TL;DR
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
GF Securities analyst Jeff Pu estimates SpaceX will be allocated 20,000 NVIDIA Vera Rubin NVL72 racks in 2027, although he says the company sought 30,000. His forecast would make SpaceX the largest individual buyer in a market he estimates at 96,000 racks that year; NVIDIA and SpaceX have not confirmed the figures.
SpaceX is expected to receive 20,000 NVIDIA Vera Rubin NVL72 racks in 2027, according to GF Securities analyst Jeff Pu, making it the largest individual buyer in his forecast despite seeking 30,000. The estimate points to a major planned investment in AI computing infrastructure, but it is an analyst projection, not a confirmed allocation from NVIDIA or SpaceX.
Pu estimates the market for NVIDIA’s NVL72 rack systems will reach 96,000 units by the end of 2027, compared with 73,000 by the end of 2026. In his account, SpaceX requested 30,000 racks but is expected to be allocated 20,000. The difference suggests demand may exceed the supply available to the company, though the report does not detail how NVIDIA determines allocations or whether the forecast has changed since it was prepared.
The analyst also projects that six major neocloud providers will procure 34,000 racks in 2027, or about 35% of the estimated 96,000-unit market. The figures describe expected industry demand, not completed purchases or deliveries. Pu’s report says Taiwan-based Foxconn may handle as much as 45% of SpaceX’s data-center orders; Foxconn is among the contract manufacturers that assemble NVIDIA GPUs into rack systems.
Pu’s outlook also includes a product ramp: his figures put 2026 demand at 67,000 GB200 and GB300 racks and 6,000 Vera Rubin VR200 racks. He expects VR200 and initial VR300 Ultra systems to drive most of the market’s growth in 2027. The report cites an expected first-pass yield above 85% for Vera Rubin, with further acceleration anticipated in the first quarter of 2027. These, too, are analyst estimates and expectations rather than company-reported results.
AI Infrastructure · Analyst Forecast · 2027
SpaceX Could Be 2027’s Biggest NVIDIA Rack Buyer
GF Securities analyst Jeff Pu estimates SpaceX may receive 20,000 Vera Rubin NVL72 racks in 2027 after seeking 30,000. The projection would make it the largest individual buyer in his forecast—though neither NVIDIA nor SpaceX has confirmed the figures.
A large buyer in a growing market
Pu’s estimates describe expected demand and supply allocation. They do not represent completed purchases, delivered systems, or deployed computing capacity.
Projected rack volumes
Request versus expected allocation
Pu’s account says SpaceX sought 30,000 racks and may be allocated 20,000. The gap suggests demand could exceed the supply available to the company.
The report does not explain NVIDIA’s allocation process or say whether the estimate reflects a finalized supply agreement.
More than a rack count
Manufacturing, product timing and component availability all affect how quickly projected demand could become operational capacity.
Foxconn’s possible role
Pu estimates Taiwan-based Foxconn may handle as much as 45% of SpaceX data-center orders. The report includes no company confirmation or detailed scope.
A generational transition
Pu’s 2026 estimate includes 67,000 GB200/GB300 racks and 6,000 VR200 racks. He expects VR200 and initial VR300 Ultra systems to drive much of 2027 growth.
Yield is a milestone
The outlook cites a first-pass Vera Rubin yield above 85%, with further acceleration expected in Q1 2027. These are analyst expectations, not reported results.
Four stages to watch
Market demand, allocated supply and delivered systems are separate steps. Each needs its own evidence.
Demand forecast
Pu estimates 96,000 racks across the market in 2027.
Supply allocation
SpaceX’s projected 20,000 allocation has not been confirmed.
Manufacturing
Vera Rubin production and partner assembly must ramp.
Delivery & use
Shipments and deployment would show operational capacity.
Financing claims need attribution
The broader discussion links AI infrastructure spending with borrowing and neocloud competition, but the report does not tie those claims to a confirmed rack purchase.
What the report attributes
What remains unknown
- Whether SpaceX has secured an allocation or placed an order for 20,000 racks
- How many systems SpaceX has already ordered or when deliveries could occur
- How pricing, memory supply and manufacturing capacity could shift the forecast
- Which six neocloud providers are included and how the market estimate was calculated
- The terms of the reported financing and how much has been spent on GPUs
“NVL72 rack TAM: 73k / 96k in 2026E/2027E.”
Jeff Pu · Market outlook posted October 27, 2026Key questions
Figures below are attributed to analyst Jeff Pu unless stated otherwise.
How many racks could SpaceX receive?
Pu estimates 20,000 NVL72 racks in 2027 after a request for 30,000. NVIDIA and SpaceX have not confirmed the figures.
What is an NVL72 rack?
A rack-scale AI computing system built around NVIDIA GPUs. Contract manufacturers, including Foxconn, assemble rack systems.
How large could the market be?
Pu estimates 96,000 racks by the end of 2027, up from 73,000 by the end of 2026. These are forecasts, not confirmed sales.
Could Foxconn assemble SpaceX systems?
Pu estimates Foxconn may handle up to 45% of SpaceX’s data-center orders. The report does not provide confirmation or specify the scope.
