Rabbit Inc. (Rabbit)
Designs and sells the AI handheld device Rabbit R1 based on the Large Action Model (LAM), aiming to simplify user interaction with digital services.
The proprietary Large Action Model (LAM) can learn and execute user actions, but its practical implementation has been poor, leaving the moat fragile and unverified.
Company Overview
Rabbit Inc. was founded in 2020 and is headquartered in Los Angeles, California, USA. It is a startup focused on consumer-grade AI hardware and interaction systems. Founder and CEO Jesse Lyu (Lü Pin) previously founded the Chinese AI voice company "Liulishuo" and has a deep understanding of AI applications. The company's core product, Rabbit R1, was unveiled at CES in early 2024, positioned as an "AI assistant in your pocket," aiming to replace smartphones for tasks such as ordering food, hailing rides, and shopping through natural language interaction. The company has not yet gone public, with total funding of approximately $30 million and investors including Khosla Ventures and others. Rabbit occupies the "terminal and interaction entry point" tier in the AI industry chain and is one of the representative players exploring consumer-facing AI hardware.
Rabbit's ambition is to create a new computing entry point for the "post-smartphone era," but the market performance of its first product, R1, fell far short of expectations, and the industry has regarded it as a case of "failed AI hardware."
The Rabbit R1 is a standalone handheld device priced at $199, featuring a 2.88-inch touchscreen, a rotating camera, a scroll wheel, and side buttons, powered by a customized Android-based system. Its core selling point is the built-in "Large Action Model" (LAM), which claims to learn from observing user operations and autonomously execute complex in-app tasks (such as booking restaurants, playing music, etc.) without requiring APIs or developer adaptation. The device must be bound to a Rabbit Hole cloud account. The initial batch of 10,000 units sold out within 24 hours, but its reputation subsequently collapsed: slow system response, low command recognition accuracy, privacy controversies (users are required to authorize LAM to simulate operations), and a wave of returns and negative reviews. Sales declined sharply after Q2 2024.
| Product Line | Revenue Share | Core Customers | Gross Margin (Estimated) |
|---|---|---|---|
| Rabbit R1 Hardware | ~100% | Early tech enthusiasts, early adopters | Approximately -10% to 5% (due to excessive costs) |
Other business attempts: The company had planned to launch a LAM API for developers and a "Teach Mode," but neither has shown signs of commercialization to date.
Technical Moat
Moat 1: Large Action Model (LAM)
Rabbit claims to have independently developed a LAM (non-traditional LLM) that directly controls third-party app interfaces through "intent understanding + action sequence generation." The underlying technical model is purportedly capable of understanding the GUI layout of web pages or apps and outputting actions such as simulated clicks and inputs. However, actual products have been found by users and evaluation organizations to be mostly LLM-based rule engines plus manually predefined scripts, rather than true autonomous learning. The real depth of this moat is seriously questioned.
Moat 2: Reverse Proxy and Screen Understanding Capability
Rabbit has filed a patent for "vision-based user interface automation," but it has not yet been granted. The R1 device relies on cloud-based parsing of user screen screenshots before simulating actions, resulting in high latency and a high error rate. Like Humane's AI Pin, this "bolt-on" interaction technology is constrained by system-level permissions in closed ecosystems such as the iPhone, making the moat very fragile.
The current technical route is essentially the consumerization of "cloud-based automated test scripts," jokingly referred to by the developer community as "buying a 199-dollar phone remote control."
| Dimension | Data |
|---|---|
| Global AI hardware terminal market share | <0.1% (estimated) |
| Industry ranking | Outside the top 7 (far below smartphones and smartwatches) |
| Main competitors | Humane AI Pin (priced at $699, also met with failure), Meta Ray-Ban smart glasses (shipments of approximately 1 million units), Apple Vision Pro (different positioning) |
| Downstream customers | Geeks and early adopters (only about 100,000 users as of end of 2024) |
| Key competitive factors | Ecosystem compatibility, AI interaction accuracy, privacy and security, pricing |
Competitive Landscape Analysis: The consumer-facing AI hardware sector has yet to produce a clear winner. Both the Humane AI Pin and Rabbit R1 were named "most disappointing tech products of the year" by the media in 2024. Traditional giants such as Apple (Siri + iPhone ecosystem), Google (Pixel + Gemini), and Meta (Ray-Ban Meta glasses) hold an absolute advantage through integrated software-hardware offerings and existing user bases. Startups face a "chicken-and-egg" dilemma: without enough users to optimize their models, the poor model experience discourages users from purchasing.
Financials and Growth
| Metric | Data |
|---|---|
| Revenue (2024 estimate) | Approximately $20 million (all from R1 pre-sales and early shipments) |
| Gross margin | Approximately -5% to 5% (hardware BOM cost around $120, negative after adding R&D, cloud, and return costs) |
| Net margin | Severe losses |
| Core growth logic | Heavily dependent on sales of the first product R1, but product reputation has collapsed and growth has stalled |
Financial background: In early 2024, Rabbit completed a $10 million seed round and a $20 million Series A round. But after the product failed, the company burned cash at an alarming rate (cloud inference costs + return processing costs). In mid-2024, Jesse Lyu admitted that "R1 is not a perfect product" and promised software updates, but market confidence was already lost. If the company cannot launch a substantial second-generation product or pivot to B2B by 2025, the risk of running out of cash is extremely high.
Key Risks:
- Product Failure Risk: The core experience of R1 fails to meet expectations, leading to high return rates, severely damaging brand reputation, and making subsequent fundraising extremely difficult.
- Technology & Patent Risk: The LAM technology has been questioned as "pseudo-innovation," lacking true moats; it holds no advantage over LLM providers such as OpenAI/Google, and faces potential patent infringement risks.
- Ecosystem & Market Risk: Tech giants like Apple and Google possess OS-level authority and can shut down "middleware-type" hardware such as Rabbit at any time. Consumer demand for standalone AI hardware has yet to be validated.
- Capital Chain Risk: Revenue is single-source and losses persist. If new funding fails to materialize, the company may go bankrupt by the end of 2025.
Core Investment Thesis / Industry Value Summary: