Magic

What Magic does

Magic (legal entity: Magic AI, Inc.) is a San Francisco-based AI startup founded in 2022 by Eric Steinberger (CEO) and Sebastian De Ro, focused on automating software engineering through frontier-scale AI coding models. The company develops proprietary Long-Term Memory (LTM) models — notably LTM-1 (5M-token context, launched June 2023) and LTM-2-mini (100M-token context, announced August 2024) — that enable reasoning over entire codebases (~10 million lines of code) in a single inference pass. Its core technical differentiator is a novel sequence-dimension algorithm that the company claims is roughly 500-1000x more efficient than standard attention mechanisms for 100M-token contexts, implemented via a vertically integrated training and inference stack including custom CUDA kernels. Magic's stated thesis is that automating AI research and code generation is the most promising path to safe AGI, and it has published an AGI Readiness Policy developed with METR outlining dangerous-capability evaluations and deployment thresholds (e.g., a 50% LiveCodeBench Pass@1 trigger).

The company operates its own frontier-scale compute infrastructure: 8,000 NVIDIA H100 GPUs plus thousands of GB200s across two AI supercomputers (Magic-G4 and Magic-G5 with NVIDIA GB200 NVL72) built in partnership with Google Cloud, with stated capacity to scale to tens of thousands of Blackwell GPUs. Magic has raised approximately $515M in total funding across a $5M seed (Summer 2022), $23M Series A led by CapitalG (Feb 2023), $117M led by NFDG Ventures (Feb 2024), and a $320M Series B led by former Google CEO Eric Schmidt (Aug 2024), at a reported $1.5B valuation. Key investors include Sequoia, Jane Street, Atlassian, Nat Friedman, Daniel Gross, and Elad Gil. The current headcount is approximately 23 engineers and researchers, with active hiring across research, RL, kernels, infrastructure, and product roles.

Magic's go-to-market combines product-led growth through magic.dev for individual developer adoption with a parallel enterprise sales motion targeting larger organizations. The revenue model is described as subscription or usage-based recurring access to its AI coding assistant, though pricing is not publicly disclosed and no customer logos are named. Primary customers are individual software developers and engineering teams seeking AI-powered coding assistance; secondary targets are enterprises seeking to transform software development workflows and address engineering talent shortages. As of the input data, no revenue figure is disclosed and the company remains in a research-and-build phase ahead of broad commercial availability.

Name

Magic

Legal name

Magic AI, Inc.

Website

https://magic.dev

Company type

Private

Founded year

2022

Operating status

Operating

Headcount range

101–250 employees

Short description

Magic (Magic AI, Inc.) is a San Francisco-based AI startup building proprietary Long-Term Memory (LTM) code models with up to 100M-token context windows. Founded in 2022, it serves individual software developers and enterprise engineering teams with frontier-scale AI coding assistants that reason over entire codebases.

Ownership category

akta.pro rank

Product category

AI Coding Assistant

NAICS

Software Publishers (513210), Computer Systems Design and Related Services (54151)

SIC

Services-Prepackaged Software (7372)

akta.pro primary industry

End-to-End Enterprise AI Platforms (MLOps & Model Lifecycle Management) (HDAEANAA)

akta.pro secondary industry

AI Application Enablement Platforms (Copilot/Agent Frameworks, SDKs) (HDAEANAJ)

Keywords

• AI coding assistant
• Code generation models
• Long-context language models
• Software engineering automation
• Developer AI tools

Headquarters

HQ city

San Francisco

HQ country

United States

HQ region

North America

Offices1 record

Markets served

GTM type

B2B

Offering type

Software

Cost components

Technology or R&D, Infrastructure, Personnel, Marketing or Sales, Operations

Revenue model

1. AI Coding Tools Subscription: Magic provides AI-powered software engineering tools that automate code generation, debugging, and software development tasks. The revenue model likely follows a subscription or usage-based approach for developer and enterprise customers accessing the AI colleague platform.

Go-to-market motion1 record

Distribution channels2 records

Marketing channels3 records

Core offering

Magic develops AI-powered software engineering tools with proprietary Long-Term Memory (LTM) models featuring up to 100M token context windows. The company's AI coding assistant acts as an AI colleague for developers, automating code generation, debugging, and entire feature development by reasoning over complete codebases. Magic operates a vertically integrated stack combining frontier-scale pre-training, domain-specific reinforcement learning, ultra-long context, and inference-time compute.