SpaceX’s Planned AI Capacity
If the forecast is borne out, SpaceX’s expected 20,000-rack allocation would make it a substantial buyer of NVIDIA’s next-generation AI infrastructure. That could support the company’s plans to provide cloud computing services and would position it among the largest customers competing for systems during a period when Pu expects demand to expand.
The projected shortfall between SpaceX’s request and allocation also illustrates the supply question behind the spending plans: a large budget does not guarantee access to the requested number of racks. NVL72 systems combine GPUs into high-density computing infrastructure, and the report describes them as expensive, with estimates often reaching millions of dollars per rack amid constrained memory-chip supply. Those price and supply pressures could affect how quickly providers build capacity and how much computing they can offer customers.
The forecast is not evidence that SpaceX has already received the hardware or secured a particular amount of revenue from it. Its significance is as a view of expected orders and manufacturing demand in 2027, rather than a measure of deployed capacity or business performance.
Products Worth Considering
Samsung 27-inch Odyssey OLED G6 QD-OLED QHD Gaming Monitor
Logitech MX Master 4 for Mac, Ergonomic Bluetooth Mouse with Haptic Feedback, Ultra-Fast Scrolling, USB-C Charging, for MacOS, iPadOS, Windo
LG 34GX90SA-W 34-inch Ultragear WQHD OLED Curved Gaming Monitor
As an affiliate, we earn on qualifying purchases.
Debt, Neoclouds and Rack Demand
The report follows public comments by Elon Musk about financing AI infrastructure and the outlook for neocloud providers, companies that offer computing capacity built around specialized AI data centers. According to the account, Musk said SpaceX had raised $40 billion in debt to finance NVIDIA GPUs and attributed its ability to borrow to its credit rating. That statement is attributed to Musk; the report does not provide independent documentation of the financing or its terms.
Musk also argued that firms without another revenue stream could face difficulty borrowing and predicted that many neoclouds would fail or consolidate. The forecast from Pu offers an estimate of where some demand may be concentrated, but it does not establish that SpaceX’s financing will fund the precise rack quantity in the report. Nor does it confirm that rival providers will fail.
NVIDIA makes the Vera Rubin GPUs, while rack systems such as the NVL72 require assembly by manufacturing partners. Pu’s projections span multiple generations: his 2026 estimate includes GB200 and GB300 systems alongside an initial VR200 volume, while his 2027 forecast anticipates VR200 and early VR300 Ultra systems. That product mix matters because the 96,000 figure refers to the analyst’s rack-market estimate, not solely to a single GPU generation.
“200+ neoclouds are going to see guppy-level death and consolidation. Some will make it through, but not many.”
— Elon Musk, as quoted in the report
Allocation and Delivery Still Unconfirmed
The report does not include confirmation from NVIDIA, SpaceX or Foxconn of the rack request, expected 20,000-unit allocation, or Foxconn’s projected share of SpaceX data-center orders. It also does not specify when deliveries would occur, how many systems SpaceX has already ordered, or whether the analyst’s allocation estimate reflects a finalized supply agreement.
The 96,000-rack market estimate and the 34,000-rack projection for six major neoclouds are forecasts, not confirmed orders. The report does not name all six providers or explain the methodology behind the total addressable market calculation. It remains unclear how pricing, memory supply, manufacturing capacity and product-ramp performance could alter the estimates.
Musk’s reported $40 billion debt figure and comments on neocloud consolidation are attributed statements. The available details do not establish the financing’s terms, how much has been spent on GPUs, or whether the borrowing is directly tied to the specific 20,000-rack projection.
Evidence to Watch in 2027
Further confirmation would likely come through company statements, supply-chain disclosures, or reported shipments as NVIDIA’s Vera Rubin systems enter production and deployment. The first-quarter 2027 ramp cited in Pu’s outlook is one near-term milestone, but the report does not give a confirmed delivery schedule for SpaceX.
Readers should distinguish forecast demand, allocated supply and delivered racks as new information emerges. Updates to NVIDIA’s production outlook, Foxconn’s assembly work, or SpaceX’s financing and cloud plans could clarify whether the projected 20,000 units are secured and when they might become operational.
Source: Wccftech
Key Questions
How many NVIDIA racks is SpaceX expected to receive?
GF Securities analyst Jeff Pu estimates SpaceX will be allocated 20,000 NVL72 racks in 2027, after seeking 30,000. Neither company has confirmed the figures.
What is an NVL72 rack?
It is a rack-scale AI computing system built around NVIDIA GPUs. The report describes NVIDIA as the GPU maker and companies including Foxconn as contract assemblers of rack systems.
How large does the analyst expect the 2027 market to be?
Pu estimates the addressable market at 96,000 NVL72 racks by the end of 2027, up from 73,000 by the end of 2026. These are analyst estimates, not confirmed sales totals.
Has SpaceX confirmed the rack order or allocation?
No confirmation from SpaceX or NVIDIA is included in the report. The request for 30,000 and expected allocation of 20,000 are attributed to Pu’s analysis.
What role could Foxconn play?
Pu estimates Foxconn may handle as much as 45% of SpaceX’s data-center orders. The report does not provide company confirmation or specify the schedule and scope of that work.
Source: Wccftech
