AI Infrastructure & Compute
The physical and financial layer under AI: data centers, chips and GPU supply chains, energy and power constraints, neoclouds and inference providers, AI capex and its financing.
Nebius shares fall as data center delays push revenue into next year
Nebius reported quarterly revenue of nearly $1 billion, up sevenfold year-over-year, but its stock dropped 15% due to power and server shortages. The company expects to land at the lower end of its current year forecast of approximately $3 billion.
Jensen Huang's departure raises concerns for AI infrastructure financing
The departure of Nvidia CEO Jensen Huang creates uncertainty for the AI sector, where Nvidia has acted as a financial backstop for neoclouds. Analysts argue the change may lead to a more conservative, shareholder-focused strategy.
SpaceX plans $24 billion Texas data center and independent power plant
SpaceX is planning a gas power plant and data center in Starbase, Texas, to bypass national grid bottlenecks. The first construction phase is estimated to cost $24 billion, though political uncertainty in Texas poses a risk.
GlobalFoundries targets quantum computing and silicon photonics amid stock decline
GlobalFoundries is positioning itself for the next wave of AI through quantum computing and silicon photonics. The company introduced its first quarterly dividend and a $500 million share buyback program as its stock price halved.
CoreWeave raises revenue guidance to $13.3 billion, easing financing concerns
CoreWeave raised its revenue guidance to up to $13.3 billion and secured a $4 billion credit line, alleviating investor fears. The news drove its stock up nearly 25%, though concerns remain about the sector's reliance on Nvidia's leadership.
Anthropic orders $3.4 billion in gas generators for data centers
Anthropic has secured a record-breaking purchase of power generation equipment to support its expanding AI infrastructure. The move allows the company to bypass utility providers and mitigate multi-year supply chain delays.
CoreWeave revenue triples to $3.5 billion amid high capital spending
CoreWeave reported quarterly revenue of $3.5 billion, nearly tripling from the previous period and exceeding expectations. The company plans to invest up to $39 billion in infrastructure this year, more than three times its revenue.
Google plans specialized TPU chips to challenge Nvidia dominance
Google is developing a new generation of Tensor Processing Units with specialized chips for government and scientific workloads. The strategy aims to improve efficiency and reduce reliance on general-purpose GPUs from competitors like Nvidia.
Volpe compares AI infrastructure build-out to 1990s fiber boom
Hanabi Capital’s Mike Volpe argues that current AI investment is more substantial than the late 1990s fiber optic surge. He predicts a market correction will occur, but asserts that the underlying compute assets will remain fully utilized.
Volpe analyzes NVIDIA acquisition strategy through Cisco lens
Hanabi Capital’s Mike Volpe compares NVIDIA’s recent acquisitions to Cisco’s 1990s model, highlighting the importance of repeatable integration processes. He views NVIDIA’s approach as a method to shape industry structure and manage customer concentration risks.
Volpe predicts shift to circuit-switching for AI data centers
Hanabi Capital’s Mike Volpe argues that current statistical multiplexing architectures are ill-suited for predictable AI workloads. He predicts a move toward circuit-switching designs that could reduce hardware costs and disrupt incumbents.
Microsoft ends Oracle deal, Google takes over Crusoe site
Microsoft declined to renew a compute rental contract with Oracle, citing aging hardware, while Google assumed control of a Wyoming data center project from Crusoe. These moves indicate that access to power and latest-generation GPUs is becoming the primary driver of infrastructure value.
Gray Wolf proposes decentralized town-level data centers
Gray Wolf founder Pete Sacco argues that the next major AI market is local inference, requiring smaller, decentralized data centers rather than massive centralized training clusters. He contends that the current neocloud model is architecturally flawed due to mismatches between long-term leases and short-term GPU lifespans.
OpenAI pledges community-friendly data center standards
Following market volatility and public resistance to AI infrastructure, OpenAI announced new rules for socially compatible data centers. The measures include ensuring local residents do not pay for data center electricity and providing data on environmental impacts.
LLMD reaches 1.0 for distributed LLM inference
LLMD, a project for distributed inference on Kubernetes, has reached version 1.0 and is now production-ready. The tool manages stateful conversation contexts to maximize hardware utilization for expensive GPU investments.
Mistral builds own infrastructure to secure compute
Mistral is vertically integrating its business by building its own data center infrastructure to address compute scarcity in Europe. The company finances this build-out through revenue commitments rather than heavy debt leverage, distinguishing itself from leveraged competitors.
