Meta Platforms is executing one of the most aggressive infrastructure buildouts in the technology sector. The driver is simple: AI compute demand requires physical assets, and physical assets require power. Meta’s strategy is to scale its global data center footprint through owned gigawatt-scale campuses and external capacity agreements, securing a competitive moat in artificial intelligence. The scale, power sourcing, and capital structure of these projects directly determine Meta’s long-term cash flows, balance sheet resilience, and ability to outpace rivals in model training and inference deployment.
As of mid-2026, Meta is simultaneously constructing record-breaking facilities in North America and India while signing strategic lease agreements with neocloud operators. This is a dual-track approach: build the permanent base and lease the surge capacity. The math is simple. Control the power, control the compute, control the market.
Key Insights
The Hyperion Campus: A $50 Billion Anchor
The most heavily corroborated data points center on Meta’s Hyperion campus in Louisiana. The project has expanded from an initial 2 GW phase 27 to a 5 GW total capacity 14,27,45,46, with total investments now exceeding $50 billion 14,46,49. Construction began in December 2024 24, with the first 2 GW phase expected by 2030 27. This is a multi-year, capital-intensive ramp. The commitment signals that Meta views Louisiana not as a satellite site, but as a core node in its AI infrastructure network.
Alberta: Regulatory Arbitrage as Competitive Advantage
Meta has launched its first major Canadian data center in Sturgeon County, Alberta, valued between C$10 billion and C$13 billion depending on the report 1,3,8,9,18,32,42. This 1 GW facility 11,21,28,29,30 is enabled by Alberta’s favorable regulatory environment, which includes a "bring your own power" framework 25,48 and access to cold climates for natural cooling 31,43. Alberta’s policies 16,32 offer regulatory and power certainty that traditional hubs like Northern Virginia 2,34,41 cannot match due to grid saturation and zoning constraints 38,40. Meta is moving to where the electrons are cheapest and the regulators are most cooperative. That is the rational play.
Power Strategy: Diversification and Self-Sufficiency
Meta’s power strategy is highly diversified and increasingly self-sufficient. For Hyperion, the company is backing gas, nuclear, and renewable energy sources 13,19,46, while Entergy Louisiana is constructing 10 new gas-fired plants specifically to serve the campus 19. The Alberta site is co-located with Project Greenlight’s natural gas facility 25 and secured through a long-term energy agreement with Capital Power 1,37. This aligns with a broader industry shift toward on-site and dedicated generation to bypass grid constraints 5,7,47. Grid independence is not a luxury. It is a prerequisite for gigawatt-scale operations.
Neocloud Leases: The Capacity Buffer
Beyond owned assets, Meta is actively engaging with neocloud providers to secure near-term capacity. The company has signed a $27 billion agreement with Nebius Group (NBIS), guaranteeing computing capacity starting in 2027 35, and maintains additional capacity deals with CoreWeave and Crusoe 6,23,36. Meta has also committed roughly 1.6 GW of capacity across Crusoe sites 6 and secured a contract with Microsoft for Qualcomm’s AI data center initiative 22. These external partnerships allow Meta to accelerate deployment while its internal campuses ramp. The $27 billion Nebius deal 35 and partnerships with Crusoe 6,26 signal a maturation of the neocloud ecosystem, where specialized AI compute providers serve as strategic capacity buffers for hyperscalers. Traditional hyperscale construction takes 2–3 years per phase 27,28,44. Leasing bridges the gap.
Capital Structure: Ownership and Off-Balance-Sheet Leverage
Contradictions exist in the financial disclosures surrounding these projects. The Alberta campus is reported at C$10 billion 28,29 and C$13 billion 1,18, while Hyperion’s total investment ranges from "more than $50 billion" 49 to "roughly $50 billion" 46. Some sources suggest Meta owns over 30 data centers globally 27,29, while others note the company continues to outsource construction through third-party lease-back arrangements with embedded exit clauses 4,10,12. This dual approach suggests Meta is using both direct ownership and off-balance-sheet financing to manage capital expenditure velocity. The use of sale-leaseback structures 20 and joint ventures 21,28,29 indicates Meta is optimizing balance sheet exposure while retaining operational control. The best hedge is ownership, but the smartest hedge is ownership with someone else’s capital on the balance sheet.
Implications
The Strategic Pivot
This cluster reveals a strategic pivot in how Meta approaches AI infrastructure. Rather than relying solely on traditional hyperscale construction, Meta is combining long-dated lease commitments with rapid external procurement to maintain leadership in AI model training. The 14 GW of total compute capacity targeted by 2027 33,39 is an industrial-scale commitment that demands both owned and leased capacity working in concert.
Financial Confidence and ROIC Risk
Meta’s willingness to commit tens of billions upfront indicates confidence in AI monetization pathways, particularly through Reality Labs, Llama model deployment, and ad-driven AI features 17,30. However, the scale of Hyperion and the Alberta project raises questions about near-term ROIC and free cash flow dilution. Sentiment is noise; the numbers will tell the story. Investors must watch whether the revenue generated by these assets justifies the capital deployed.
Geopolitical Positioning
Meta’s expansion into Alberta and Louisiana reflects a broader trend toward onshore, sovereign-aligned infrastructure. Louisiana’s energy infrastructure 15,19 and Alberta’s regulatory framework position Meta to capture AI workloads that require data residency, low-latency training clusters, and uninterrupted power. This is infrastructure as geopolitical strategy.
What Investors Should Monitor
- Lease utilization rates and neocloud execution risk: The $27 billion Nebius agreement and partnerships with Crusoe, CoreWeave introduce counterparty risk. Delays compress near-term AI compute output.
- Power delivery timelines: Hyperion’s 2030 ramp and Alberta’s permitting schedule are critical path items. Power is the binding constraint.
- Margin expansion: The $50 billion+ Hyperion commitment and the Alberta project must translate into monetizable compute capacity. Capital allocation without returns is capital destruction.
The bottom line: Meta is building the railroad. The question is whether the freight justifies the track.
