MangoBoost
Category: AI Infrastructure
Full-stack agentic AI infrastructure company spanning proprietary DPUs, inference software, servers and serverless inference, spun out of Seoul National University research. MangoBoost was founded in 2022. The company is led by Jangwoo Kim. Based in Bellevue, Washington, USA (R&D in Seoul, Korea; offices in San Jose and Tokyo). Team size: 101-500. Total funding raised: $100.0M. Latest round: Venture Round. Key investors include Stonebridge Ventures, IMM Investment, Shinhan Venture Investment, Samsung, KT Investment, BonAngels Venture Partners, Aju IB Investment, Aion Asset Management.
- Founded
- 2022
- Headquarters
- Bellevue, Washington, USA (R&D in Seoul, Korea; offices in San Jose and Tokyo)
- Team size
- 101-500
- Total funding
- $100.0M
Value proposition
Vendor-neutral full-stack AI infrastructure, from DPU silicon to inference software to server systems, that cuts TCO and raises utilization for AI inference across heterogeneous GPU/NPU environments.
Products and solutions
Mango BoostX DPUs (RoCE AI, NRT, Base), LLMBoost inference optimization software, Mango Inference (serverless inference platform, launched July 2026), Mango Alphonso GPU server, Mango Kesar distributed storage, Agent Serve.
Unique value
Proprietary in-house DPU plus software and systems stack with MLPerf-validated inference performance on AMD GPUs, including claimed 5.1x faster response than an NVIDIA B300 on MI355X.
Target customer
Cloud service providers, AI data centers, enterprises and sovereign AI deployments, plus developers via serverless inference.
Industries served
Cloud and data center infrastructure, enterprise AI, sovereign AI
Technology advantage
15+ years of SNU DPU/systems research, 50+ patents, MLPerf records (v4.1 to v6.1), heterogeneous compute and prefill-decode disaggregation expertise.
How they differentiate
Vendor-neutral, Ethernet-based heterogeneous full stack that combines DPU, software, systems and API service in one offering, rather than DPU silicon alone.
Main competitors
NVIDIA (BlueField DPU), AMD Pensando (also a partner), Marvell OCTEON
Key partnerships
AMD (key AI ecosystem partner), FuriosaAI (MOU with AMD, Nov 2026), Preferred Networks (MOU at SC26, Nov 2026, evaluating GPU prefill plus MN-Core decode), Supermicro, Ultra Ethernet Consortium, UALink Consortium, Linux Foundation OPI, DAOS
Major milestones
2022 founded, seed, 2023 Series A $55M, 2024 joined UEC, UALink and Linux Foundation OPI, 2025 LLMBoost launch and MLPerf 5.0/5.1 records, 2026 MLPerf 6.0 multi-region hetero inference, Mango Inference launch (July), AMD/FuriosaAI MOU (Nov), PFN MOU at SC26 (Nov 17).
Growth metrics
120+ technical experts; 50+ patents; 20+ investors; Series B (~$100M at roughly $500M valuation, reported) in progress as of mid-2026; Series A valuation of about ₩403B.
Market positioning
Korean-founded, US-headquartered AI infrastructure challenger positioning against NVIDIA's stack, aligned with the AMD ecosystem.
Geographic focus
US, Korea, Japan, with expansion to the Middle East and Southeast Asia
Patents and IP
50+ patents rooted in SNU research (company-reported)
About Jangwoo Kim
Professor, Seoul National University (2017-); Professor, POSTECH (2010-2017); CPU/Server Architect, Sun Microsystems/Oracle (UltraSPARC T4); PhD ECE Carnegie Mellon; MEng and BE Cornell. 40+ papers at ISCA/MICRO/ASPLOS/HPCA/OSDI; ISCA and MICRO Hall of Fame.
Latest news about MangoBoost
More AI Infrastructure companies
Official website: https://www.mangoboost.io