January 1, 2026
January 2026 Update: Elon Musk announced xAI purchased a third building in Memphis, expanding Colossus to 2 gigawatts total capacity. The facility will house 555,000 NVIDIA GPUs purchased for approximately $18 billion—making it the world's largest single-site AI training installation.
TL;DR
xAI's Colossus expansion to 2 GW represents an unprecedented concentration of AI compute. The 555,000 GPU deployment ($18B) exceeds any other single-site AI facility globally. With on-site gas power generation and 19-day buildout timelines, xAI demonstrates a construction model that compresses what typically takes 4 years into weeks. Infrastructure planners must reckon with this new benchmark for scale and speed.
What Happened
On December 30, 2025, Elon Musk revealed xAI purchased a third building near its Colossus 2 data center in Memphis, Tennessee.1 The expansion brings total site capacity to nearly 2 gigawatts.
Colossus footprint:
| Facility | Status | GPUs | Power |
|---|---|---|---|
| Colossus 1 | Operational | 230,000 (incl. 32K GB200s) | ~500 MW |
| Colossus 2 | Operational | 550,000 GB200s/GB300s | ~1 GW |
| Building 3 ("MACROHARDRR") | Purchased | Planned expansion | ~500 MW |
| Total | 555,000+ | ~2 GW |
The new building sits adjacent to Colossus 2 near Southaven, Mississippi, close to a gas-fired power plant xAI is also constructing.2
Musk named the third structure "MACROHARDRR"—extending his "Macrohard" naming convention, a dig at Microsoft.3
Why It Matters
Scale Without Precedent
The 2 GW Colossus complex dwarfs every other AI training facility:4
| Facility | Power | GPUs | Operator |
|---|---|---|---|
| xAI Colossus (Memphis) | 2 GW | 555,000+ | xAI |
| Meta AI Research Center | ~500 MW | ~150,000 | Meta |
| Microsoft Azure AI | ~400 MW | ~100,000+ | Microsoft |
| Google TPU clusters | ~300 MW | TPU v5 equivalent |
xAI's facility represents 4x the power of the next-largest dedicated AI training site.
$18 Billion GPU Investment
The 555,000 GPU purchase at approximately $18 billion implies:5
- Average cost: ~$32,400 per GPU
- Mix includes NVIDIA's latest: GB200s and GB300s
- July 2025: "First batch of 550k GB200s & GB300s" went live at Colossus 2
For context, $18B exceeds the annual capex of most tech companies and represents roughly 3% of NVIDIA's total GPU shipments concentrated in a single customer.
Construction Speed
NVIDIA CEO Jensen Huang called the original Colossus buildout "superhuman"—operational in 19 days versus the typical 4-year timeline.6
| Milestone | Traditional | xAI Colossus |
|---|---|---|
| Site selection to groundbreaking | 6-12 months | Weeks |
| Construction | 2-3 years | 19 days |
| Power provisioning | 1-2 years | On-site generation |
| GPU installation | 3-6 months | Concurrent with build |
This velocity comes from vertical integration: xAI builds its own power generation on-site rather than waiting for utility interconnection.
Technical Details
Power Infrastructure
The Memphis site bypasses traditional utility constraints through on-site generation:7
- Gas-fired power plant under construction adjacent to data center
- 2 GW total load—equivalent to powering ~1.5 million homes
- Avoids ERCOT-style interconnection queues
- Self-contained power generation + consumption
GPU Configuration
Based on Musk's disclosures:8
| Generation | Count | Notes |
|---|---|---|
| GB200 | ~520,000 | First batch operational July 2025 |
| GB300 | ~30,000 | Latest Blackwell variant |
| H100/H200 (legacy) | ~30,000 | Colossus 1 original install |
The GB200-NVL72 configuration (72 GPUs per rack) suggests approximately 7,700+ compute racks at full deployment.
