March 10, 2026 — I’m leading with what’s new and material: a record funding round, fresh carrier and storage builds, and on-chain agentic tooling. Collectively, these extend last week’s agent-first infrastructure momentum into hard deployment and capital scale.
🧭 Key Highlights
💰 Nscale raises $2B — European record for AI infrastructure
🔗 HPE unveils 1.6T AI connectivity for GPU clusters
📡 SoftBank outlines Telco AI Cloud for “Physical AI”
🔬 Andrej Karpathy open-sources autoresearch for AI-driven research
⛓️ Injective delivers MultiVM optimizations and on-chain agentic tools
Capital & Core Infrastructure
💰 Nscale raises $2B Series C to expand vertically integrated AI infrastructure
According to Siliconrepublic, Nscale raised $2 billion (€1.7 billion) Series C to expand vertically integrated AI infrastructure across Europe, North America, and Asia, spanning GPU compute, networking, data services, and orchestration.
This marks a European record for AI infrastructure funding, validating vertically integrated models spanning the full stack.
💾 Coredge adopts Lightbits software-defined NVMe/TCP storage for AI workloads
According to Businesswire, Coredge adopted Lightbits software-defined NVMe/TCP storage on OpenShift in India to power AI training and inference workloads.
Software-defined storage over standard TCP/NVMe reduces hardware dependencies for AI infrastructure.
🔗 HPE unveils 1.6T AI connectivity solutions at MWC 2026
According to Markets, HPE unveiled 1.6T AI connectivity solutions at MWC 2026 to relieve networking bottlenecks in large-scale GPU clusters.
Higher-speed networking directly addresses the communication bottleneck in distributed training and inference.
📡 SoftBank outlines Telco AI Cloud vision with AI-RAN MEC
According to Thefastmode, SoftBank outlined a Telco AI Cloud vision using AI-RAN MEC for in-network model deployment and “Physical AI,” with a proof-of-concept with Ericsson for robot connectivity.
Carriers are positioning networks as AI infrastructure, not just transport.
Open Source & Research Infra
🔬 Andrej Karpathy open-sources autoresearch for AI-driven research loops
According to Radicaldatascience, Andrej Karpathy open-sourced autoresearch for AI-driven research loops on small, single-GPU LLM setups.
Democratizing AI research infrastructure enables broader experimentation outside large-scale compute environments.
🎨 AIMomentz launches image-model evaluation platform
According to Unifuncs, AIMomentz launched an image-model evaluation platform featuring “AI Models That Can Die” and CAP-SRP to track refusals for safety benchmarking.
New evaluation frameworks address the lack of standardized safety and refusal metrics for image models.
Decentralized & Agent Infrastructure
⛓️ Injective’s Mainnet Upgrade delivers MultiVM and on-chain agentic tools
According to X, Injective’s Mainnet Upgrade delivered MultiVM optimizations and onchain agentic tools for decentralized AI execution.
Blockchain infrastructure is evolving to support AI agents executing on-chain.
🌐 aZen details AI-native DePIN environment for distributed AI
According to X, aZen detailed an AI-native DePIN environment to coordinate compute, contributors, and infra for distributed AI.
DePIN models are expanding beyond storage/bandwidth into AI compute coordination.
🚀 Based Pages: Agent-native deployment continues momentum
According to X, Based Pages continues the agent-native deployment thread by turning prompts into live subdomains with x402 settlement.
Agent-to-production deployment without DevOps loops is gaining traction.
👛 Keystrata surfaces as AI-native wallet infrastructure
According to X, Keystrata surfaced as AI-native wallet infrastructure following Privy’s acquisition by Stripe.
Agent-native financial infrastructure is emerging to support autonomous transactions.
Enterprise Traction & Headwinds
🏦 Axiom Trust emerges from stealth with $11.8M for AI-native trust platform
According to X, Axiom Trust emerged from stealth with $11.8M to build an AI-native trust administration platform.
Domain-specific AI-native platforms are emerging beyond horizontal infrastructure.
📈 AI-native infra adoption outpaces prior dev tooling waves
According to X, commentary noted AI-native infra adoption outpacing prior dev tooling waves, forecasting agent-built apps.
The shift to agent-built applications is accelerating infrastructure adoption.
⚠️ Oracle and OpenAI reportedly pause Texas AI datacenter expansion
According to Computing, Oracle and OpenAI reportedly paused plans to expand their Abilene, Texas AI datacenter amid financing and demand disagreements.
Even record funding faces real-world constraints around capital allocation and demand forecasting.
🔍 Infra Insights
Today’s core trends: Record capital validates vertical integration, Carriers position networks as AI infra, Agent infrastructure extends to payments and deployment, Real-world constraints temper expansion.
Nscale’s $2B raise marks a European record for AI infrastructure, validating vertically integrated models that span GPU compute, networking, data services, and orchestration. This isn’t just capital for GPUs — it’s funding full-stack control, from hardware to software to operations. The vertical model addresses fragmentation but requires massive scale to justify.
HPE’s 1.6T AI connectivity and SoftBank’s Telco AI Cloud represent carrier-grade infrastructure catching up to AI workloads. As GPU clusters scale, networking becomes the bottleneck — 1.6T interconnects and in-network model deployment via AI-RAN MEC acknowledge that AI infrastructure isn’t just servers, it’s the fabric connecting them.
Karpathy’s autoresearch and AIMomentz’s evaluation platform signal maturing tooling for AI research and safety. Democratizing research loops and standardizing refusal metrics address foundational gaps in how we build and evaluate AI systems.
Injective’s MultiVM, aZen’s DePIN, Based Pages, and Keystrata collectively extend agent infrastructure from coordination (yesterday’s SwarmBase, Qubic) into execution, deployment, and payments. The agent stack is filling out: compute → coordination → deployment → payments.
Oracle and OpenAI’s paused Texas expansion is a reality check: even record funding faces constraints around capital allocation, demand forecasting, and infrastructure timing. Not all planned capacity will be built, and not all capacity built will be utilized efficiently.
Together, these developments show AI infrastructure moving beyond initial hype into sustained build-out — with real capital, real constraints, and real technical challenges across networking, storage, compute, and agent operations.