FriendliAI Closes $20 Million Seed Extension Funding Round

Listen to this article

FriendliAI, a Redwood City, California-based AI inference platform startup, secured $20 million in a seed extension round. The round was led by Capstone Partners, with participation from new investors Sierra Ventures, Alumni Ventures, KDB, and follow-on investment from KB Securities. This funding builds on a $6 million seed round raised in 2021, also led by Capstone Partners. The capital is aimed at accelerating FriendliAI’s growth in the generative AI inference market, particularly in North America and Asia, by enhancing its technology and expanding its customer base.

Strategic Use of Funds

The $20 million will be allocated to several key areas:

• Sales and Marketing Expansion: FriendliAI plans to bolster its sales and marketing teams to accelerate customer acquisition, targeting enterprises transitioning AI models from experimentation to production.
• Operational Scaling: The funds will support scaling operations to meet growing demand, particularly from large clients like SKT and NextDay AI, who rely on FriendliAI’s platform for cost-efficient AI inference.
• Product Development: Investments will enhance the Friendli Engine, a software platform that optimizes AI inference workloads, reducing GPU costs by up to 90% and improving response times.
• Hardware Procurement: The company will acquire additional GPUs to support its cloud services, ensuring scalability and performance for its clients.

Founded in 2021 by CEO Byung-Gon Chun at Seoul National University, FriendliAI specializes in generative AI infrastructure, focusing on efficient deployment and scaling of large language models (LLMs) and multimodal models. Its flagship Friendli Engine leverages proprietary technologies like Iteration Batching (also known as Continuous Batching), Friendli DNN Library, Friendli TCache, and Native Quantization to achieve significant cost savings (50-90%) and performance improvements (up to 10.7x higher throughput and 6.2x lower latency compared to alternatives like vLLM). The company serves 25 to 30 large clients, including South Korea’s SKT, which reported 5x higher LLM throughput and 3x cost savings, and NextDay AI, which reduced GPU usage by over 50%.

Market Context and Competitive Advantage

The AI inference market is growing rapidly as organizations shift from AI development to production deployment. FriendliAI estimates that 80-90% of GPU resources are dedicated to inference, highlighting the critical need for cost-effective and high-performance solutions. The company’s platform addresses key challenges such as high infrastructure costs, slow inference speeds, and the complexity of managing LLMs in production. Strategic partnerships with LG AI Research (for the EXAONE 4.0 model) and Hugging Face (for deploying open-source models) enhance FriendliAI’s market reach and technological credibility.

Financial Outlook

While FriendliAI is not yet profitable, it maintains strong gross margins and anticipates significant revenue growth in 2025, projecting a 6x to 7x increase over 2024. This growth is driven by the adoption of its Friendli Suite, which includes Dedicated Endpoints, Serverless Endpoints, and Friendli Container, catering to diverse enterprise needs. The company’s focus on cost efficiency and performance has attracted clients like Scatter Lab and Upstage, further solidifying its market position.

FriendliAI, a generative AI infrastructure company headquartered in Redwood City, California, with roots in Seoul, South Korea, announced a $20 million seed extension round. This funding round marks a significant milestone for the company, which has positioned itself as a leader in optimizing AI inference for large language models (LLMs) and multimodal models. Below is a detailed analysis of the funding round, its implications, and FriendliAI’s strategic positioning in the AI inference market.

Funding Round Details

The $20 million seed extension round was led by Capstone Partners, a prominent investment bank, with participation from new investors Sierra Ventures, Alumni Ventures, KDB, and follow-on investment from KB Securities. This follows a $6 million seed round in 2021, also led by Capstone Partners, bringing FriendliAI’s total funding to approximately $26 million. While the company declined to disclose its valuation, CEO Byung-Gon Chun confirmed it has increased since the 2021 round, reflecting investor confidence in FriendliAI’s growth potential. The funding was announced across multiple reputable sources, including Crunchbase, SiliconANGLE, and Finsmes, ensuring the reliability of the information.

Strategic Allocation of Capital

FriendliAI plans to deploy the $20 million across several strategic initiatives to strengthen its market position and scale its operations:

• Sales and Marketing Expansion: The company aims to accelerate customer acquisition by expanding its sales and marketing teams, particularly targeting enterprises in North America and Asia. This focus aligns with the growing demand for AI inference solutions as businesses transition from experimental AI to production-grade deployments.
• Operational Scaling: With a client base of 25 to 30 large enterprises, including South Korea’s leading telecom provider SKT and NextDay AI, FriendliAI will scale its operations to support increasing demand. This includes enhancing its cloud infrastructure to handle high-throughput inference workloads.
• Product Development: A significant portion of the funds will be invested in improving the Friendli Engine, FriendliAI’s proprietary software platform that optimizes AI inference. The platform’s key technologies—Iteration Batching, Friendli DNN Library, Friendli TCache, and Native Quantization—enable up to 90% GPU cost savings, 10.7x higher throughput, and 6.2x lower latency compared to competitors like vLLM. These metrics were validated in testing conducted by FriendliAI in October 2023 using Llama-2-13B and AWQ-ed Mixtral 8x7B models on NVIDIA A100 80GB GPUs.
• Hardware Procurement: To support its cloud services, FriendliAI will procure additional GPUs, ensuring scalability and performance for its clients. This is critical as the company serves high-volume clients like NextDay AI, which processes over 3 trillion tokens per month.

