The largest AI infrastructure company tracked in the wiki. Builds frontier models (Gemini), open-weights models (Gemma 4), agentic dev environments (Firebase Studio), creative tools (Google Stitch), and the underlying inference platforms (Vertex AI, Genkit). Google AI is split across Google DeepMind (research, model development) and Google Cloud (infrastructure, dev tools, deployment).
- Founded: Google 1998; DeepMind 2010 (acquired 2014)
- Corporate structure: Alphabet Inc. subsidiary (publicly traded)
- AI org structure: Google DeepMind (research + model dev) + Google Cloud (infrastructure + dev products) + Google Labs (experimental product surfaces) + Waymo (autonomous vehicles, sister Alphabet bet)
- Sites: deepmind.google, ai.google.dev, cloud.google.com, labs.google
Products and tools tracked in this wiki
Models
- Gemini — frontier model family. Per Matthew Berman’s framing: best for search and deep research, with unique video ingestion (the only frontier model that natively reads video at the time of the Berman survey).
- Gemma 4 — open-weights model family. Apache 2.0. Sizes E2B (2B effective) → E4B (4B effective) → 26B A4B (MoE, ~4B active) → 31B (dense). 31B ranks #3 globally on Arena AI. Hardware sizing covered in gemma-4-vram-requirements (Gemma4Guide reference).
Dev tools
- Firebase Studio — cloud-based agentic dev environment, rebranded from Project IDX at Google Cloud Next 2025. Combines Firebase, Genkit, and Gemini into a single full-stack workspace. Free tier with 3 workspaces (30 for dev program members).
- Google Stitch — voice-to-UI design tool that exports agent-readable
design.md. Featured in Nate B Jones’s “command-line creative stack” thesis as the canonical example of design moving to the command line. - Genkit — Google’s GenAI framework with Python/Go support, RAG, multimodal orchestration. Bundled into Firebase Studio. Not yet a standalone wiki page.
- Vertex AI — Google Cloud’s enterprise AI platform; provides access to Gemini, Imagen 3, and open-source models including Llama and Mistral. Underpins Firebase Studio’s model selection.
Adjacent (Alphabet)
- Waymo World Model — simulation and prediction at scale. Originally framed for self-driving but Fireship’s death-spiral analysis cites the same primitives as obsolescing traditional SaaS dashboards (forecasting, logistics, risk, operations). Not a wiki entity yet but referenced in the SaaS death-spiral analysis.
Cross-cutting threads
- The command-line creative stack (Nate B Jones’s thesis): Google Stitch (voice→UI, design.md), Remotion, Blender MCP. Stitch is Google’s contribution and the wiki’s canonical example of “design follows development to the command line.”
- Open-weights leadership: Gemma 4 is one of the most-tracked open-weights families in the wiki, with its own hardware sizing reference. Google is the only big-tech-frontier lab releasing open weights at this scale (vs Meta’s Llama which is closer to corporate-strategic open weights, and Anthropic/OpenAI which release no open weights).
- Agentic coding platform proliferation: Firebase Studio is Google’s entry in the fragmenting agentic-coding market alongside Claude Code, Cursor, Augment Agent, Archon OS, and DeepCode.
How Google differs from OpenAI and Anthropic in the wiki’s coverage
| Anthropic | OpenAI | ||
|---|---|---|---|
| Frontier model | Gemini | Claude | ChatGPT / GPT-5 |
| Open-weights | Yes (Gemma 4) | No | No |
| Coding agent | (none direct; uses Gemini via Firebase Studio) | Claude Code | Codex |
| Cloud-IDE play | Firebase Studio | None | None |
| Creative tool | Google Stitch | None | DALL-E / Sora |
| Wiki-tracked policy positions | None | None | Industrial Policy paper |
| Adjacent bets | Waymo (autonomy), DeepMind (research) | None tracked | None tracked |
Google’s breadth is its distinctive feature in the wiki — every other org tracked is a single-product or single-bet company by comparison. Google is the only entry that simultaneously ships frontier models, open-weights models, cloud-hosted dev environments, creative tools, autonomous-vehicle infrastructure, and a full enterprise cloud stack.
Editorial framing the wiki applies to Google sources
- First-party Google content (the Gemma 4 launch post, Firebase Studio Cloud Next demos, Stitch product launch) is positioning from an interested party — same conventions as OpenAI/Anthropic.
- Third-party coverage of Google products (Berman on Gemma, WorldofAI on Firebase Studio, Nate B Jones on Stitch) is more credible for capability claims, but check for affiliate/access biases as usual.
- The “Beats Cursor + Bolt” framing in the Firebase Studio coverage is consistent with the WorldofAI channel pattern of new-tool-vs-incumbent positioning — discount.
- Google’s open-weights play is genuinely useful and not self-serving in the same way the Industrial Policy paper is — Gemma 4 has no equivalent regulatory-moat structure; it’s just released under Apache 2.0.
See Also
- gemini, gemma-4, gemma-4-vram-requirements — model family + hardware sizing
- firebase-studio, google-stitch — dev and creative tools
- saas-death-spiral — references Waymo World Model + Firebase Studio context
- anthropic, openai — the other two frontier-lab orgs
- Source: Google Drops Gemma 4
- Source: Gemma 4 VRAM Requirements
- Source: Firebase Studio Walkthrough
- Source: A Markdown File Just Replaced Your Design Meeting
- Source: SaaS Death Spiral — Waymo World Model context
- Source: Every AI Model Explained