All articles tagged with #ai infrastructure
OpenAI is pausing Stargate UK, its plan to host sovereign AI compute in UK data centers with NVIDIA, citing high energy costs and regulatory hurdles, and says it still sees potential for the UK’s AI future and will resume the project when regulation and energy costs enable long-term investment.
Intel and Google announced a multiyear collaboration to advance AI and cloud infrastructure, with Google Cloud continuing to deploy Intel Xeon processors across workloads and expanding joint development of custom ASIC-based IPUs to offload networking, storage and security tasks from host CPUs. The goal is a balanced CPU-IPU platform that boosts performance, energy efficiency and total cost of ownership in hyperscale AI environments, enabling more predictable performance and scalable data-center architectures, as leaders Lip-Bu Tan of Intel and Amin Vahdat of Google emphasize the central role of CPUs and IPUs in modern AI systems.
NVIDIA and Marvell announced a strategic partnership to connect Marvell to NVIDIA’s AI factory and AI-RAN ecosystem via NVLink Fusion, enabling a heterogeneous, rack-scale AI infrastructure. Marvell will supply custom XPUs and NVLink Fusion-compatible scale-up networking; NVIDIA will provide Vera CPU, ConnectX NICs, Bluefield DPUs, NVLink interconnect and Spectrum-X switches, and NVIDIA has invested $2 billion in Marvell. The collaboration aims to accelerate AI workloads and transform telecom networks into AI infrastructure, including AI-RAN for 5G/6G and related optical interconnect innovations.
French AI startup Mistral has secured $830 million in debt financing from a seven-bank consortium to fund a Paris-area data center powered by Nvidia GPUs (13,800 GB300s, about 44 MW initial capacity), with plans to reach 200 MW of European compute capacity by 2027. The center is slated to be operational in Q2 and supports Mistral’s broader push to build AI infrastructure in Europe, including a 1.2‑billion-euro plan for data centers in Sweden.
Crusoe unveiled a new 900 MW AI factory campus in Abilene, Texas to support Microsoft AI workloads, bringing the full Abilene site to about 2.1 GW with two new buildings and an on-site power plant, designed for high-density compute and water‑efficient cooling; energization of the first building is planned for mid-2027, as the expansion aims to accelerate giga-scale AI infrastructure and boost the local economy through jobs and tax revenue.
Alphabet unveiled three AI memory-bottleneck algorithms, led by TurboQuant, aimed at compressing memory demands for high-volume AI inference. Summit Research frames this as a meaningful catalyst for Credo Technology Group Holding Ltd, given Credo's focus on memory connectivity and high-bandwidth interconnects, and sees alignment with Credo's roadmap (ALCs, ZeroFlap Optics, Weaver). This could broaden Credo’s growth beyond copper interconnects into higher-value adjacencies.
The analysis argues Micron’s current stock surge reflects a temporary, cyclical memory-demand spike driven by AI, not lasting growth. As memory capacity expands, margins and earnings are likely to retrace, making MU overvalued versus secular growers with recurring revenues (e.g., data-center REITs like Equinix) and utilities. The piece highlights how capacity expansions have historically crushed cyclical highs, and Micron’s own capex plan will boost supply further, supporting a return to equilibrium rather than permanent gains. Investors should rotate from MU’s immediate-revenue burst to companies with durable, recurring revenue.
An investor under the alias Agar Capital says Microsoft is undervalued despite a 21% year-to-date drop, citing rising earnings versus a lower forward P/E, a triple‑A credit rating, and aggressive AI/cloud investments that could unlock meaningful upside, with price targets near $600 and consensus suggesting roughly 55% upside.
Micron’s results show price-driven growth in DRAM and NAND driven by AI infrastructure demand, with DRAM up mid-60% and NAND up high-70% year over year, pushing gross margins toward 80% as supply remains tight; strategic 5-year contracts improve revenue visibility and AI-driven memory needs in servers, workstations, and high-memory devices expand the market beyond a traditional cycle. Despite solid fundamentals, MU trades around 7.9x FY2026 EPS and 4.8x FY2027 EPS, which the author views as a mispricing: an AI infrastructure stock rather than a pure memory cyclicality play.
Nebius Group (NBIS) plans a $3.75 billion convertible senior notes offering to fund AI cloud and data center expansion, split into $2 billion due 2031 and $1.75 billion due 2033, with shares down in premarket trading.
NVIDIA announced BlueField-4 STX, a modular storage architecture featuring the new CMX context memory storage platform to bring AI-friendly, low-latency data access for agentic AI, delivering up to 5x token throughput, 4x energy efficiency, and 2x faster data ingestion; with a broad ecosystem of storage providers, manufacturing partners, and leading labs planning adoption, STX-based systems are expected in the market in the second half of the year.
NVIDIA unveils the Vera CPU, the first processor purpose-built for agentic AI and reinforcement learning, claiming about twice the efficiency and 50% faster performance than traditional CPUs. Designed for rack-scale AI factories, Vera supports large-scale concurrent workloads (e.g., 256-core racks), uses high-bandwidth NVLink-C2C interconnects, LPDDR5X memory up to 1.2 TB/s, and features 88 Olympus cores. Backed by broad ecosystem partners and major customers, Vera is set for full production in the second half of 2026.
Meta has signed a five-year deal to spend up to $27 billion with Netherlands-based Nebius for dedicated and additional AI infrastructure capacity, as Meta plans up to $135 billion in AI-related capex this year amid a wave of hyperscaler investments. Nebius will deploy Nvidia’s Vera Rubin chips across multiple locations, signaling Nebius’s growth as a European AI cloud player and following past deals with Microsoft and Nvidia.
Palantir teams with NVIDIA to launch the Sovereign AI OS Reference Architecture (AIOS-RA), a turnkey on-prem AI datacenter that layers Palantir’s software stack (Foundry, Apollo, Rubix, AIP) on NVIDIA hardware and software, including Blackwell Ultra GPUs and Spectrum-X networking, to enable latency-sensitive, geo-distributed, data-sovereign AI workloads from hardware procurement to application deployment.
UK-based AI infrastructure startup Nscale raised $2 billion in a Series C at a $14.6 billion valuation, with Nvidia among lead investors alongside Aker ASA and 8090 Industries and other backers; the funds will accelerate a vertically integrated AI infrastructure buildout—GPU compute, networking, data services and orchestration—across Europe, North America and Asia, as the company expands data centers in the UK, U.S., Norway, Portugal and Iceland and eyes an IPO, while pursuing partnerships with Microsoft and OpenAI.