Recommended: Nuro Raises $203 Million In Series E Funding Round

FriendliAI’s Technology and Market Differentiation

Founded in 2021 by Byung-Gon Chun, a researcher with over a decade of experience in accelerating AI model execution at Seoul National University, FriendliAI has developed a robust platform for generative AI inference. The Friendli Suite comprises three core offerings:

1. Friendli Dedicated Endpoints: Provides high-performance inference with dedicated GPU resources, auto-scaling capabilities, and a user-friendly dashboard for monitoring endpoint performance. This is ideal for enterprises with stringent service-level agreements (SLAs).
2. Friendli Serverless Endpoints: Enables rapid deployment of AI models with built-in tools like web search and Python interpreters, supporting side-by-side model comparisons for optimal selection.
3. Friendli Container: Allows clients to run models within their own infrastructure, ensuring data privacy and seamless integration with systems like Kubernetes, AWS EKS, and SageMaker.

FriendliAI’s technology addresses critical challenges in AI inference, including high infrastructure costs, slow inference speeds, and the complexity of managing LLMs in production. Its proprietary Iteration Batching (Continuous Batching) optimizes dynamic and uneven LLM inference traffic, allowing faster processing and reduced costs. The Friendli DNN Library and TCache further enhance performance by optimizing GPU kernels and reusing computational results, while Native Quantization (supporting FP8, INT8, and AWQ) ensures efficient serving without sacrificing accuracy.

Market Context and Competitive Landscape

The AI inference market is experiencing explosive growth, with FriendliAI estimating that 80-90% of GPU resources are allocated to inference rather than training. This shift underscores the importance of cost-effective and high-performance inference solutions. FriendliAI competes with platforms like vLLM but claims superior performance, as demonstrated by its 10.7x higher throughput and 6.2x lower latency in controlled tests. The company’s strategic partnerships with LG AI Research (exclusive provider of API access to the EXAONE 4.0 model) and Hugging Face (for deploying open-source models) enhance its ecosystem and market reach. Additionally, FriendliAI’s integration with Weights & Biases and AWS partnerships strengthens its position as a production-grade AI infrastructure provider.

Financial Performance and Growth Outlook

FriendliAI is not yet profitable but maintains strong gross margins, prioritizing efficient scaling over immediate profitability. The company projects a 6x to 7x revenue increase in 2025 compared to 2024, driven by the accelerating adoption of generative AI in production environments. Its client base, including Scatter Lab (which reduced infrastructure costs for its Zeta LLM service) and Upstage (leveraging FriendliAI for its Solar foundation model), demonstrates the platform’s value in delivering cost savings and performance gains. With 25 to 30 large clients and a reported 600% revenue growth potential in 2025, FriendliAI is well-positioned to capitalize on the growing demand for AI inference solutions.

Notable Achievements and Recognition

FriendliAI has garnered significant industry recognition, including selection as one of 50 “Baby Unicorn” companies by the South Korean Ministry of SMEs and Startups in 2024, which provides market exploration funds and special guarantees. The company’s participation in high-profile events like NVIDIA GTC 2024 and the AI Engineer World’s Fair in San Francisco highlights its technological leadership. Additionally, a settled patent dispute with Hugging Face in January 2025, related to FriendliAI’s patented batching technology, underscores its innovative contributions to the AI inference space.

Potential Risks and Considerations

While FriendliAI’s technology and funding position it for growth, several risks warrant consideration:

• Market Competition: The AI inference market is crowded, with competitors like Fireworks AI and Dify.AI offering similar services. FriendliAI’s ability to maintain its performance edge will be critical.
• Profitability Challenges: The company’s focus on scaling and reinvestment may delay profitability, requiring careful financial management to sustain investor confidence.
• Technological Dependence: Heavy reliance on GPU infrastructure and partnerships (e.g., NVIDIA, AWS) exposes FriendliAI to supply chain risks and partner dynamics.

Summary Table: FriendliAI Funding Round Overview

FriendliAI’s $20 million seed extension round positions the company to capitalize on the growing demand for efficient AI inference solutions. By leveraging its proprietary technologies and strategic partnerships, FriendliAI addresses critical pain points in AI deployment, delivering significant cost savings and performance improvements. With plans to expand in North America and Asia, enhance its Friendli Engine, and scale its client base, FriendliAI is poised for substantial growth in the generative AI infrastructure market.

Please email us your feedback and news tips at hello(at)techcompanynews.com