026, a step suppliers describe as "technically much harder" than the move from 8 to 12 layers (The Korea Herald, December 29, 2025). SK hynix's and Samsung's next quarterly results, and any disclosure of HBM4 qualification progress, will show whether commitments like the Google-SpaceX lease are tightening an already-sold-out order book or simply re-confirming demand the market has priced.
This article is for informational purposes only and does not constitute investment advice or a recommendation to buy or sell any security. Figures are drawn from the cited sources as of publication; readers should verify against primary disclosures before making decisions.
