What Trillion Labs does

Trillion Labs is a Seoul-based AI research lab founded in 2024 by former NAVER executive Jay Shin, building specialized foundation models and agentic AI systems for industrial environments, primarily power plants and AI data centers. The company's product portfolio centers on the Gravity series of lightweight STEM-specialized MoE models (16B-A3B, Gravity Nano, Gravity Flash) designed to deliver frontier-class performance in industrial reasoning at sub-frontier compute cost, alongside the Tri series of Korean-language foundation models (7B to 70B parameters) and Trida-7B, Korea's first diffusion-based large language model. Supporting infrastructure includes the proprietary XLDA training technology (reducing learning costs to one-twelfth of standard levels), the rBridge framework for 100x-700x cost reduction in pre-training optimization (published on arXiv 2509.21013 in collaboration with KAIST AI), and PowerBench, an evaluation suite covering Knowledge, Reasoning, and Decision tiers for power plant operations where all current models score 0% on the Decision tier.

The company deploys these capabilities through direct enterprise sales, embedding with strategic industrial partners such as GS Group to develop AI agents that interpret engineering diagrams (P&IDs, logic diagrams, electrical schematics), diagnose equipment issues, and execute end-to-end industrial workflows. Compute infrastructure is accessed primarily through NVIDIA technology partnerships (Inception program, Omniverse, Nemotron, NeMo Curator) and NIPA's national AI computing program (80 H200 GPUs). The company is also competing for selection in South Korea's sovereign AI foundation model project alongside Upstage, LG AI Research Institute, and SK Telecom.

Trillion Labs is venture-backed at the seed stage with $4.2 million raised in August 2024 from Strong Ventures (lead), Kakao Ventures, and other domestic and foreign investors. The business model relies on custom enterprise contracts for industrial AI deployments, with all models and frameworks also distributed open-source via Hugging Face and GitHub as a top-of-funnel developer relations strategy. Pricing is not publicly disclosed, revenue is not yet reported, and the company operates with 1-10 employees from its Gangnam district headquarters in Seoul.

Name

Trillion Labs

Legal name

Trillion Labs

Website

https://trillionlabs.co

Company type

Private

Founded year

2024

Operating status

Operating

Headcount range

1–10 employees

Short description

Trillion Labs is a Seoul-based AI research lab that builds specialized foundation models and agentic AI systems for industrial environments, primarily power plants and AI data centers, serving enterprise operators through direct deployment partnerships such as GS Group.

Ownership category

akta.pro rank

Product category

Industrial AI Foundation Models

NAICS

Computer Systems Design and Related Services (54151)

SIC

Services-Computer Integrated Systems Design (7373)

akta.pro primary industry

Foundation Model Developers (LLM/Multimodal Model Labs) (HDAAACAA)

akta.pro secondary industries

Prompt Engineering, Orchestration & LLMOps Tooling (HDAAACAD), Model Development & Training Platforms (AutoML, Notebooks, Feature Stores) (HDAEANAB)

Keywords

• Industrial AI foundation models
• AI for energy
• AI factory infrastructure
• Domain-specialized language models
• Korean large language models

Headquarters

HQ city

Seoul

HQ country

South Korea

HQ region

Asia

Offices1 record

Markets served

GTM type

B2B

Offering type

Software

Cost components

Technology or R&D, Infrastructure, Personnel

Revenue model

1. Enterprise AI Solutions: Domain-specialized foundation models and AI agents deployed to industrial environments such as power plants and AI data centers. Revenue generated through enterprise contracts with direct engagement model.

Go-to-market motion2 records

Distribution channels1 record

Marketing channels5 records

Core offering

Trillion Labs develops domain-specialized foundation models and AI agents for industrial environments such as power plants and AI data centers, with a particular focus on the energy infrastructure underlying AI factories. Its product portfolio includes the Gravity series of STEM-specialized reasoning models, the Tri and Trida series of Korean language foundation models, and supporting methodologies like rBridge, delivered via enterprise contracts with energy operators and government sovereign AI initiatives.

Product overview

Trillion Labs is an AI research lab building specialized foundation models and agentic AI systems for industrial environments, particularly power plants and AI data centers. The product portfolio centers on the Gravity series (16B-A3B, Flash, Nano) — STEM-specialized models for industrial reasoning — combined with the Tri series (7B-70B) for Korean language foundation models, and specialized tools like rBridge for efficient model evaluation. The company offers an AI for Energy platform targeting power plant operations, which includes PowerBench for evaluation and specialized agents for end-to-end industrial tasks. All products are designed around the company's mission of recursive self-improving AI infrastructure for AI factories.

Differentiator

Problem solved

Functional benefit

Brands

• Gravity: Foundation model series for STEM and agentic tasks, including Gravity Nano (16B-A3B)
• Tri
• Trida
• rBridge
• gWorld

