Magic
AI coding-model lab building foundation models with ultra-long context windows for software development — the LTM (Long-Term Memory) architecture and the LTM-2-mini 100M-token context model released August 2024. Founded 2022 by Eric Steinberger (CEO) and Sebastian De Ro. Cumulative external capital approximately $465M through the August 2024 $320M round (Eric Schmidt, CapitalG, Atlassian, Jane Street, Nat Friedman / Daniel Gross, Sequoia, Elad Gil); Google Cloud partnership on NVIDIA-powered AI supercomputer infrastructure.
The Business
Magic is a privately-held AI coding-model lab building foundation models with ultra-long context windows for software development. The company was founded in 2022 by Eric Steinberger (CEO, previously at FAIR / Facebook AI Research on poker-AI and multi-agent reinforcement learning) and Sebastian De Ro (CTO). Magic’s technical thesis centres on the LTM (Long-Term Memory) model architecture, which is designed to handle ultra-long context windows efficiently — LTM-2-mini, the first 100M-token context model, was announced in August 2024 alongside the $320M funding round. Per Magic’s own blog, LTM-2-mini’s sequence-dimension algorithm is approximately 1000x cheaper than the attention mechanism in Llama 3.1 405B for a 100M-token context window, and the model requires a small fraction of a single H100’s HBM per user against the 638 H100s required for Llama 3.1 405B at the same context length.
Cumulative external capital is approximately $465M through the August 2024 $320M round that brought in Eric Schmidt, Alphabet’s CapitalG, Atlassian, Jane Street and Sequoia alongside existing investors Nat Friedman, Daniel Gross and Elad Gil. Magic has announced a Google Cloud / NVIDIA partnership to build AI supercomputers on Google Cloud Platform for the next-generation LTM training pipeline.
Customers and Distribution
Magic has not released a generally-available commercial coding product as of mid-2026. LTM-2-mini was announced in August 2024 with developer-trial access through Magic’s partnership channels (including the Atlassian partnership and the Google Cloud distribution channel); no GA paid product has been announced. Distribution-readiness for a future commercial release sits across several potential pathways: a direct Magic developer product competing head-to-head with Cursor and GitHub Copilot; a partnership-distribution play through Atlassian (Jira / Bitbucket integration) or Google Cloud channels; a model-API play competing with Anthropic and OpenAI on coding-specific capability; or a strategic outcome (acquisition, partnership-deal) that monetises the LTM model investment.
The competitive set is the coding-AI cohort — GitHub Copilot inside Microsoft (the dominant IDE-distribution incumbent), Cursor / Anysphere (the AI-native IDE leader), Anthropic Claude Code, OpenAI Codex / GPT-5 (the frontier-lab competitors) and Cognition Devin (the autonomous-coding-agent flank). The active variables on commercial trajectory are the cadence of model releases after LTM-2-mini, the Google Cloud / NVIDIA supercomputer-build progress, the next priced round at any post-August-2024 re-mark, and any partnership or distribution announcement that would convert the LTM model thesis into commercial coding-product traction.
Model Strategy
Magic is a Verticals-first foundation-model play on coding-specific models with the LTM ultra-long-context architecture as the differentiating technical thesis. The strategic frame is that coding-foundation models with 100M-token context windows can hold all of a customer’s code, documentation and libraries in context simultaneously — including code not on the public internet — and that this is a defensible technical differentiator against the generalist frontier-lab cohort (OpenAI GPT-5, Anthropic Claude, Google Gemini) for code-synthesis applications. The competitive cadence from frontier-lab long-context releases through 2025 and 2026 (Claude Sonnet 4 / Opus 4 with extended context, GPT-5 with long-context capabilities, Gemini with 1M+ token context) is the principal active pressure on the LTM thesis; the technical gap has narrowed but the 100M-token specialisation remains the distinctive Magic position.
Above the model layer, the Google Cloud / NVIDIA AI supercomputer partnership is the principal infrastructure commitment and the channel-distribution pathway. The capital-intensity of the supercomputer build and the LTM pre-training pipeline is materially expensive; the $465M cumulative capital base gives multi-year runway but the burn rate against the infrastructure commitment is the principal funding watch-item.
At A Glance
The Numbers
Headcount (FTE)
Funding to date
Leadership Team
Magic is privately held and founder-led with Eric Steinberger (CEO) and Sebastian De Ro (CTO) as the principal public voices. The team is small for the $465M capital base and the AI-supercomputer infrastructure commitment; senior engineering hires have come from frontier-lab and FAIR alumni networks. CFO and CRO appointments are not separately public; the research-lab positioning emphasises depth-of-bench over commercial-organisation scale.