Product overview

Magic offers an AI-powered software engineering platform centered on its core AI coding assistant. The product portfolio includes LTM-1 (5 million token context model) and LTM-2-mini (100 million token context model) as key model releases within the LTM (Long-Term Memory) family, plus the HashHop evaluation benchmark. Magic's approach is vertically integrated—they conduct research, train foundation models, build products, and engage directly with users. The company combines frontier-scale pre-training, domain-specific reinforcement learning, ultra-long context windows, and inference-time compute to achieve their goal of automating software engineering toward safe AGI.

Differentiator

Problem solved

Functional benefit

Products and services

• Magic AI Coding Assistant AI-powered software engineering assistant that acts as an AI colleague for software developers. Uses frontier-scale code models with up to 100M token context windows to automate code generation, debugging, and feature development across entire codebases. Targets individual professional developers and enterprise engineering organizations seeking to accelerate software development and address talent shortages.

Quantifiable outcome

• 100M token context model with ~10 million lines of code capacity
• +1 more outcomes

Segments2 records

Ideal customer profiles2 records

Technology focussed Yes

API detail

Has API

No

API docs

API detail

Core technology

AI maturity

App detail

AI capability4 records

Feature4 records

Partnerships

Two partnerships are on record, tiered core.

• NVIDIAcoreTechnology or Integration · 30 August 2024NVIDIA providing GB200 NVL72 rack-scale design and GPU technology for Magic's supercomputers on Google Cloud, enabling training, deployment, and scaling of large language models.
• Google CloudcoreStrategic or Co-development Partner · 29 August 2024Magic partnered with Google Cloud to build two AI supercomputers: Magic-G4 powered by NVIDIA H100 Tensor Core GPUs and Magic-G5 powered by NVIDIA GB200 NVL72, with ability to scale to tens of thousands of Blackwell GPUs over time.

Scale indicators5 records

Recent moves6 records

Expansion highlights6 records

Direct peers

• Anysphere (Cursor): Anysphere builds Cursor, a leading AI coding assistant IDE that competes directly with Magic's core product. Both target software developers with AI-native coding experiences and have raised significant venture funding to compete on model quality and developer workflow integration.
• Cognition AI (Devin): Cognition builds Devin, an autonomous AI software engineer that takes multi-step actions to complete coding tasks. Both Magic and Cognition pursue the "AI colleague" framing for software engineering rather than autocomplete, and both are well-funded frontier-AI startups.
• Codeium (Windsurf): Codeium offers Windsurf, an AI coding IDE with autocomplete, chat, and agentic capabilities. It competes directly with Magic on AI coding assistance for individual developers and enterprises, with strong free-tier adoption among developers.
• Poolside AI: Poolside AI is building foundation models purpose-built for software engineering with large-context capabilities and reinforcement learning. Like Magic, Poolside targets enterprise developers and pursues a vertically integrated AI coding stack.
• Sourcegraph (Amp): Sourcegraph's Amp is an AI coding agent focused on large codebases, directly competing with Magic's value proposition of reasoning across entire repositories. Sourcegraph also brings enterprise distribution and code search infrastructure to the AI coding space.
• Tabnine: Tabnine provides AI coding assistance focused on enterprise developers with an emphasis on personalization and on-premise deployment. Both Magic and Tabnine target professional software engineering teams with AI-powered productivity tools.

Broad incumbents

• GitHub Copilot: GitHub Copilot, backed by Microsoft and OpenAI models, is the dominant AI coding assistant with massive distribution through GitHub and enterprise channels. It competes with Magic on the same developer productivity use case but as part of a broader platform portfolio.
• Replit: Replit offers an AI-powered cloud development environment with integrated coding agents. While broader than Magic's offering (it spans IDE, hosting, and deployment), Replit competes for the same developer productivity use case and is expanding AI capabilities aggressively.
• Anthropic: Anthropic's Claude models (especially Claude Code) are widely used for AI coding assistance and compete with Magic's LTM models on long-context code reasoning. Anthropic is a foundational model provider with broader enterprise reach than Magic but overlapping target market.
• OpenAI: OpenAI's GPT models power many AI coding tools (including GitHub Copilot) and ChatGPT itself is used heavily by developers. As the dominant foundation model provider, OpenAI competes with Magic's vertically integrated approach on the model layer and via its coding-focused products.

Market position

Strengths5 records

Weaknesses5 records

Competitive moat5 records

Key risks6 records

Key highlights6 records

Customer concentration

• Xx.com/magicailabs
• LinkedInlinkedin.com/company/magicailabs

Revenue estimate

Valuation estimate

Number of profiles

Profiles2 records

Funding overview

Funding rounds5 records

Investors9 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 Magic