Hugging Face

Hugging Face is the principal global open-source artificial intelligence platform and developer community, founded in 2016 by Clément Delangue, Julien Chaumond, and Thomas Wolf. The company is headquartered in Paris with offices in New York and operates the Hugging Face Hub, the world's largest repository of open-weights AI models, datasets, training code, and AI applications. Hugging Face is the principal distribution channel for open-weights releases from Meta AI / FAIR, Mistral AI, DeepSeek, Alibaba Qwen, Allen Institute for AI, and other open-AI organizations, and develops its own SmolLM, SmolVLM, and earlier BLOOM model lines through its in-house research organization. In February 2026, Hugging Face acquired GGML.ai, bringing the llama.cpp local-inference runtime and the GGUF quantization format into the company.

At a glance

• Founded: 2016 in New York City by Clément Delangue, Julien Chaumond, and Thomas Wolf, all French AI entrepreneurs. Headquarters relocated to Paris with continued New York presence.
• Status: Private. Series D in August 2023 valued the company at $4.5 billion. Reported cash position from earlier rounds; no recent disclosed Series E as of April 2026.
• Funding: Approximately $400 million raised in cumulative private rounds. Series D of $235 million in August 2023, led by Salesforce, valued the company at $4.5 billion. Reported approximately half of the $400 million remained on the balance sheet as of late 2025.
• CEO: Clément Delangue (co-founder; CEO since founding).
• Other notable leadership: Julien Chaumond (co-founder, Chief Technology Officer), Thomas Wolf (co-founder, Chief Science Officer; co-author of the Hugging Face Transformers library and "Natural Language Processing with Transformers" textbook). Georgi Gerganov joined as senior leader for the Local AI organization following the February 2026 GGML.ai acquisition.
• Open weights: Yes. Hugging Face is the principal global open-weights distribution platform, with the Hub hosting hundreds of thousands of open-weights models, datasets, and AI applications.
• Flagship products: Hugging Face Hub (the platform), Transformers library, SmolLM3 (small efficient language models), SmolVLM (small vision-language models), llama.cpp and the GGUF format (post-February-2026 acquisition).

Origins

Hugging Face was founded in 2016 by Clément Delangue, Julien Chaumond, and Thomas Wolf, three French entrepreneurs who started the company in New York City as a chatbot for teenagers branded as a virtual friend. The original consumer-product framing was abandoned within two years as the founders pivoted toward building open-source NLP infrastructure for the broader machine-learning community. Delangue had previously been at Moodstocks (acquired by Google) and at eBay; Chaumond had a software-engineering background; Wolf brought research credentials from the academic NLP community.

The pivotal moment came with the November 2018 release of the Transformers library, an open-source implementation of the BERT architecture and other transformer-based language models. The library quickly became the de-facto standard for working with transformer models in the academic and industry research communities, and Hugging Face built the Hub platform around it as the distribution channel for community-shared models.

The 2019 to 2022 period saw Hugging Face become the principal global open-source AI platform. Funding rounds expanded the company's resources for platform development, research, and ecosystem support. The Hub grew to host tens of thousands of community-contributed models. Hugging Face also led BigScience, the international research collaboration that produced the BLOOM 176-billion-parameter multilingual model in 2022.

The 2023 Series D of $235 million, led by Salesforce with participation from Google, NVIDIA, Amazon, AMD, IBM, Intel, and Qualcomm, valued the company at $4.5 billion and provided capital for platform growth. The investor mix combined US technology incumbents and US AI-and-cloud strategic partners, an unusually broad strategic-investor base.

The 2024 to 2026 period saw Hugging Face emphasize in-house model research alongside the Hub platform. The SmolLM series launched in July 2024 with 135M, 360M, and 1.7B parameter open-weights variants, framed as "blazingly fast and remarkably powerful" small models for local-device deployment. SmolLM2 in December 2024 and SmolLM3 in 2025 expanded the line, with SmolLM3 supporting 6 languages, long context, and strong function calling. SmolVLM brought small-vision-language capability to the family.

The February 20, 2026 acquisition of GGML.ai was a structurally significant transaction. GGML.ai, founded by Georgi Gerganov, develops the ggml C tensor library, llama.cpp (the inference runtime used by millions of developers for running large language models on consumer hardware), whisper.cpp (the equivalent for speech-to-text), and the GGUF quantization format that has become the de-facto standard for local AI deployment. The acquisition consolidated under Hugging Face the principal local-AI infrastructure stack, with all components remaining open-source under their existing MIT licenses.

In late 2025, CEO Clem Delangue gained attention for publicly characterizing the AI industry as in an "LLM bubble" rather than a broader AI bubble, with Delangue arguing that AI value creation extends well beyond chatbots and that LLM-specific valuations may face a 2026 correction. The framing reinforced Hugging Face's positioning as the principal platform for AI diversity beyond the LLM-centric narrative.

