Developer Tools

Category Research Report

Your code, your keystrokes, your development patterns — collected by default, used to train models you don't control, and sold back to you as "AI-powered" features. This is the landscape, the data, and the opportunity.

---

The Landscape

The core software development tools market is valued at $6.4-7.5 billion in 2025, growing at 14-17% CAGR, projected to reach $13.7 billion by 2030 (Mordor Intelligence, 2025). The AI coding tools sub-segment alone reached $7.37 billion in 2025, with GitHub Copilot holding 42% market share and Cursor at 18%. The global developer population reached 20.8 million professionals in 2025 (JetBrains Developer Ecosystem Report, 2025).

Source Code Hosting

GitHub now hosts 630 million repositories and added 36 million new developers in 2025 alone — more than one per second on average (GitHub Octoverse 2025). India overtook the US in total open-source contributor count for the first time. GitLab's market cap sat at $4.98B as of February 2026, down 57.6% over the prior year, with its Ultimate tier comprising 54% of ARR (GitLab Q3 FY26 earnings).

Sources: GitHub About, GitHub Octoverse 2025, GitLab MacroTrends, Kinsta GitHub Statistics

AI Coding Assistants

Cursor (Anysphere) is the fastest-growing SaaS product in history, reaching $100M ARR in January 2025 without marketing spend, $500M by June 2025, and crossing $1B by November 2025. It raised $2.3B in Series D at a $29.3B valuation (November 2025, Accel/Coatue). GitHub Copilot is used by 90% of the Fortune 100 and added 5 million users in Q2 2025 alone (TechCrunch, July 2025). Copilot now represents over 40% of GitHub's revenue growth and is a larger business than all of GitHub was at Microsoft's 2018 acquisition.

GitHub introduced metered billing for premium requests in June 2025 at $0.04 per overage request. Cursor's July 2025 switch from 500 fixed requests to usage-metered caps on the $20 Pro plan provoked complaints about unexpected charges; the company rolled back limits and promised refunds.

Sources: TechCrunch Copilot 20M Users, TechCrunch Cursor $9.9B, Anysphere $29.3B, Mind the Product Copilot $2B

IDEs and Editors

VS Code first took the #1 spot in 2018 at 35% usage. It reached 75.9% in 2025, up from 73.6% in 2024 (Stack Overflow Developer Survey 2025, 49,000 respondents, 177 countries). Over 7,593 companies use VS Code in enterprise settings. However, AI-native editors like Cursor and Claude Code are "attracting interest from developers already using VS Code" (Visual Studio Magazine, August 2025).

Sources: Stack Overflow Dev Survey via Visual Studio Magazine, 6sense VS Code Market Share, Second Talent IDE Statistics

Deployment and Hosting Platforms

Vercel's revenue grew from $144M (2024) to ~$200M (2025). It powers 4M+ websites and processes 30 billion requests weekly. Its v0 AI tool has 3.5M users. Netlify has not raised since its 2021 Series D and its employee count shrank 4% last year to 177.

Railway claims 31% of Fortune 500 companies use its platform, with just 7-10 employees. Render grew to 4.5M+ developers with 250,000 joining monthly. Fly.io has not raised since its 2023 Series C.

Sources: Vercel Statistics, Getlatka Vercel, CNBC Render $1.5B, Railway Series B, Getlatka Netlify

Replit

Replit overhauled its pricing in February 2026: Core dropped from $25/month to $20/month, a new Pro tier launched with tiered credits ($100-$4,000/month), and the Teams plan was sunset. CEO Amjad Masad publicly acknowledged that v2 pricing left them "out of whack" because AI inference costs from Anthropic and OpenAI exceeded revenue. Heavy users report spending $100-$300/month on top of base plans, with cost unpredictability being the primary complaint.

Sources: Replit Blog Pro Plan, Replit Blog Effort-Based Pricing, Hackceleration Replit Review

Open-Source Self-Hosted Alternatives

Forgejo was forked from Gitea in late 2022 after Gitea's acquisition by a private company. The Zig programming language migrated from GitHub to Codeberg in 2025, signaling growing confidence in community-governed forges. Forgejo 1.3.0 introduced CI/CD pipelines via community runners and is compatible with GitHub Actions (Gitea 1.23+, Forgejo community runners).

Sources: Self-Hosted Git Platforms 2026, GitHub Alternatives

---

The Enshittification Timeline

• 2018: Microsoft acquires GitHub for $7.5B. The open-source community reacts with alarm. Codeberg, Gitea, and GitLab see temporary migration spikes. Microsoft promises independence. GitHub was free for public repos; private repos required a paid plan.

• 2019-2020: The golden period. Microsoft makes private repos free for all users. GitHub Actions launches with generous free tiers. Trust slowly rebuilds. Developer count crosses 56 million.