IM Framework Scoring
IM’s structured assessment of Magic’s competitive position. The summary below is the headline; expand “Show the full analyst-grade analysis” near the bottom for the per-dimension reasoning and evidence. Methodology →
Funding History
| Date | Round | Raised | Post-money | Lead investor(s) |
|---|---|---|---|---|
| Aug 2024 | Funding round | $320M | — | Eric Schmidt, CapitalG, Atlassian, Jane Street, Nat Friedman, Daniel Gross, Sequoia, Elad Gil |
| 2024 | Prior round | $117M | $500M+ | Nat Friedman, Daniel Gross, CapitalG, Elad Gil |
| 2023 | Seed and Series A | ~$28M cumulative | — | Nat Friedman, Daniel Gross |
Cumulative external capital approximately $465M through the August 2024 $320M round that brought in Eric Schmidt, Alphabet’s CapitalG, Atlassian, Jane Street and Sequoia alongside existing investors Nat Friedman, Daniel Gross and Elad Gil, per the TechCrunch coverage and the Magic blog post announcing the round. Valuation at the August 2024 round was not formally disclosed by the company; named-press reports placed the post-money in the $1.5B–$1.6B range but Magic has not confirmed a specific figure. We hold the disclosed round size of $320M and note the valuation as not formally disclosed by primary sources.
Competitive Landscape
| Competitor | Positioning | Distribution edge | Threat profile |
|---|---|---|---|
| GitHub Copilot ((Microsoft / NASDAQ: MSFT)) |
Dominant AI coding assistant inside VS Code, JetBrains and Visual Studio; Copilot Workspace and Enterprise extend the suggestion-level product into agentic and codebase-grounded workflows. | Bundled into GitHub paid plans and Microsoft / GitHub enterprise contracts; reaches every developer organisation with no separate procurement. | High — the dominant coding-assistant incumbent with deep IDE-distribution depth across Visual Studio Code, the GitHub developer base and the Microsoft enterprise channel; the principal competitive substitution risk for any Magic-built coding product. |
| Cursor (Anysphere) | Closed-source AI-first IDE with frontier-model routing (Claude Sonnet 4 / Opus 4, GPT-5, Gemini 2.5 Pro) and agent mode; positions itself as the AI-native replacement for VS Code and the fastest-growing coding-AI product by revenue in 2025-2026. | Direct download with bottom-up developer adoption; per-seat self-serve ramps into Business and Enterprise tiers. | High — the AI-native IDE leader with deep developer mindshare and a multi-model approach across Anthropic Claude, OpenAI GPT and open-source models; competes for the same developer-application surface that a Magic commercial product would target. |
| Anthropic (Claude Code) | CLI-and-IDE coding agent built on Claude Sonnet 4 / Opus 4; positions itself as the terminal-native engineering teammate with deep Claude-Constitutional-AI grounding. | Distributed inside the Anthropic account (claude.ai Team / Enterprise and the Anthropic API) plus AWS Bedrock and Google Cloud Vertex AI reseller channels. | High — the frontier-lab coding competitor with Claude Code and the Claude Sonnet / Opus coding capabilities; the principal head-to-head on coding-foundation-model capability and on developer-platform distribution. |
| OpenAI (Codex / GPT-5) | Codex as the cloud-hosted coding agent surface on top of GPT-5 and the o-series reasoning models; positions itself as the OpenAI-native engineering teammate inside ChatGPT and the OpenAI API. | ChatGPT consumer and Enterprise tiers, the OpenAI API and Azure OpenAI Service; broadest developer-AI distribution surface in the cohort. | High — the dominant frontier-lab competitor on long-context and coding capability via the GPT-5 series; competes head-to-head on the technical thesis that drives Magic’s strategic positioning. |
| Cognition (Devin) | Autonomous-engineer positioning — Devin as the hosted teammate that takes Jira / Linear tickets end-to-end; the closest direct mirror to Magic’s superhuman-engineer thesis. | Direct enterprise sales into engineering orgs plus per-seat subscriptions for Devin Team; the Founders Fund / Khosla narrative anchors the channel. | Medium — AI software-engineer flank with Devin as the autonomous-agent positioning; competes on coding-agent execution capability though with a different technical thesis (autonomous-agent execution rather than ultra-long-context model). |
Potential Risks
No released commercial product
Magic has not released a generally-available commercial product as of mid-2026. LTM-2-mini was announced in August 2024 alongside the $320M funding round with developer-trial access through Magic’s partnership channels, but the company has not announced general-availability for a paid coding product. The $465M capital base gives multi-year runway against the Google Cloud / NVIDIA supercomputer commitments but the absence of revenue and the absence of a released GA product are the principal structural risks against the IDE-incumbent and frontier-lab competitors.