Mission and strategy

Hugging Face's stated mission is to "democratize good machine learning." The framing has been remarkably consistent since the company's pivot in 2018, with the Transformers library and the Hub platform as the principal mechanisms for the democratization premise.

The strategy combines five threads. First, the Hugging Face Hub as the global open-source AI platform, hosting models, datasets, training code, and AI applications across the full diversity of AI research and applications. Second, open-source software libraries (Transformers, Datasets, Tokenizers, Diffusers, PEFT, Accelerate, and other) that provide the practical infrastructure for working with AI models. Third, in-house model research through the SmolLM, SmolVLM, and earlier BLOOM lines, demonstrating the platform's capability and contributing models to the open-weights ecosystem. Fourth, enterprise services through Hugging Face Enterprise and the Inference Endpoints platform, providing paid commercial services for the platform's enterprise users. Fifth, post-February-2026 the Local AI organization through llama.cpp, GGML, whisper.cpp, and the GGUF format, providing the local-inference infrastructure complementing the Hub's cloud distribution.

The competitive premise is that open-source AI is the structurally important contribution to the field that closed-source AI cannot fully replace. The Hub's network effects (model creators upload to the Hub because users are there; users download from the Hub because models are there) produce a structural moat, and the in-house Smol model line plus the local-AI infrastructure expand Hugging Face's value-add beyond pure platform hosting.

The "LLM bubble" framing publicly articulated by Delangue aligns with the strategic positioning that Hugging Face benefits from AI diversity (small models, local inference, multimodal, scientific AI) rather than from the closed-weights LLM scale-economics race.

Models and products

• Hugging Face Hub. Global open-source AI platform hosting hundreds of thousands of models, tens of thousands of datasets, and tens of thousands of community-built AI applications (Spaces). The principal distribution channel for open-weights models from across the AI research community.
• Transformers library. Open-source Python library for working with transformer-based AI models. The de-facto standard for academic and industry transformer research.
• Datasets, Tokenizers, Diffusers, PEFT, Accelerate, TRL. Open-source libraries covering specific aspects of the AI development workflow.
• SmolLM3. Latest small-language-model release. 3-billion-parameter dual-reasoning model with 6-language support, long context, and strong function calling.
• SmolLM2 and SmolLM. Earlier small-language-model variants spanning 135M, 360M, and 1.7B parameter scales.
• SmolVLM. Small vision-language models including a 256-million-parameter variant.
• BLOOM. 176-billion-parameter multilingual language model produced through the BigScience collaboration in 2022. The first publicly available multilingual model at this scale.
• Idefics. Multimodal model line covering 9-billion and 80-billion parameter variants released in 2023.
• Zephyr. Instruction-tuned variants distributed open-weights.
• llama.cpp and ggml. Local-AI-inference infrastructure post-February-2026 acquisition. Powers consumer-hardware deployment of open-weights models worldwide.
• whisper.cpp and GGUF. Local speech-to-text and the GGUF quantized-model format.
• Inference Endpoints and Hugging Face Enterprise. Paid commercial services for enterprise customers.

The principal commercial channels are Hugging Face Enterprise (paid platform tier), Inference Endpoints (managed inference), and direct enterprise sales for hosted-Hub deployments and AI infrastructure.

Benchmarks and standing

Hugging Face's in-house model line emphasizes small-and-efficient deployment characteristics rather than frontier-tier benchmark leadership. SmolLM3 reports competitive performance on small-model benchmarks among 3-billion-parameter models, with the differentiating capabilities being multilingual coverage, long context, and function-calling support.

The Hub platform's standing is overwhelming: Hugging Face is widely characterized as the principal global open-source AI platform, with no peer organization at the same scale of model and dataset hosting, community engagement, and developer-tooling integration. Industry coverage frequently characterizes the Hub as the GitHub of AI, although Hugging Face's research-and-models-and-datasets focus is structurally different from GitHub's code-and-repositories focus.

The post-February-2026 GGML.ai acquisition consolidated Hugging Face's standing in the local-AI-inference market. llama.cpp had been the most-used open-source local-inference runtime; bringing it under the Hugging Face umbrella consolidates the platform's central role across cloud-and-local AI.

Leadership

As of April 2026, Hugging Face's senior leadership includes:

• Clément Delangue, Chief Executive Officer and co-founder. Public face for Hugging Face on platform strategy, AI policy, and the broader open-source AI movement. Has been particularly visible in 2025 to 2026 with public statements on the "LLM bubble" framing.
• Julien Chaumond, Chief Technology Officer and co-founder. Senior technology leadership for the Hub platform and other infrastructure.
• Thomas Wolf, Chief Science Officer and co-founder. Senior research leadership; co-author of "Natural Language Processing with Transformers" and senior figure in the global AI research community.
• Georgi Gerganov, senior leader of the Local AI organization following the February 2026 GGML.ai acquisition. Creator of llama.cpp and the GGML / GGUF infrastructure.

