SpaceX (SPCX): SpaceX’s planned GB300 deployment could lift compute capacity to about 2.7 GW in early 2027

Overview

This company update assesses SpaceX’s expanding AI-compute infrastructure. Deutsche Bank concludes that planned GB300 cluster deployments could materially increase available capacity, supporting existing cloud-hosting commitments and growing internal inference and training needs.

Core views

Elon Musk indicated that an additional 220,000 NVIDIA GB300 GPUs would become fully operational during the week of the report, followed by another 220,000 in November and potentially a further 220,000 by late December. Deutsche Bank estimates each 220,000-unit GB300 cluster represents 440 MW of compute. Against 1.4 GW of compute at the end of 2Q26 and guidance for more than 2 GW at the end of 2026, the firm estimates SpaceX could exceed roughly 2.3 GW by the end of November and reach approximately 2.7 GW in early 2027. The report argues that this incremental capacity should be sufficient to support cloud-hosting agreements alongside rising inference requirements, including Cursor and Grok Bot, and training for Grok 5. The infrastructure expansion is also supported by what the report describes as the largest grid-connected battery pack bank in the United States, with 3.3 GWh of capacity. For neocloud contracts, Deutsche Bank estimates that five signed customers generate a USD 54.5 billion annualized revenue run rate. Management has not disclosed the precise compute allocation to these agreements, but the firm estimates 1.2-1.4 GW is committed, equating to USD 40-45 billion of pricing per GW. This sits toward the upper end of the report’s USD 30-50 billion per-GW range. With further capacity coming online and industry constraints persisting, Deutsche Bank expects at least several additional large deals to emerge. It also views the growth of agentic AI offerings as an additional driver of compute demand.

Analysis framework

The report starts with management’s stated GPU deployment timeline, converts planned GB300 clusters into megawatts of compute capacity, and compares the resulting capacity path with existing cloud-hosting, inference and training requirements. It then estimates neocloud contract revenue by combining assumed compute allocations with implied revenue per GW pricing.

Methodology notes

• Other

  Compute-capacity and neocloud contract run-rate analysis

  The report translates GPU cluster counts into MW of compute, then uses estimated GW allocated to signed contracts and annualized revenue to derive implied pricing per GW.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• SpaceX (SPCX)

  Primary covered company; its compute buildout is expected to support neocloud hosting, inference and AI training demand.

  Strengths

  Rapid planned GB300 deployment, substantial projected compute capacity, signed neocloud contracts and 3.3 GWh of supporting battery storage.

  Weaknesses

  Management has not disclosed the exact compute allocation to signed neocloud contracts.

  Comparison

  Implied USD 40-45 billion per-GW neocloud pricing is toward the upper end of the USD 30-50 billion per-GW range cited by the report.

Key data

• Compute capacity at end-2Q261.4 GWReported starting capacity level.
• Year-end 2026 compute guidance>2 GWCompany guidance cited by the report.
• GB300 cluster capacity440 MW per 220,000 GB300 GPUsDeutsche Bank estimate.
• Estimated capacity by end-November~2.3 GWBased on planned GB300 deployments.
• Estimated capacity in early 2027~2.7 GWIncludes potential additional late-December deployment.
• Signed neocloud customer run rateUSD 54.5 billion annualizedEstimate for five signed customers.
• Estimated contract compute allocation1.2-1.4 GWManagement has not disclosed the exact allocation.
• Implied neocloud pricingUSD 40-45 billion per GWToward the upper end of the report’s USD 30-50 billion per-GW range.
• Grid-connected battery pack bank3.3 GWhDescribed as the largest in the United States.

Impact & implications

Deutsche Bank believes the planned capacity ramp can accommodate both contracted cloud-hosting demand and expanding AI inference and training workloads. The firm expects additional large neocloud agreements as capacity is added amid industry constraints and growing agentic-AI compute demand.