February marked a period of significant growth for Hyperbolic, with record user adoption, expanded model usage, new partnerships, and product milestones that strengthened our position as an open-access AI cloud for developers, researchers, and AI teams.

Record-Breaking Growth

February was a month of exceptional expansion across our services.

We welcomed over 57,000 new users to the platform, a 78% increase from January, bringing our total user base to more than 100,000 developers and researchers.

Our inference services now support over 22,000 unique users, up 51%, who collectively processed more than 32 billion tokens, up 12%.

We facilitated over 3,600 GPU rentals, up 116%, totaling more than 64,000 hours of compute time, up 260%.

These numbers reflect growing demand for accessible, high-performance AI infrastructure that gives developers reliable compute and model access without the cost, complexity, or long procurement cycles of traditional providers.

Powering the Open-Source AI Revolution

The AI landscape changed significantly this month with the release of DeepSeek R1, a breakthrough model that quickly captured the attention of developers worldwide. Following its launch and onboarding to Hyperbolic, we experienced a 150% surge in inference usage as developers came to our platform to access the model.

Hyperbolic was among the first to make DeepSeek R1 available at BF16 precision, unlike providers that rely on FP8. This gives developers access to higher-quality model performance and lets them build with greater confidence. Our no data storage policy also allows teams to experiment with DeepSeek R1 without compromising control over their information.

Hyperbolic in the Spotlight

Andrej Karpathy's Shout Out

In one of the most validating moments of the month, former Tesla AI Director and OpenAI founding member Andrej Karpathy gave Hyperbolic a remarkable endorsement. In his latest deep-dive video on LLMs, Karpathy named Hyperbolic as his "favorite place to interact with base models," highlighting that we're "the only place that hosts my <3 Llama 3 405B Base, and in BF16 precision."

This recognition from one of the most respected figures in AI underscores our commitment to supporting the development of truly innovative AI applications by providing access to the most powerful base models at the highest quality.

Hugging Face Partnership

In another major milestone, Hyperbolic was selected as one of the few serverless inference providers on Hugging Face—joining an exclusive group chosen for performance, precision, and scale. This integration makes it even easier for developers to leverage our infrastructure, with the latest huggingface_hub Python library (v0.29.0) now supporting Hyperbolic.

Business Insider Coverage

Our mission to make AI infrastructure more accessible caught the attention of Business Insider, which featured Hyperbolic in an article about access to powerful AI models following the release of DeepSeek R1. The piece highlighted Hyperbolic’s role in connecting developers with advanced AI models while maintaining a focus on data security, performance, and verifiability.

LMSYS SGLang Partnership

LMSYS is a nonprofit advancing open-access AI research, developing large language models, and optimizing system performance. It collaborates with leading institutions and has pioneered projects like Vicuna LLM, FastChat, and SGLang, an advanced framework for high-performance model inference and system optimization.

We’re excited about our recent partnership with LMSYS to support SGLang for efficient large-scale model inference with DeepSeek V3 and R1.

SWE Arena: Revolutionizing Code Evaluation

This month, the Compute Intelligence Project launched SWE Arena, powered by Hyperbolic’s Inference Service. SWE Arena is an open evaluation platform designed for real-time coding evaluation and model comparison.

Unlike traditional frontend-focused coding applications, SWE Arena can execute a wide range of programs, allowing users to compare the coding capabilities of different AI models more accurately in real time.

Hyperbolic provides the performance and reliability that a complex evaluation platform requires to serve users across diverse use cases.

Fueling Autonomous AI Agents

In February, we continued to support partners building practical AI agents for engineering, developer workflows, and technical automation.

Soleng Partnership

Soleng, an autonomous solutions engineering and DevRel agent, is now powered by Hyperbolic’s compute, inference, and agent infrastructure. The integration enables Soleng to evaluate technical systems, generate recommendations, and scale autonomous workflows across complex developer environments.

With Hyperbolic’s infrastructure, Soleng can help teams analyze code, identify potential issues earlier, and automate technical review workflows that would otherwise require manual engineering time.

PMF or Die Agent Hackathon

To further accelerate the development of useful AI agents, we hosted the PMF or Die Hackathon
To further accelerate the development of useful AI agents, we hosted the PMF or Die Hackathon, challenging developers to build AI agents that generate revenue and deliver real-world utility. With $5,000 in bounties across three categories — Highest Revenue Generator, Most Creative Revenue Strategy, and Best Agent Framework Innovation — the hackathon reflected our commitment to supporting builders creating practical, outcome-driven AI applications.

Global Community Building

February also brought opportunities to connect with developers, researchers, and AI builders across the global technology community.

  • Hype(r)House at ETH Denver: We hosted a full day of workshops, panels, and networking focused on the intersection of AI and blockchain, featuring speakers from Solana, Optimism, NEAR Foundation, and more. Check out our live stream here.

  • Hype(r)House at Consensus Hong Kong: Our AI Workshops and panels brought together leaders from across AI landscapes to discuss the future of decentralized intelligence. Check out our live stream here.

March-ing On

As we move into March, we’re more committed than ever to building an open and accessible AI cloud for developers, researchers, and teams building with AI.

The growth we saw in February is just the beginning. With more open-source models being added regularly and continued improvements to our GPU compute and inference platform, Hyperbolic is helping more builders access the infrastructure they need to develop, test, and scale AI applications.

Join the 100,000+ developers already building with Hyperbolic at app.hyperbolic.ai.

About Hyperbolic

Hyperbolic is the Open-Access AI Cloud, giving researchers, startups, developers, and AI-native companies fast, flexible access to high-performance GPU capacity. The platform helps teams start on demand, scale programmatically, and grow into reserved infrastructure without long waitlists, rigid contracts, or complex procurement cycles.

Founded by award-winning math and AI researchers from UC Berkeley and the University of Washington, Hyperbolic is committed to making advanced AI infrastructure more accessible to builders around the world.

Website | X | LinkedIn | YouTube | GitHub | Documentation