The company has hired aggressively across research, engineering, product, and policy roles, with offices in Paris, New York, and other locations. Senior research-team leadership includes Hugging Face researchers contributing to the SmolLM line, the Transformers library development, and other research programs.

Funding and backers

Hugging Face's funding history includes approximately $400 million in cumulative private rounds. The Series D of $235 million in August 2023, led by Salesforce, brought the valuation to $4.5 billion and is the most recent major round. Earlier rounds included a Series C of $100 million in May 2022, a Series B of $40 million in March 2021, and Series A and earlier rounds from the company's founding period.

The investor base is unusually broad in strategic alignment. Series D participants included Salesforce (lead), Google, NVIDIA, Amazon, AMD, IBM, Intel, Qualcomm, and earlier investors Sequoia Capital, Coatue, Lux Capital, Addition, and Betaworks. The diverse strategic investor mix reflects Hugging Face's positioning as a platform for the broader AI ecosystem rather than as a competitor to any single large investor.

The reported approximately half of the $400 million remaining on the balance sheet at late 2025 indicates strong capital efficiency and reduced dependence on near-term fundraising. The company has not announced a Series E as of April 2026.

Industry position

Hugging Face occupies a structurally distinctive position as the central global open-source AI platform. The combination of the Hub platform's network-effect moat, the open-source software libraries, the in-house model research, the post-February-2026 local-AI-infrastructure consolidation through the GGML.ai acquisition, and the broad strategic-investor support produces a profile that no other AI organization matches across all dimensions.

Industry coverage has consistently characterized Hugging Face as the most strategically important open-source AI organization and as the principal counterweight to closed-weights frontier labs in the AI ecosystem. The "LLM bubble" framing publicly articulated by Delangue in late 2025 reinforced the platform's strategic positioning.

Strategic risks include the dependence on continued open-weights model releases from third-party labs (the Hub's value is anchored by external model releases), competition from cloud-platform alternatives (AWS, Google Cloud, Azure offering managed AI infrastructure), and the open question of whether the in-house Smol model line and Local AI infrastructure can grow into a meaningful commercial business beyond platform-fee revenue. Strategic strengths include the network-effect moat, the founder-team and senior-leadership stability, the strong capital position, and the breadth of open-source software contribution.

Competitive landscape

Hugging Face is structurally a platform that competes with and collaborates with most other AI organizations. The principal competitive and collaborative relationships include:

• Allen Institute for AI, EleutherAI, LAION, BigScience, MILA, Nous Research. Peer open-AI-research organizations, with collaboration through Hub-based distribution and joint research initiatives.
• Meta AI / FAIR, Mistral AI, DeepSeek, Alibaba Qwen, Cohere. Commercial open-weights model providers. Hugging Face hosts their releases on the Hub; the model providers benefit from platform distribution.
• OpenAI, Anthropic, Google DeepMind. Closed-weights frontier labs. Less direct platform competition; Hugging Face hosts adjacent open-source materials related to these organizations' research.
• GitHub. Indirect competitor as a code-hosting platform for AI projects, though Hugging Face's models-and-datasets focus is structurally distinct.
• Cloud-platform AI services (AWS SageMaker, Google Vertex AI, Azure ML). Compete with Hugging Face Inference Endpoints for managed-AI-inference customers.
• AI-development-tool companies including Replicate, Modal, Together AI, and other organizations. Compete in specific submarkets within the broader Hugging Face footprint.

Outlook

Several open questions affect Hugging Face's trajectory in 2026 and 2027:

• The integration of GGML.ai and the development of the post-acquisition Local AI organization, including any new product directions combining the Hub and llama.cpp / GGUF infrastructure.
• The SmolLM4 release timing and capability profile; sustaining the SmolLM3 dual-reasoning capability is the central technical question for the in-house model line.
• A potential Series E funding round and any further valuation movement, particularly given the 2025 to 2026 public-listing wave (Z.ai, MiniMax) that may pressure private-company valuations.
• The development of Hugging Face Enterprise and Inference Endpoints commercial revenue.
• The platform's response to potential AI-policy and AI-content-regulation evolution in the United States and Europe.
• Continued senior-talent recruitment across research, platform, and Local AI organizations.
• The validation or rejection of Delangue's "LLM bubble" framing through 2026 and 2027 capital-markets dynamics.

Sources

• Hugging Face Blog: GGML and llama.cpp join HF to ensure the long-term progress of Local AI. February 2026 acquisition announcement.
• Hugging Face Blog: SmolLM3. 2025 SmolLM3 release.
• Hugging Face Blog: SmolVLM. SmolVLM release.
• TheAIInsider: Hugging Face CEO Warns of an Emerging "LLM Bubble". November 2025 LLM-bubble framing.
• SitePoint: GGML Joins Hugging Face: What This Means for Local Model Optimization. GGML acquisition strategic context.
• Synthedia: Hugging Face Lands $235 Million in Funding and a Giant New Valuation. August 2023 Series D.
• Wikipedia: Hugging Face. Comprehensive company history reference.