October and November brought meaningful shifts in how the major companies are building and allocating AI infrastructure. OpenAI signed a $38b agreement with AWS to secure long‑term compute capacity and expand beyond Azure, marking its first major diversification of cloud supply. Anthropic outlined a $50b plan for new U.S. data centers in Texas and New York focused on long context and high‑volume inference. CoreWeave accelerated expansion with new multi‑gigawatt capacity commitments and a $1.17b storage partnership with VAST Data. Across AWS, Microsoft, and Google, the most important updates centered on lowering inference cost, improving governance and evaluation tooling, and making deployments more predictable for enterprises; several leaders also said the power grid, not chips, is now the bottleneck.

Startups operating across infrastructure, training, inference, silicon, and agent tooling also raised meaningful capital. The pattern is straightforward: secure compute, commit to power, build physical infrastructure, and ship the governance layers enterprises expect before rolling out agents.

🛠️ Spotlight releases

  • OpenAI shifted the balance of its cloud footprint by signing a multi‑year compute agreement with AWS. The deal reserves substantial accelerator and CPU capacity specifically for OpenAI workloads, its first major future capacity outside Azure. (Reuters)
  • Anthropic announced a $50b build‑out of data centers in Texas and New York designed for sustained memory and long‑context inference. (Anthropic)
  • CoreWeave strengthened its position as the scaled independent GPU cloud by signing a $1.17b storage partnership with VAST Data and committing to multi‑hundred‑megawatt phases at a new 2 GW Texas campus where it serves as the anchor tenant. (Reuters)
  • AWS activated its Project Rainier supercluster with ~500k Trainium2 chips and continued to push Inferentia2 for cost‑efficient inference. The net effect: broader, cheaper, and more predictable access paths for large workloads. (Data Center Dynamics)
  • Microsoft focused on enterprise readiness in Copilot Studio by adding pre‑deployment evaluation, monitoring, and governance controls in October updates and subsequent rollouts. (Microsoft)
  • Google increased TPU availability (v5e/v5p/Trillium, with Ironwood slated for late 2025) and shipped Gemini features across Workspace that emphasize retrieval quality, policy controls, and efficiency. (Google Cloud)
  • Databricks continued to wire agent steps into its data environment via Agent Bricks and the Multi‑Agent Supervisor, reducing friction between model output and production jobs. (Databricks Documentation)

💰 Funding activity

  • d‑Matrix raised $275m (Series C) at a $2b valuation to scale its memory‑centric inference hardware.
  • Fireworks AI raised $250m (Series C) at a $4b valuation to expand high‑volume inference infrastructure.
  • Modal raised $87 m in a Series B round at a $1.1 b valuation to grow its serverless compute platform for large scale model and agent workloads.
  • LangChain raised $125m at a $1.25b valuation to expand its agent framework, including tooling for memory, evaluation, and structured workflows used inside enterprise applications.

🔍 Notable developments across the stack

  • Mistral launched Mistral AI Studio (private beta) to make building on its open‑weight catalog easier for enterprises; focus areas included operational observability and governance.
  • Together AI expanded training and inference services emphasizing predictable latency and transparent pricing, including earlier Batch Inference API updates that cut large‑scale costs.
  • Lambda announced a multi‑billion‑dollar agreement with Microsoft to deploy tens of thousands of NVIDIA GPUs (including GB300 NVL72), a direct expansion of U.S. capacity outside hyperscalers.
  • Groq published fresh low‑latency inference results relevant to real‑time agent loops, underscoring its LPU differentiation.
  • Scale AI and Snorkel advanced evaluation tooling: Snorkel co‑released Terminal‑Bench 2.0 for agent evaluation, while Scale updated enterprise evaluation offerings and methods for stabilizing noisy judge variance.
  • Hugging Face expanded evaluation and agent tooling with RTEB (retrieval benchmark) and the OpenEnv agentic environment hub, reinforcing its role as the primary distribution layer for open‑weight development.

🧠 What matters going into 2026

The past few months have made one thing clear: AI is entering an operational phase where infrastructure decisions matter more than model announcements. The companies shaping the market are securing power, expanding data-center capacity, and putting governance and evaluation tooling in place because these are now the gating factors for enterprise adoption. Conversations with buyers consistently center on predictable cost, dependency management, deployment safety, and the ability to plug agents into existing systems without introducing new operational risk.

The companies with real infrastructure footprints are beginning to define the standards for everyone else. They control the variables that matter at scale: availability, latency, unit economics, and deployment trust. For startups, the opportunity is increasingly tied to these constraints. Lowering inference cost, reducing operational overhead, and aligning with enterprise integration paths have become the fastest routes to revenue. New model development still plays a role, but the strongest commercial pull is going to teams that can make agents dependable, observable, and easy to adopt.

As we head into 2026, the competitive edge will come from execution against infrastructure realities rather than theoretical capability. The winners will be the companies that internalize power, cost, governance, and integration as first-order requirements and build for the environments where AI is actually being deployed.

I’m a Principal at Work-Bench, a Seed stage enterprise-focused VC fund based in New York City. Our sweet spot for investment at Seed correlates with building out a startup’s early go-to-market motions. In the cloud-native infrastructure and developer tool ecosystem, we’ve invested in companies like Cockroach Labs, Run.house, Prequel.dev, Autokitteh and others.