Building AI Infrastructure Responsibly: ER Steel on the Evolving Demands of Data Center Expansion

Micron and Anthropic sign a multi-year AI memory supply deal

Micron Technology ký thỏa thuận nhiều năm cung cấp HBM, DRAM, SSD cho trung tâm dữ liệu của Anthropic, cùng hợp tác tối ưu kiến trúc bộ nhớ cho AI, và đầu tư chiến lược vào vòng Series H của Anthropic. Thỏa thuận này phản ánh xu hướng ngành khi các nhà sản xuất chip và đám mây vừa là nhà cung cấp vừa là cổ đông của các phòng thí nghiệm AI.

Lập trình viên nên đọc bài này để hiểu cách các công nghệ xử lý bộ nhớ (HBM, DRAM) và kiến trúc lưu trữ mới đang định hình hiệu suất, tiết kiệm năng lượng cho các mô hình AI lớn, từ đó tìm hiểu cách tối ưu hóa ứng dụng của mình với những tiến bộ này.

Meta, like SpaceX, looks to turn excess AI compute into cash

Meta is reportedly developing a cloud infrastructure business called Meta Compute to monetize excess AI compute capacity and models, following a similar move by SpaceX's xAI. The initiative would compete directly with AWS, Google Cloud, and Microsoft Azure. Meta has committed $182.9 billion to AI infrastructure, including massive data center projects in Louisiana and Ohio, but has yet to generate significant standalone revenue from its AI products like Meta AI or Llama. The new business may sell raw compute capacity similar to CoreWeave, as well as hosted model access similar to AWS, including its recently launched closed-weight model Muse Spark.

Empowering Mission Success: Public Sector Highlights and Key Takeaways from Cisco Live US 2026

Cisco Live US 2026 public sector highlights cover government and education sessions focused on AI-ready infrastructure, cybersecurity, and network modernization. Key topics include AI-enabled government services, AI-native campus design at UNC Charlotte, personalized learning through AI, and federal discussions on aging infrastructure, cloud adoption, and Post-Quantum Cryptography. Yale University was recognized with the Modernized Infrastructure Maestro Award for replacing its legacy network with a secure Cisco architecture. On-demand sessions are available for those who missed the event.

Digital Realty to buy Blackstone’s stake in three Virginia data centres

Digital Realty is acquiring Blackstone's 64% equity stake in three fully leased Northern Virginia data centres for $3.5 billion ($1.2B cash + $2.3B in stock), valuing the portfolio at $7.8 billion. The three facilities — two in Manassas and one in Sterling — collectively provide 288 megawatts of IT capacity, all leased to investment-grade hyperscale tenants. The deal reflects surging AI-driven demand for contracted data centre capacity in Northern Virginia, where power constraints make fully energised, occupied assets especially scarce and valuable. For Digital Realty, it's a fast path to scale without construction risk; for Blackstone, it's a partial exit that keeps exposure to the sector via Digital Realty shares.

MongoDB embeds reranking into Atlas as enterprises look to simplify AI stacks for scale

MongoDB has introduced Native Reranking as a public preview feature in Atlas, powered by Voyage AI. It runs directly within the MongoDB aggregation pipeline and can improve retrieval quality by up to 30%. Analysts highlight that embedding reranking natively into the database reduces developer overhead by eliminating separate orchestration layers, retry logic, and failure handling. For CIOs, it simplifies AI stack governance and can lower inference costs by trimming irrelevant context before it reaches expensive LLMs. However, trade-offs include potential vendor lock-in and limited value for enterprises whose data spans multiple repositories. The move reflects a broader industry trend of database and data warehouse vendors consolidating AI, transactional, and analytical capabilities to reduce the integration complexity from multi-vendor AI stacks.

Learnings from VT Data Center Summit

Key takeaways from the 2026 Virginia Tech Data Center Summit covering the massive power demands driven by AI workloads, construction and deployment challenges, and emerging efficiency metrics. Northern Virginia hosts 13% of global data center capacity, with Dominion Energy managing 48.5 GW of contracted data center capacity in its pipeline. GPU-driven AI workloads have pushed rack power draw from ~3 kW to 50+ kW, with NVIDIA exploring 800V DC architecture to support 1 MW racks by 2027. Construction bottlenecks center on power availability and labor efficiency, with modularization and offsite prefabrication as key strategies. Virginia's new GS-5 rate class ensures large data center operators bear infrastructure costs rather than residential ratepayers. Emerging AI workload types — training, inference, and agentic — each present distinct engineering challenges, and 'tokens per watt-hour' is gaining traction as a more meaningful efficiency metric than traditional PUE.

Announcing our $800M Series C to accelerate the shift to open-source AI

Together AI has raised $800M in Series C funding from investors including NVIDIA, Aramco Ventures, Vista Equity, and General Catalyst, plus commitments for over 500 MW of compute capacity. The company argues that closed frontier LLMs become economically unsustainable at production scale, while open-weights models like DeepSeek and Nemotron deliver 6x–20x cost reductions. Together AI positions itself as a full-stack open-source AI platform covering inference, training, and accelerated compute, and highlights recent research output including FlashAttention-4 for NVIDIA Blackwell and Together Megakernel.