GAIB 🟠 | RWAiFi (@gaib_ai): State of AI Infra – Week of Mar 10

State of AI Infra – Week of Mar 10 This week’s edition highlights the massive capital expenditure required to secure AI compute, as operators utilize creative financing, from ATM equity programs to billion-dollar credit facilities, to fund unprecedented hardware expansion. Simultaneously, the strategic acquisition of utility companies confirms that power generation remains the ultimate bottleneck and risk-mitigation strategy in the AI era. IREN Secures 50,000 NVIDIA B300 GPUs for AI Cloud Expansion (March 9) Data center operator IREN announced a massive expansion of its AI Cloud services through the purchase of 50,000 NVIDIA B300 GPUs, increasing its total fleet to 150,000. Once fully deployed by late 2026 across its air-cooled facilities in British Columbia and Texas, IREN projects its AI Cloud business will generate an ARR exceeding $3.7 billion. To support this scale, the company anticipates a capital expenditure of approximately $3.5 billion for hardware, networking, and labor. This expansion is being financed through a $6 billion at-the-market (ATM) equity program. While securing necessary capital, it sparked some investor concern regarding stock dilution, leading to a dip in share prices. (Source) Together AI Seeks $1B in New Funding for Training Scale (March 7) Together AI is reportedly seeking to raise $1 billion in a new funding round, potentially valuing the company at ~$7.5 billion, more than doubling its previous $3.3 billion valuation from February 2025. To date, the AI-focused cloud platform has raised $537 million from investors including General Catalyst, Kleiner Perkins, Nvidia, and Lux Capital. The new capital will support the expansion of compute clusters and accelerate product development as it competes with both specialized AI clouds and traditional hyperscalers. The fundraising effort reflects investor demand for alternative AI infrastructure providers that support open-source ecosystems and offer cost-efficient compute. (Source) Morgan Stanley Backs Core Scientific with $1B Credit Facility (March 6) Core Scientific secured up to $1 billion in strategic financing from Morgan Stanley to support data center expansion and accelerate its transition toward AI and high-performance computing workloads. The facility includes an initial $500 million commitment with an accordion feature to increase to $1 billion, providing flexible capital for new capacity and site upgrades. The funding will primarily develop large-scale data center campuses and GPU-ready infrastructure capable of supporting AI training and inference. The deal reflects growing institutional confidence in AI infrastructure as a long-term asset class, particularly as former crypto-mining operators reposition to serve the rapidly expanding market. (Source) Perplexity Diversifies Cloud Footprint with CoreWeave Deal (March 5) Perplexity AI signed a cloud capacity agreement with CoreWeave to support the scaling of its AI services. Perplexity will run a portion of its inference workloads on CoreWeave’s dedicated NVIDIA GB200 NVL72 GPU clusters, enabling higher performance and lower latency. This agreement follows Perplexity's earlier $750 million, three-year cloud agreement with Microsoft. It reflects a broader industry trend where AI-native companies increasingly rely on specialized “AI clouds” rather than traditional hyperscalers for large-scale inference workloads, highlighting the rising strategic importance of GPU-optimized infrastructure. (Source) GIP and EQT’s $10.7B Acquisition of AES (March 4) A consortium led by BlackRock-owned Global Infrastructure Partners (GIP) and EQT has agreed to acquire U.S. power producer AES Corporation, valuing the equity at $10.7 billion. AES is a major supplier of renewable energy, with existing power agreements supporting Google, Microsoft, and Meta. The buyers plan to leverage AES’s 34.7GW portfolio of generation assets to expand energy supply for digital infrastructure outside the constraints of public markets. This reflects that the most valuable asset in the AI era is not only the chip, but the power generation behind it. As grid operators struggle with transmission backlogs, owning the utility source is the ultimate risk-mitigation strategy. (Source) VNET Secures Major Data Center Deal with ByteDance (March 3) Vnet Group reportedly secured a major data center capacity agreement with ByteDance that could provide roughly 500MW of capacity in China. The agreement will supply computing and storage infrastructure to support ByteDance’s expanding AI workloads, including the development of its Doubao chatbot. The scale of the deal represents a significant portion of Vnet’s existing 783MW wholesale capacity. It highlights the rapidly rising demand for large-scale AI infra among major tech companies, underscoring how leading AI developers are securing long-term capacity to support LLM requirements. (Source) CoreWeave Anticipates Doubling 2026 Capex (March 3) CoreWeave plans to add 5GW of additional data center capacity by 2030 as part of an aggressive expansion strategy to support surging demand for AI compute. The company expects its CapEx to more than double in 2026 to approximately $30–35 billion, aiming to more than double its active power capacity to over 1.7GW by year-end. The company’s financial growth reflects this push, with 2025 revenue surpassing $5.1 billion and contracted backlog reaching $66.8 billion. This highlights the massive demand for specialized AI cloud providers as enterprises race to secure dedicated compute capacity. (Source)