Framework Momentum: What Developers Are Actually Using
GitHub star counts are an imperfect but leading indicator of developer adoption. Unlike downloads (which can be inflated by bots and CI/CD pipelines), stars represent intentional human interest. Contributor counts and fork rates confirm whether that interest translates to active use. The data below represents public GitHub repository stats as of May 2026.
| Framework / Tool | GitHub Stars May 2026 |
Star Growth Jan 2024 baseline |
Contributors May 2026 |
Trajectory |
|---|---|---|---|---|
| Ollama | 88K+ | +520% | 1,200+ | Accelerating |
| LlamaIndex | 38K+ | +310% | 890+ | Accelerating |
| LangChain | 100K+ | +340% | 3,100+ | Plateauing (high base) |
| Hugging Face Transformers | 140K+ | +280% | 2,800+ | Steady growth |
| AutoGen (Microsoft) | 34K+ | +410% | 1,100+ | Accelerating |
| vLLM | 28K+ | +680% | 720+ | Fastest growing |
| TensorFlow | 185K+ | +18% | 4,200+ | Declining relative |
| PyTorch | 82K+ | +85% | 4,800+ | Stable |
Source: GitHub public repository statistics. Stars and contributor counts approximate (GitHub rounds large numbers). Growth calculated from January 2024 baseline. "Accelerating" = growth rate increasing QoQ; "Plateauing" = growth rate declining from high base. Accessed May 2026.
The fastest-growing category is not orchestration frameworks (LangChain) or training frameworks (PyTorch) — it is inference serving infrastructure (vLLM +680%, Ollama +520%). This reflects a maturation of the market: developers are moving from "how do I build with AI" to "how do I run AI efficiently at scale without paying per-token API costs." This shift toward self-hosted inference has major implications for enterprise AI spending — it suggests a coming challenge to hyperscaler API revenue as organizations build local inference capacity.
Enterprise Purchasing: What Companies Actually Buy
Developer preference and enterprise procurement are not the same decision. Enterprise AI spending is governed by procurement processes, compliance requirements, security review, and existing vendor relationships. The result is a significant divergence between what developers use and what organizations officially license.
| AI Platform Category | Share of Enterprise Contract Value IDC AI Software Market 2025 |
Developer Usage Share Stack Overflow 2025 |
Gap |
|---|---|---|---|
| Hyperscaler AI platforms (Azure OpenAI, Vertex AI, Bedrock) | 61% | 33% | +28pt enterprise premium |
| Specialized AI vendors (Anthropic, Cohere, Mistral direct) | 18% | 22% | Roughly aligned |
| Open-source / self-hosted | 21% | 45% | -24pt developer preference gap |
Enterprise contract value: IDC Worldwide AI Software Market 2025 (summary data). Developer usage: Stack Overflow Developer Survey 2025, "AI tools used in primary work." Percentages are approximate; IDC figures represent contract value, not usage volume. Accessed March–April 2026.
Stack Overflow Developer Signal
Stack Overflow's 2025 Developer Survey (90,000+ respondents) provides the most representative view of developer AI tool usage. Key findings relevant to the stack shift:
- 76% of developers use AI coding assistants (up from 44% in 2023)
- ChatGPT/OpenAI API: used by 64% of AI-using developers
- GitHub Copilot: 55% (highest satisfaction score, 76% "would recommend")
- Hugging Face models: 38% (primarily ML/data science specialists)
- Local models (Ollama, LM Studio): 28% — up from 9% in 2024
- LangChain or similar orchestration: 41% of developers building AI applications
The local model usage jumping from 9% to 28% in one year is the most significant signal in the dataset. It confirms that Ollama's GitHub growth reflects genuine adoption, not just developer curiosity.
Methodology
GitHub data: Public repository statistics for repositories identified as primary frameworks for AI/LLM development. Stars, forks, and contributor counts from GitHub public API. January 2024 baseline from Wayback Machine snapshots and open-source tracking projects (star-history.com). Accessed May 2026.
Enterprise spending: IDC Worldwide AI Software Market 2025 publicly available summary data. "Contract value" represents annual software contract value, not usage. Hyperscaler category includes resold OpenAI API capacity (Azure OpenAI Service) in addition to native hyperscaler AI services.
Developer usage: Stack Overflow Developer Survey 2025. Self-reported tool usage in "primary work" context. 90,000+ respondents, globally representative. Survey conducted January–February 2025. Accessed March 2026.
Gap calculation: Enterprise contract value share minus developer usage share. Positive gap = enterprises over-weight relative to developers; negative gap = developer preference not reflected in enterprise procurement.