• 2021: Copilot launches as technical preview. Trained on all public GitHub repositories — billions of lines of open-source code. Developers who contributed that code were not asked, compensated, or notified. The product is free during preview to drive adoption.

• 2022: Copilot goes paid at $10/month. The code it was trained on remains free; the product built from that code costs money. Doe v. GitHub class-action lawsuit filed in November 2022 seeking $9 billion in damages for DMCA violations and breach of open-source licenses.

• 2023: JetBrains removes continuity discounts for new subscriptions (effective January 2025), effectively a 20-40% price increase over time. VS Code telemetry collects usage data by default with an opt-out (not opt-in) model.

• 2024: AI pricing proliferates. Cursor launches and reaches $100M ARR by January 2025. Vercel shifts to hybrid usage-based pricing, introducing complexity and surprise bills. Netlify stagnates — no new funding since 2021.

• 2025: The price squeeze. JetBrains announces 11-30% price increases effective October 2025 across all IDEs. GitHub Copilot introduces metered billing at $0.04 per premium request overage (June 2025). Cursor's Pro plan switches from fixed requests to usage-based caps, then rolls back after backlash. GitHub Actions introduces a $0.002/minute platform charge for 2026. Replit's CEO admits AI costs exceed revenue under their pricing model.

• 2025: The acquisition wave. OpenAI acquires Windsurf/Codeium for ~$3 billion. Cursor (Anysphere) acquires Graphite (code review, ~$290M). GitHub Copilot CEO Thomas Dohmke steps down.

• 2026: Usage-based becomes the norm. Vercel, Cursor, Replit, and GitHub all move toward consumption-based pricing. Predictable monthly bills become harder to maintain. Replit overhauls pricing again in February 2026. GitHub Actions adds cloud platform charges. The Pro-to-Enterprise pricing "cliff" at Vercel draws criticism — enterprise contracts start at $20-25K/year.

Sources: JetBrains Price Increase Blog, GitHub Actions Pricing Changes, Flexprice Vercel Breakdown, PriceTimeline Vercel

---

The Data Audit

VS Code telemetry collects by default:

• Crash reports, error telemetry, and usage data
• A MacAddressHash to identify users (hashed, but persistent)
• Feature usage patterns, command invocations, execution durations
• Third-party extension telemetry is not controlled by VS Code's settings — each extension collects independently
• Opt-out is available but not opt-in; some data leaks before a user can disable it
• Users cannot view what data has been collected about them — Microsoft says there is no reliable mechanism for this because VS Code has no sign-in experience

GitHub data collection:

• All code pushed to public repositories is used as Copilot training data
• GitHub Advanced Security charges $30/committer for Code Security and $19/committer for Secret Protection — scanning your code for vulnerabilities is a paid upsell
• GitHub Actions processed 23 million jobs/day as of early 2024, generating detailed workflow metadata

AI coding assistant data:

• Copilot sends code context (surrounding lines, file contents) to cloud models for completion
• Cursor's autocomplete and Windsurf's Supercomplete send keystroke-level requests to cloud endpoints
• Replit's Agent uses AI inference from Claude Sonnet 4 and GPT-4o — every interaction is a cloud request with associated data transmission
• JetBrains AI offers local model support (Qwen2.5-Coder, DeepSeek-Coder, CodeStral) for offline-capable completions — a notable exception

The Copilot training data controversy:

• Copilot was trained on public GitHub repos without explicit consent from code authors
• The resulting tool strips copyright notices and license terms from generated output
• Academic research found 29.1% of Python and 24.2% of JavaScript code generated by Copilot contained security weaknesses across 43 CWE categories (including SQL injection, insufficient randomness)

VSCodium exists as a privacy-focused VS Code distribution with telemetry stripped out — but loses access to Microsoft's proprietary extension marketplace.

Sources: VS Code Telemetry Docs, GitHub Copilot Quality Research, ACM Copilot Correctness Study

---

Code Training Data Lawsuits and Regulation

Doe v. GitHub (Filed Nov 2022, ongoing through 2026)

The class-action lawsuit alleges GitHub Copilot violates the DMCA, breaches open-source licenses, and monetizes developers' intellectual property without consent. Potential statutory damages exceed $9 billion.

Court rulings so far: Judge Jon S. Tigar dismissed the DMCA 1202(b) claim, ruling that Copilot's output is not "identical enough" to constitute a DMCA violation — a narrow interpretation requiring exact replication. This decision was appealed to the Ninth Circuit (September 2024). As of January 2026, discovery is ongoing.

The outcome will set precedent for whether AI training on publicly available code constitutes fair use.