Products and services

• AI for Energy Industrial AI platform comprising AI agents and foundation models specifically designed for power plant operations, including interpretation of engineering diagrams, operational data analysis, equipment relationship reasoning, and end-to-end maintenance procedure execution for energy operators.
• Gravity-16B-A3B Lightweight STEM-specialized MoE foundation model for power plant operations. Achieves state-of-the-art performance in the power plant domain with significantly lower training cost, demonstrating that frontier-class industrial capability does not require frontier-scale compute.
• Gravity Nano (16B-A3B-Preview) Pareto-frontier reasoning foundation model for STEM and agentic tasks using a new Mixture of Experts (MoE) series trained from scratch.
• Gravity Flash Specialized model designed to deliver frontier-class performance at sub-frontier cost for power plant domain applications; first internal PowerBench results expected in Q3 2026.
• Tri Series (7B to 70B) First generation of open Korean foundation models with 7B, 21B, and 70B parameter versions trained from scratch for Korean language understanding and generation.
• Trida-7B Korea's first diffusion-based large language model using parallel generation architecture for faster inference and computational efficiency, built using 80 NVIDIA H200 GPUs through NIPA's national AI computing infrastructure.
• Tree-70B Open-source 70B parameter foundation model developed as part of South Korea's national AI foundation model project competition.
• gWorld-32B Mobile world model for enabling AI systems to understand, simulate, and optimize complex industrial environments.
• rBridge Methodology enabling tiny proxy models (≤1B parameters) to reliably predict the reasoning performance of much larger models (13× to 32× larger), reducing pre-training data and hyperparameter optimization costs by over 100× to 700×.

Quantifiable outcome

• Training time reduced from 24 hours on CPU to 3.4 hours using 8x H100 GPUs for 1.7 trillion tokens
• +5 more outcomes

Named customers1 record

Segments3 records

Ideal customer profiles3 records

Technology focussed Yes

API detail

Has API

No

API docs

API detail

Core technology

AI maturity

App detail

AI capability10 records

Feature9 records

Partnerships

Three partnerships are on record, tiered core and minor.

• NVIDIAcoreTechnology or Integration · 8 June 2026Active partnership through NVIDIA Inception program. Using NVIDIA Omniverse libraries and NVIDIA Nemotron open models for Industrial World Models development. Access to H100 and H200 GPU infrastructure through NIPA's national AI computing infrastructure. Deployed NVIDIA NeMo Curator for GPU-accelerated data curation.
• GS GroupcoreStrategic or Co-development Partner · 8 June 2026Collaborating on critical energy infrastructure including power plants as part of Industrial World Models for AI Factories initiative. GS Group hosted technology cooperation event with Trillion Labs among 17 AI startups in their investment portfolio, focusing on physical AI and robotics integration into energy plants, construction sites, and distribution logistics centers.
• KAIST AIminorStrategic or Co-development Partner · 1 February 2026Academic collaboration on AI research with co-authorship on the rBridge paper. Authors include Woosung Koh, Juyoung Suk, Sungjun Han, Se-Young Yun from Trillion Labs and KAIST AI.

Scale indicators3 records

Recent moves6 records

Expansion highlights6 records

Direct peers

• Upstage: Seoul-based foundation model startup also competing in Korea's sovereign AI national representative selection with Solar LLM family. Direct head-to-head competitor for Korean enterprise/government foundation model mandates, with overlapping GTM in chaebol accounts.
• Mistral AI: European open-source foundation model developer with comparable approach: releasing competitive open-weight models (7B-70B+) while pursuing enterprise GTM. Closely comparable on open-source distribution strategy, model sizing philosophy, and enterprise foundation model positioning.
• Cohere: Enterprise-focused foundation model company building specialized models for business use cases with secure deployment and customization. Comparable on enterprise foundation model positioning, B2B sales motion, and emphasis on private/customized deployments over consumer chat.
• Sakana AI: Japan-based foundation model lab focused on efficient, smaller-parameter models for non-English languages and specialized domains. Closely comparable on philosophy of frontier-class performance at sub-frontier scale, and on East Asian sovereign AI positioning adjacent to Korea.

Broad incumbents

• Naver Cloud (HyperCLOVA X): Naver's flagship foundation model developed by former employer of Trillion Labs' founder Jay Shin. Competing in the same Korean sovereign AI race with materially larger resources, and serves as the incumbent Trillion Labs is positioning to displace for Korean-language enterprise AI.
• LG AI Research (EXAONE): LG Group's foundation model lab developing EXAONE, a major competitor in Korea's national sovereign AI selection. Backed by LG conglomerate resources and serves as one of the four named competitors for the Korean representative AI designation.
• SK Telecom (A.X / Sovereign LLM): SK Telecom's sovereign AI initiative with Korean-language foundation models, one of the named competitors alongside Upstage, LG AI Research, and Trillion Labs for Korea's national AI representative status. Brings telco distribution and capital scale.
• Anthropic: Frontier AI lab whose Claude Opus model is the explicit benchmark Trillion Labs compares Gravity-16B-A3B against in PowerBench. Represents the generic-frontier-model competitive threat that Trillion Labs' domain specialization is positioned to outmaneuver in industrial verticals.

Emerging players

• Adept AI: AI agent company building models that execute end-to-end enterprise workflows, similar to Trillion Labs' positioning of Gravity as an agent that executes full industrial operations tasks (diagnosis, work plans, procedure execution). Comparable on agentic AI for complex enterprise workflows.

Others

• NVIDIA (Nemotron): Technology partner through the NVIDIA Inception program and source of the Nemotron open models and Omniverse libraries Trillion Labs builds on. Functions simultaneously as critical infrastructure provider and adjacent competitor with its own domain-specialized open models.

Market position

Strengths5 records

Weaknesses5 records

Competitive moat5 records

Key risks6 records

Key highlights7 records

Customer concentration

• GitHubgithub.com/trillion-labs
• LinkedInlinkedin.com/company/trillion-labs

Revenue estimate

Valuation estimate

Number of profiles

Profiles3 records

Funding overview

Funding rounds2 records

Investors7 records

Funding detail is available on the Subscription and Enterprise plan.Contact sales →

M&A

Investments

M&A and investment is available on the Subscription and Enterprise plan.Contact sales →

Frequently asked questions about Trillion Labs