Meta and BlackRock Form $14B El Paso AI Data Centre Venture

Meta and BlackRock Form a $14 Billion AI Data Centre Joint Venture on 28 July 2026

On 28 July 2026, Meta Platforms and BlackRock announced a joint venture to develop, own, and operate a 1 gigawatt AI data centre campus currently under construction in El Paso, Texas. The total expected development cost of the project is approximately $14 billion. BlackRock is acting through its Global Infrastructure Partners and HPS Investment Partners subsidiaries — both acquired by BlackRock in recent years — which together will hold an 80 per cent stake in the venture. Meta will retain the remaining 20 per cent. The transaction is expected to close within days of the 28 July announcement, with the campus expected to begin bringing capacity online in 2028. The deal is the largest single AI data centre investment jointly disclosed by a technology company and an infrastructure fund to date.

The Financial Structure: Equity, Assets, and $12.5 Billion in Debt

The deal's capital structure is characteristic of infrastructure fund investments in physical assets with long operating lives and predictable cash flows. Meta is contributing the El Paso land and its in-progress construction assets to the venture, valued at approximately $2.3 billion. BlackRock is providing approximately $4.9 billion in cash equity for its 80 per cent ownership interest. The remaining and largest portion of the $14 billion total development cost is financed through approximately $12.5 billion in project debt raised against the venture entity. This structure enables both parties to commit to a facility of this scale without deploying the full capital in equity: Meta contributes assets it was already building, BlackRock commits cash at its target ownership percentage, and long-term infrastructure debt covers the remainder. For Meta, the joint venture structure also moves the majority of the data centre's balance sheet exposure off its own books and onto the venture entity, freeing corporate capital for AI model development and product expenditure.

What 1 Gigawatt of AI Compute Capacity Actually Means

A 1 gigawatt data centre campus is significantly larger than any single AI data centre currently operational. Most large-scale AI training facilities today draw between 200 and 500 megawatts at peak. A 1 GW campus at full draw would represent approximately 1 per cent of the estimated peak power consumption of the entire US data centre fleet. The El Paso campus will support AI model training, fine-tuning, and inference workloads at a scale that Meta would otherwise need to procure from cloud providers at significantly higher per-unit cost. Meta's own models — including the Muse Spark family launched in July 2026 at competitive API pricing — require sustained large-scale compute to serve the volumes Meta's global user base generates daily. Owning that compute at construction and operating cost, rather than renting it at cloud margins, is central to Meta's ability to price its AI inference output competitively while sustaining viable economics across multi-billion daily query volumes.

Why El Paso

El Paso offers a combination of renewable energy access — West Texas has among the highest solar and wind resource density in the United States — lower land and operating costs than Northern Virginia or Phoenix, and proximity to Mexico's electricity grid for hybrid power sourcing options. The existing construction activity that Meta contributed to the venture, valued at approximately $2.3 billion, indicates this site had been selected and developed before the BlackRock partnership was finalised. Bringing an infrastructure fund in at this construction stage reflects a deliberate balance sheet decision: rather than carrying the full $14 billion as a corporate capital expenditure, Meta converts a portion of its owned construction programme into a joint venture with institutional investors specifically structured to hold long-duration physical infrastructure assets.

What This Means for Indian AI Teams

For Indian AI product companies and software teams, the Meta-BlackRock announcement carries a direct downstream implication: frontier AI inference pricing will face continued downward pressure as Meta and other hyperscalers bring owned compute online at scale. Meta's Muse Spark 1.1 model, launched in early July 2026 at $1.25 per million input tokens, is already priced at approximately one-quarter of comparable offerings from OpenAI and Anthropic. Owned compute at El Paso scale is what makes sustained sub-two-dollar inference pricing viable for Meta across enormous daily query volumes. Indian engineering teams building AI features on top of frontier model APIs should factor this trajectory into cost planning: the expected economic life of the El Paso campus runs well into the 2030s, and multiple hyperscalers are building similar owned compute at comparable scale. Teams currently priced out of frontier model inference at production volume should plan for meaningful per-token cost reductions from 2028 onward as this wave of owned capacity comes online globally.

The Bottom Line

On 28 July 2026, Meta Platforms and BlackRock announced a $14 billion joint venture to build a 1 gigawatt AI data centre campus in El Paso, Texas, expected to begin coming online in 2028. BlackRock — through its Global Infrastructure Partners and HPS Investment Partners subsidiaries — holds an 80 per cent stake; Meta retains 20 per cent. Meta contributed approximately $2.3 billion in land and in-progress construction assets; BlackRock committed approximately $4.9 billion in cash. $12.5 billion in project debt covers the balance. At 1 GW, the El Paso campus will be the largest single AI data centre by power draw, positioning Meta to run its AI model family at a structural cost advantage that will flow through to global inference pricing through the end of this decade.