The Expansion

Hugging Face expanded its free GPU tier on Spaces in May 2025, making NVIDIA T4 GPUs (16GB VRAM) available for free with a 48-hour session limit per Space. Previously, free Spaces only offered CPU execution — sufficient for lightweight models but inadequate for running modern LLMs, image generation models, or video AI in real time. The T4 tier change meant that models like Llama 3.3 7B, Stable Diffusion XL, and Whisper Large could be hosted and demoed publicly without cost.

Spaces as the AI App Store

Hugging Face Spaces functions as the de facto app store for open-source AI. As of May 2025, over 350,000 public Spaces existed — each a shareable web interface for an AI model or tool. Spaces could be built with Gradio, Streamlit, or static HTML, deployed from a simple repository structure, and shared as a permanent URL. The addition of free GPU meant the most impactful open-source models became immediately accessible to anyone with a browser.

The Compute Economics

Hugging Face's free GPU offering was made possible by its Series D funding and strategic partnerships with infrastructure providers. The company was explicit that the free tier was designed to drive adoption and premium upgrades — Zero GPU (the Spaces GPU system) with persistently allocated GPU time was available on the Pro plan at $9/month, making it affordable for serious developers who needed more than the shared free tier.

Impact on Open Source AI

The broader impact was accelerating open-source AI adoption. Researchers could now deploy demos alongside papers on the same day of publication, allowing peer reviewers and the public to interact with results rather than just reading about them. This shift toward "demo or it didn't happen" was increasingly shaping which models gained community momentum.

What This Means for Indian Businesses

Hugging Face Spaces is how most Indian AI researchers and developers currently share their work publicly. The expanded free GPU tier means Indian AI developers — especially those building demos for research papers, hackathons, or portfolio projects — can host computationally intensive models without personal cloud bills. This is particularly significant for the IIT/IISc research community, where compute budget constraints often limit what can be demonstrated publicly.