Arena AI Hits $100M Revenue Mark as Anthropic Deploys Claude on NVIDIA Blackwell Ultra GPUs

Arena, the AI model evaluation platform spun out of UC Berkeley, has reached $100 million in annualized run-rate revenue just eight months after launching its commercial service, the company confirmed on June 29, 2026. The startup, which began as a research project in 2023, generates over 10 million user evaluations on its free crowdsourced leaderboard.

Arena's commercial service, AI Evaluations, launched in September 2025 and provides model labs and enterprises with deep-dive performance analytics. The company says its revenue milestone reflects surging demand for post-training optimization as AI providers strive to maximize model performance.

Crowdsourced evaluation becomes a business

Arena differentiates itself by letting users type a prompt sent to two anonymous models, then vote on which performed better. That crowdsourced data, now covering text, coding, vision, image generation, and agent-mode workflows, forms the backbone of its paid analytics platform.

• Arena raised $150 million Series A in January 2026 at a $1.7 billion valuation, with total funding of $250 million from investors including Felicis, Andreessen Horowitz, Kleiner Perkins, and Lightspeed Venture Partners.
• Co-founders include CEO Anastasios Angelopoulos, CTO Wei-Lin Chiang, and Ion Stoica, the UC Berkeley professor and Databricks co-founder.
• Annualized revenue grew from $30 million in January to $100 million in June, though CEO Angelopoulos noted the revenue is consumption-based, not traditional recurring subscription revenue.
• The company competes for the same dollar as human labeling startups Mercor, Surge, and Scale AI.

Anthropic and Qwen expand enterprise options

Separately, Anthropic's Claude models are now generally available on Microsoft Azure running NVIDIA GB300 Blackwell Ultra GPUs. The deployment, hosted in Microsoft Foundry, gives Azure-native enterprises access to Claude for building autonomous and domain-specific AI agents, according to NVIDIA's blog.

Meanwhile, Qwen's latest model, Qwen 3.6 27B, is gaining traction as a local development option. The model hit the front page of Hacker News on June 29, with developers praising its balance of size and capability for running on consumer-grade hardware, a sweet spot for prototyping and testing AI applications without cloud dependencies.

The convergence of crowdsourced evaluation, hyperscaler GPU access, and local-capable models underscores a fragmented but rapidly maturing AI infrastructure market. As Arena's Angelopoulos told TechCrunch, many still view the company as an open source project, underscoring the gap between perception and commercial reality.