HSBC launches ETF on China robotics sector
HSBC launched the MSCI China Robotics Usage ETF to provide exposure to China's dominant physical AI sector. The fund tracks 38 companies, with the largest holding being a precision gearbox maker, despite the sector's recent underperformance.
Voters reject data center proposals, threatening AI infrastructure expansion
Forty-one of forty-two anti-data center measures passed in U.S. ballots this week, marking the first time regular Americans have wielded significant leverage over Silicon Valley infrastructure plans. The rejection threatens the business models of levered AI compute companies that rely on incremental expansion.
Cloudera builds sovereign AI stack with Mistral and NVIDIA
Cloudera has partnered with Mistral and NVIDIA to create a data sovereignty-focused AI stack for European and Middle Eastern markets. The initiative leverages Cloudera's 30 exabytes of data to enhance agentic actions across SaaS systems while maintaining strict regulatory compliance.
AI infrastructure debt faces liquidity crunch and credit constraints
Credit providers are retreating from multi-billion dollar AI data center deals due to saturation with exposure to major AI labs. This tightening is forcing neoclouds to confront severe funding gaps and structural maturity mismatches.
OpenAI rebrands data centers as compute campuses to ease local opposition
OpenAI is adopting the term Compute Campus to make its facilities sound less industrial and more community-oriented. The move aims to mitigate public resistance in the United States while highlighting the sector's growing energy demands.
Google plans $15 billion Malaysia data center investment
Google is set to invest $15 billion in new data center campuses in Malaysia during 2027 and 2028. The move is driven by regulatory friction in the United States, shifting significant AI infrastructure capacity to Southeast Asia.
Germany lags US and China in AI data center capacity
A new study reveals Germany ranks far behind the US and China in computing power relative to GDP. The country faces a significant investment gap and heavy reliance on US cloud providers.
CoreWeave debt crisis triggers neocloud sector selloff
Rising interest rates and CoreWeave's debt load have triggered a selloff in the neocloud sector. Investors are now reassessing the viability of specialized AI compute providers amid financing concerns.
Blockworks pivots to AI agent data infrastructure
Blockworks has transformed from a media company into a data and analytics firm following its acquisition of Massar. The company now provides data infrastructure for AI agents, launching an MCP and API for crypto and financial data.
Dell reinvents itself from PC maker to AI infrastructure leader
Dell Technologies has evolved from a personal computer manufacturer into a major provider of AI infrastructure. The company’s history reflects a shift from supply chain efficiency to navigating the challenges of cloud computing and AI demand.
OpenAI struggles to raise $30B, DeepSeek raises $16B
OpenAI is facing difficulty securing a $30 billion round at a $1.4 trillion valuation, while DeepSeek raised $16 billion at a lower valuation. The contrast highlights divergent capital environments for US and Chinese AI labs.
AIE Shanghai launches as first China conference, highlighting East-West tech gap
The first AI Engineer conference in China opened in Shanghai, organized by local partners due to regulatory restrictions. The event aims to bridge the information gap between US and Chinese tech communities.
AI debt issuance surges as hyperscaler cash flows lag
AI capital expenditure is outpacing the free cash flows of major tech firms, forcing a reliance on debt markets and large-scale leasing commitments. This shift has heightened sensitivity to market sentiment, with Oracle’s stock falling after announcing a $20 billion capital raise.
Together AI targets inference layer with new model partnerships
Together AI is expanding its inference partnerships with Moonshot, Minimax, and Cursor while arguing that the inference layer is the primary source of value in AI. The company is also leveraging AI tools to accelerate kernel engineering, a process that requires deep domain expertise to remain effective.
DeepSeek delays V4 release to optimize for Huawei hardware
DeepSeek has prioritized inference efficiency on domestic Huawei chips over early deployment on Nvidia hardware, delaying the release of its V4 model. This strategic shift allows the company to achieve gross margins comparable to Anthropic despite significant performance gaps in the underlying silicon.
Altman argues local AI usage drives global memory shortages
OpenAI CEO Sam Altman stated that individuals running local AI models contribute to rising storage prices by underutilizing limited DRAM supply. He described this practice as inefficient from a global resource allocation perspective, while noting that OpenAI is developing robots to build self-replicating data centers.
Modal raises $750 million at $15.75 billion valuation
AI inference platform Modal closed a $750 million Series D led by Excel, reaching a $15.75 billion valuation after hitting $1 billion in annual recurring revenue. CEO Eric Bischoff stated the company is profitable on a product-line basis but faces negative cash flow due to heavy GPU capital expenditure.