Cooling Requirements
2 GW of GPU compute generates approximately 1.8 GW of heat requiring dissipation:9
- Liquid cooling mandatory at this density
- Estimated 50,000+ gallons per minute cooling capacity
- Memphis location provides water access via Mississippi River watershed
Competitive Implications
AI Training Arms Race
Musk stated days before the announcement: xAI aims to have "more AI compute than everyone else."10
The 2 GW facility supports this goal:
| Company | Estimated AI Training Compute | Status |
|---|---|---|
| xAI | 2 GW (555K GPUs) | Largest single site |
| OpenAI/Microsoft | ~1.5 GW (distributed) | Azure infrastructure |
| ~1 GW (TPU + GPU) | Distributed globally | |
| Meta | ~800 MW | Multiple facilities |
| Anthropic | ~500 MW | AWS + FluidStack |
Grok Model Training
Colossus exists primarily to train xAI's Grok models. The expanded capacity enables:11
- Larger model parameter counts
- Faster training iteration cycles
- Multi-model parallel training runs
xAI's target: 1 million GPUs total. The 555,000 at Colossus represents 55% of that goal at a single site.
What's Next
2026 Timeline
- Q1 2026: Building 3 conversion to data center begins
- Q2-Q3 2026: Additional GPU deployment
- 2026: Gas power plant completion
Expansion Path
Musk has indicated plans for 1 million+ GPUs total. Potential paths:12
- Memphis site expansion beyond 2 GW (requires additional power)
- Second major site (location TBD)
- Acquisition of existing data center capacity
Industry Impact
The Colossus model—on-site power generation, compressed timelines, massive single-site scale—may become the template for frontier AI training facilities. Traditional data center development cycles appear inadequate for AI training demand.
For large-scale GPU deployment and data center infrastructure, contact Introl.
References
-
Bloomberg. "Musk's xAI to Expand 'Colossus' Data Center, Information Reports." December 30, 2025. https://www.bloomberg.com/news/articles/2025-12-30/musk-s-xai-to-expand-colossus-data-center-information-reports ↩
-
SiliconANGLE. "Elon Musk reveals plan to expand xAI's 'Colossus' data center to 2 gigawatts." December 30, 2025. https://siliconangle.com/2025/12/30/elon-musk-reveals-plan-expand-xais-colossus-data-center-2-gigawatts/ ↩
-
Benzinga. "Elon Musk Says xAI Purchased Third Building For Massive AI Expansion." December 2025. https://www.benzinga.com/markets/tech/25/12/49642773/elon-musk-says-xai-purchased-third-building-for-massive-ai-expansion-as-company-takes-on-openai-anthropic-property-name-takes-dig-at-microsoft ↩
-
Tom's Hardware. "Musk to expand xAI's training capacity to a monstrous 2 gigawatts." December 2025. https://www.tomshardware.com/tech-industry/artificial-intelligence/musk-purchases-third-building-at-memphis-site-to-expand-xais-training-capacity-to-a-monstrous-2-gigawatts-announcement-comes-days-after-musk-vows-to-have-more-ai-compute-than-everyone-else ↩
-
Techzine Global. "xAI expands Colossus megadata center to 2 gigawatts." December 2025. https://www.techzine.eu/news/infrastructure/137578/xai-expands-colossus-megadata-center-to-2-gigawatts/ ↩
-
SemiAnalysis. "xAI's Colossus 2 - First Gigawatt Datacenter In The World." 2025. https://newsletter.semianalysis.com/p/xais-colossus-2-first-gigawatt-datacenter ↩
-
The Edge Malaysia. "Musk's xAI buys building to expand 'colossus' data centre." December 2025. https://theedgemalaysia.com/node/787706 ↩
-
WebProNews. "Elon Musk's xAI Doubles Colossus Supercomputer to 2GW in Memphis." December 2025. https://www.webpronews.com/elon-musks-xai-doubles-colossus-supercomputer-to-2gw-in-memphis/ ↩
-
Yahoo Finance. "Musk's xAI buys third building to expand AI compute power." December 2025. https://finance.yahoo.com/news/musks-xai-buys-third-building-221629820.html ↩
-
Stocktwits. "Elon Musk Is Ramping AI Compute At Breakneck Speed." December 2025. https://stocktwits.com/news-articles/markets/equity/elon-musk-is-ramping-ai-compute-at-breakneck-speed/cL7BTsJREzG ↩
-
Bloomberg. "xAI Colossus Data Center." December 2025. ↩
-
Tom's Hardware. "xAI training capacity expansion." December 2025. ↩