EU AI Act (GPAI rules effective August 2, 2025)

All GPAI model providers must:

• Publish a "sufficiently detailed summary" of training content (template published July 24, 2025)
• Maintain a copyright compliance policy throughout each model's lifecycle
• Respect machine-readable rights signals (robots.txt)
• Provide a contact point for copyright holder complaints

Open-source models are not exempt from copyright and training data summary obligations. Enforcement begins August 2, 2026, with fines up to 3% of global annual turnover or EUR 15 million.

Stack Overflow Data Licensing

Stack Overflow's monthly question volume collapsed from 200,000+ at peak (2014) to fewer than 4,000 by December 2025 — a 78% year-over-year drop. The platform now primarily monetizes through Stack Internal (enterprise AI product, 25,000 companies) and data licensing to AI companies. Stack Overflow's CEO stated that LLM developers violate their terms of service by failing to attribute individual contributors under the Creative Commons license.

Developer trust in AI accuracy fell to 29% (from 40% prior year) even as 84% report using or planning to use AI tools (Stack Overflow Developer Survey 2025).

Sources: Doe v. GitHub Patent AI Lab, Saveri Law Firm Copilot Litigation, EU GPAI Code of Practice, Stack Overflow Decline, TechCrunch Stack Overflow AI Data Provider

---

Developer Tool Spending

According to a DX survey of 275 engineering leaders (2025):

• 38.4% spend $101-$500 per developer/year on AI developer tools
• 10.5% spend $501-$1,000/dev/year; another 10.5% spend over $1,000/dev/year
• Full-stack AI tool access costs $500-$3,000+/dev/year
• 46.5% of leaders allocate 1-3% of total engineering budgets to AI tools
• 86% of leaders feel uncertain about which tools provide the most benefit
• 40% lack sufficient data on adoption and impact to build an ROI case

By 2026, AI tool spending is shifting from discretionary/exploratory to a recurring budget line item with clear ROI expectations. Global IT spending is expected to top $6 trillion for the first time in 2026 (Gartner).

Sources: DX 2026 AI Tooling Budget, DX Budget Podcast, Splunk IT Spending

---

Vulnerability Score

Overall vulnerability: Medium-High.

The category splits into sub-segments with different vulnerability profiles. AI coding assistants are highly vulnerable (low network effects, low switching costs, open-weight model alternatives). Deployment platforms are moderately vulnerable (technical complexity is the primary barrier). Source code hosting is the hardest to displace — GitHub's 180M-user network effect is formidable, though Gitea/Forgejo/Codeberg prove self-hosted alternatives are technically viable.

---

The Your 99 Blueprint

The developer tools category is broad. The highest-impact entry point is an AI-augmented development environment — combining code editing, AI assistance, and deployment in a user-owned package.

Revenue model: $12/month for individual developers. $25/user/month for teams. Built on open-weight models (DeepSeek-Coder, Qwen2.5-Coder, CodeStral) with optional cloud model access. Self-hosted deployment supported.

Draft Contribution Map:

Economics at scale:

(Assumes $12/month individual average, ~37.5% compute costs, ~5% operating costs, 1%/10%/89% split)

Key differentiator beyond ownership: No telemetry by default. Local-first AI completions with open-weight models. Code never leaves your machine unless you choose cloud inference. Full transparency on what data flows where. Community governance over any model training decisions.

Minimum viable feature set: VS Code-compatible editor (fork VSCodium as base), local AI code completion via open-weight models, self-hostable Git forge (Forgejo-based), basic deployment pipeline. Phase 2: Cloud AI model access (user-controlled), team collaboration, CI/CD. Phase 3: Community-curated extension marketplace, specialized domain models trained with community consent.

---

Open Questions

• GitHub's 180M-user network effect is the elephant in the room. Is the strategy to compete with GitHub directly, or to build complementary tools (AI assistant, deployment) that work alongside it while the self-hosted forge grows?
• Local AI model quality is the critical variable. Can DeepSeek-Coder or Qwen2.5-Coder match Copilot's acceptance rate (currently ~55% faster coding, 70% problem-solving rate) for everyday tasks? The gap is narrowing but not yet closed.
• Compute costs for AI inference are the primary expense. At 500,000 users doing regular AI-assisted coding, what is the real infrastructure cost? Does local-first inference change the economics meaningfully?
• Should this be one integrated product or a suite of interoperable tools? The all-in-one approach (like Cursor or Replit) vs. the composable approach (like the current open-source ecosystem) involves real trade-offs in user experience vs. flexibility.
• Developer tools have a unique advantage: the users are builders. The community itself can contribute extensions, integrations, and improvements. How does the Contribution Map account for this without creating perverse incentives?
• The EU AI Act's training data requirements (effective August 2025, enforced August 2026) create an opportunity: a development platform that is transparent about training data by design has a regulatory advantage in Europe.