Colossus Memphis — SpaceXAI / xAI: A vast training cluster needs useful output after renewal costs

FJI source-qualified: four-document gate met, including project-specific government evidence. Qualification evidence and decisions →

Project status: Dated large-scale training operation documented; current configuration unreconciled. Capacity scope: 100,000 Hopper GPUs documented October 2024; operator claims 200,000 GPUs February 2025. How we verify evidence.

Compare this project → · Track its catalysts → · Evidence guide →

Expanded analysis · Original disclosures linked below · CAD and USD preserved as labeled · Company statements attributed; interpretations labeled.

The one-minute investment takeaway

Our analysis: A vast training cluster needs useful output after renewal costs. Colossus supplies unusually specific hardware evidence and a named public supplier. Its scale claims still need dating and reconciliation, while the original Memphis facility must remain separate from later Southaven developments.

Readiness basis: Classification of evidence in the reviewed profile; strongest documented phase, not full-campus completion. Dated large-scale training operation documented; current configuration unreconciled Readiness source →

What the sources establish

• Supplier corroboration. NVIDIA’s October 28, 2024 announcement describes a 100,000-Hopper-GPU Memphis cluster using Spectrum-X networking for Grok training. This is dated evidence, not a current inventory. Primary source →
• Later scale claims. The current operator page claims 200,000 GPUs by February 2025 but also displays a separate 180,000 H100 statistic. Those figures are not internally reconciled; a million-GPU roadmap is not achieved capacity. Primary source →

Who receives the money?

Our analysis of disclosed roles. Revenue routes are business-model interpretations, not estimates of booked project revenue.

How financially meaningful is it?

Our analysis: Model training infrastructure consumes cash through accelerators, networking, cooling, power and periodic replacement. A rapid build can reduce time to first useful training, but it does not establish the all-in capital cost or the future value of model improvements. Investor analysis should examine training output, failure rates, network efficiency and subsequent commercial contribution. NVIDIA’s confirmed hardware and network role is much stronger evidence than guessing vendors at another AI campus, yet the release gives neither a project-specific gross margin nor all subsequent order volumes. SpaceXAI is not assigned an invented public ticker. SpaceX announced its acquisition of xAI on February 2, 2026. Parent-company ownership is distinct from the dated Grok training evidence and campus-specific commercial arrangements. The advertised million-GPU ambition remains an option requiring more capital, power and accepted demand.

Milestones that would change the assessment

What could break the investment case?

Our analysis: GPU obsolescence, network inefficiency and unreliable power can impair useful output despite impressive device counts. Inconsistent operator inventory figures weaken precision. Southaven power assets are not automatically Memphis campus assets.

What would change our view?

Our analysis: Dated acceptance, disciplined incremental capital and useful output at competitive resource cost would strengthen our assessment. Idle equipment, repeated delivery slippage or capital and renewal spending ahead of economic benefits would weaken it. We change the stage only on project-specific evidence, and would improve public-company financial attribution only with contract scope and cash-flow detail. A larger headline by itself would not justify a stronger return conclusion.

What remains unknown

Current accepted IT and facility MW, workload mix, utilization, site-attributable income, contract prices, tariffs, hardware renewal schedule and phase-specific capital allocation remain unestablished in the reviewed sources unless explicitly disclosed above. Targets remain expectations until confirmed. The map pin represents the locality rather than a verified individual building. Supplier, fund and corporate exposures must be modeled separately.

Project research for information, not a buy recommendation. Financial illustrations use assumptions, not disclosed site results or management guidance.

SpaceX parent-company connection

SpaceX announced its acquisition of xAI on February 2, 2026. Colossus is presented by its operator under SpaceXAI branding. We identify SpaceX as the parent company while retaining Colossus Memphis as the campus identity. This does not establish a Starlink project connection.

Official acquisition announcement → · ownership evidence reviewed October 11, 2026