[AINews] SpaceX is already a $28B/yr Neocloud

2026-06-23 · Source: Latent.Space - Www.latent.space · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Cloud Computing & IT Infrastructure, Cybersecurity & Data Privacy · Depth: Advanced, long

Summary

SpaceX has emerged as a "neocloud" provider, securing compute deals totaling \$2.32 billion per month, or \$28 billion annually, with Reflection AI (\$6.3 billion), Anthropic, and Google/xAI, implying Blackwell GPU pricing above \$10 per hour. Concurrently, Baseten announced a \$1.5 billion Series F, reinforcing the trend of enterprises seeking "owned intelligence" through post-trained open models. OpenAI expanded its Daybreak cyber program with GPT-5.5-Cyber, shifting to closed-loop patch generation, while Sakana introduced Fugu for learned model orchestration, though it faced immediate criticism for opaque benchmarks. GLM-5.2 is gaining traction as a frontier-adjacent open-weight model, ranking #3 on GDPval-AA at 1524 Elo and demonstrating cost-effective, robust performance in real-world agentic tasks. Google also promoted its Interactions API for Gemini agents, and Hermes continues to expand as a local agent platform, highlighting a broader shift towards evaluating agents as complex systems rather than simple chatbots.

Key takeaway

For AI Engineers evaluating model deployment strategies, the rise of "neocloud" providers like SpaceX and the strong performance of open-weight models like GLM-5.2 suggest a re-evaluation of compute sourcing. You should consider integrating frontier-adjacent open models into your agentic workflows, leveraging their cost-effectiveness and robust real-world performance. Additionally, explore local inference options for enhanced privacy and control, especially for high-utilization or sensitive workloads.

Key insights

The AI landscape is rapidly evolving with new "neocloud" compute providers, advanced open-weight models, and sophisticated agent orchestration systems.

Principles

• Compute capacity is a strategic asset.
• Open-weight models are reaching frontier capabilities.
• Agent evaluation requires system-level metrics.

Method

Sakana Fugu proposes a method of learned orchestration for AI models, involving selection, delegation, verification, and synthesis across multiple frontier models via a single API.

In practice

• Evaluate agent systems with real-world harnesses.
• Explore local inference for privacy and control.

Topics

• AI Compute Infrastructure
• Open-weight Models
• AI Agents
• Model Orchestration
• Cybersecurity AI
• Local Inference

Best for: Machine Learning Engineer, NLP Engineer, CTO, Director of AI/ML, AI Engineer, Investor

Related on AIssential

• not much happened today — Local-first AI models and self-improving agents are disrupting cloud-dependent services and driving demand for open data and specialized hardware.
• Key Changes in AI and Space Exploration — The AI industry is rapidly evolving with significant investments, regulatory shifts, and infrastructure challenges.
• not much happened today — AI development is shifting towards multi-cloud distribution, cost-aware agentic workflows, and specialized hardware for inference.
• SpaceX inks compute deal with Reflection AI, an open source AI lab — Open-weight AI labs are securing significant compute infrastructure, challenging closed models and leveraging available data center capacity.
• We’re feeling cynical about xAI’s big deal with Anthropic — xAI's deal to sell its compute capacity to Anthropic signals a strategic shift from AI model development to infrastructure provision.
• The AI Race Just Got a SpaceX-Sized Twist — AI competition is shifting from raw model capability to productized agent orchestration and ambitious compute infrastructure.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Latent.Space - Www.latent.space.