Frontier-lab long-context capability substitution
Anthropic Claude (Sonnet 4 / Opus 4 with extended context), OpenAI GPT-5 (with long-context capabilities) and Google Gemini (with 1M+ token context) have all materially expanded long-context capabilities through 2025 and 2026. The LTM-2-mini 100M-token thesis is technically differentiated against the frontier-lab cohort but the competitive gap has narrowed. The competitive substitution from frontier-lab long-context releases is the principal long-run structural risk against the LTM technical thesis.
Capital-intensity and supercomputer-build risk
The Google Cloud / NVIDIA AI supercomputer partnership commits Magic to a capital-intensive infrastructure stack. The $465M capital base is generous against research-lab-stage operating costs but the supercomputer-build and the LTM pre-training pipeline are materially expensive; the next priced round at any post-August-2024 re-mark is the principal funding watch-item against the burn rate.
Small-team scale relative to capital base
Magic’s team size is small for the $465M capital base and the AI-supercomputer infrastructure commitment. Senior engineering hires have come from frontier-lab alumni networks but the team has not been disclosed in the multi-hundred range typical of frontier-lab competitors with comparable capital. Bench-depth on the LTM model training pipeline and the inference-architecture stack is the principal technical-execution variable.
Commercial-product-release pathway risk
The value-creation pathway from research-lab depth to commercial revenue is undefined. Possible pathways include GA release of a Magic commercial coding product, partnership-distribution through Atlassian (Jira / Bitbucket) or Google Cloud channels, an acquisition outcome by a frontier-lab competitor or a hyperscaler, or a developer-platform play that competes head-to-head with Cursor and GitHub Copilot. None has been publicly disclosed as of mid-2026.
Recent IM Coverage
- Coding AI — sector landing Jun 2026.
- AI Tracker — Coding AI cohort Jun 2026.
- AI Tracker methodology Jun 2026.
Show recent press coverage of Magic
- Aug 2024 — Generative AI coding startup Magic lands $320M investment from Eric Schmidt, Atlassian and others (TechCrunch)
- Aug 2024 — 100M Token Context Windows (Magic Blog)
- Aug 2024 — Ex-Google CEO Schmidt Joins Magic’s $320 Million Raise to Develop AI Coding (PYMNTS)
- Aug 2024 — AI Coding Startup Magic Secures $320M Investment from Eric Schmidt, Atlassian & Leading Investors (The AI Insider)
- 2024 — Magic (Sequoia portfolio company page) (Sequoia Capital)
Show the source register for the figures on this page
IM operates a primary-source-where-possible discipline. The figures above come from:
- Revenue: Magic does not disclose revenue. The company is in pre-product-GA-release stage as of mid-2026 with developer-trial access to LTM-2-mini through partnership channels but no announced GA paid product. We decline-to-publish a revenue figure pending a primary disclosure or commercial-product-release announcement.
- Cumulative capital: Cumulative external capital approximately $465M through the August 2024 $320M round per TechCrunch and the Magic blog post. Lead investors Eric Schmidt, Alphabet’s CapitalG, Atlassian, Jane Street, Sequoia, Nat Friedman, Daniel Gross and Elad Gil. Valuation at the August 2024 round was not formally disclosed by primary sources.
- Headcount: Magic does not publicly disclose precise headcount. The team is small for the $465M capital base and the Google Cloud / NVIDIA supercomputer commitment; senior engineering hires from FAIR, OpenAI, Google DeepMind and frontier-lab alumni networks per LinkedIn-visible disclosures. We decline-to-publish a precise figure and reference the Magic careers page as the canonical entry point.
- Funding to date: Cumulative external capital approximately $465M through the August 2024 $320M round per the named-press cycle. The August 2024 round brought in Eric Schmidt, Alphabet’s CapitalG, Atlassian, Jane Street and Sequoia alongside existing investors Nat Friedman, Daniel Gross and Elad Gil. Prior 2024 round disclosed at $117M; Seed and Series A cumulative approximately $28M. Valuation at the August 2024 round was not formally disclosed.
Methodology & Disclaimer
For metric definitions, source-tier hierarchy, and decline-to-publish rules, see the tracker methodology. Confidence dots (• green / • amber / • red) follow the same convention as the AI Tracker.
Spotted a figure you believe is wrong? Send corrections to info@informationmatters.net.
Information Matters Framework scores are the considered opinion of the IM team — human and AI — applied to publicly-available evidence under a disclosed methodology. They are not statements of fact about the companies scored and they are not investment advice.
