AI Wearables - The Hardware Strategy for AI Defensibility
AI Wearables - The Hardware Strategy for AI Defensibility
Why Rabbit chose to build hardware instead of an App.
Is AI defensible?
That’s the latest question in tech.
AI technologies develop so fast that many question whether AI-first companies can build real moats. A new "10X" startup today, may not be that impressive tomorrow.
Value creation doesn't always mean value capture.
Defensibility ultimately determines long-term value capture. - Andreesen Horowitz
Example: many companies who have built the "ChatGPT for X" may no longer have a product advantage because now OpenAI has plugins and the equivalent of an "App Store."
That week even raised the question of whether foundation models are defensible. A team of Stanford students took a very clever approach of starting with LLaMA 7B, an open source LLM, and fine-tuning it on 52,000 instruction-following demonstrations from OpenAI’s text-davinvi-003. The resulting LLM, dubbed Alpaca 7B, can run on a MacBook and has roughly comparable performance to OpenAI’s GPT-3.5, which relies on massive cloud compute. It took OpenAI 4.5 years and over ~$1B raised to launch GPT-3. Alpaca? Training costs totaled less than $600 (not a typo). It’s a very big deal. -
In this post, we'll look at:
The strategy at play: build a device, not an app
Why it matters for defensibility
3 additional examples of AI wearable startups
First, what is a moat? competitive advantage? defensibility?
Competitive advantage - The leverage a business has over its competitors.
Defensibility - The ability to maintain a competitive advantage.
Moats - The economic or market moats represent long-term defensibility.
“Competitive advantages” help your company become successful. “Defensibility” helps you stay there. Both add value to your business. - NFX
The Strategy
The first wave of hardware tech with AI started a decade ago with smart speakers such as Alexa and Google Home. The 2nd wave is now wearables with AI.
The strategy - build a device and accompanying software instead of another app.
Let’s take a look at Rabbit, your new AI pocket companion that decided to build a wearable instead of an app.
This was a strategic business decision for them with a significant driver being defensibility.
The strategy helps make AI-first companies more defensible while simultaneously opening options for how the technology is used. New hardware = new possibilities. Hardware = harder to build = harder to copy.
Why it Matters
AI-first startups are hard to defend as we outlined before. I believe we'll start to see more personal devices and AI delivered via hardware in general as AI proliferates.
Hardware is more defensible for the following reasons:
Higher barrier to entry/copy - Building hardware & software is more complex, harder, and requires additional skill sets, thus more defensible.
Reduced competition - The Rabbit won't need to compete with every app in the App Store. Though, the trade-off to this is competing for space in your pocket.
Higher switching costs - Switching to another app is easier and less expensive compared to switching hardware.
Optionality - Specialized or purpose-built hardware could be added that Apple may not want to add to the iPhone. Thus, other apps would be limited to the hardware that Apple can provide. For example, Hu.ma.ne's device has a small projector that displays information in the palm of your hand.
Better control of IP (Rabbit’sLarge Action Model) - Apple won't be able to see Rabbit's code.
AI requires a new approach to how we interact with the technology.
Examples in the Wild: AI Wearables
Rabbit
A personal AI device that leverages LAM (Large Action Model) to understand human intentions and to get things done for you.
WHAT!?! Thought we were all talking about LLMs...
Here's the difference - with most AI apps, you provide an input and you get an output with one app. Ex. "What are the best interview questions to ask "
But, what if you want an output that requires interacting with multiple apps and steps? Ex. "I'm planning a business trip to San Francisco. Please book the optimal plane ticket (based on time, convenience, and cost), and hotel, and prearrange an Uber pick-up to and from the airport. Update my calendar with the itinerary. "
This would require interacting with multiple apps: Airline, Expedia (for hotels), Uber, Google Calendar...
Rabbit uses a LAM to operate apps on your behalf. You don't have to open those apps or do anything else, just push a button and give it instructions in natural language - like a walkie-talkie. The cool part about this is that you can also teach it new actions.
The Rabbit is AI hardware that replaces app-based operating systems by introducing a new experience between humans and machine.
The Numbers:
Funding: $30M
Cost: $199
Subscription: TBD
Total revenue: $4M (20,000 sold as of 11/11/2023)
Investors:
You can check out the announcement and pitch here.
Rewind AI Pendant
Rewind is a personalized AI powered by everything you've seen, said, or heard as you use your computer. (Note: I'm a small angel investor.)
Rewind currently has a desktop app that records everything on screen, heard, or what you've said. The data is compressed and stored on your computer. You can ask Rewind.ai about anything you've captured. Ex. "What was the topic of conversation and next steps with Johnny back in March?"
The Pendant extends this capture ability to your person so you can rewind on a greater surface area of captured data around what you do.
The Numbers:
Funding: $33M
Cost: $59
Subscription: $19/month (if paid annually)
Total revenue: not disclosed
Investors: First Round, Andreessen Horowitz, NEA
Humane AI
Hu.ma.ne’s AI Pin is built by ex-Apple designers and engineers and is taking a different approach. Eliminate the screen and use a projector instead.
Not only is there a projector from a differentiation standpoint, but you can also use gestures to control the device in addition to spoken language.
AI gets to know you and helps filter information based on what's most important - like trusted vs. non trusted contacts.
There's a ton of additional use cases:
Take photos of the moment
Scan a fruit to get nutritional value
Messaging
Interpreter
The Numbers:
Funding: $230M
Cost: $699 for base model
Subscription: $24/month (if paid annually)
Total revenue: not disclosed
Investors: Kindred Ventures, Microsoft, Volvo Cars Tech Fund, Tiger Global, Qualcomm Ventures, and OpenAI CEO, Sam Altman
Tab
Tab has the same use case as Rewind AI's pendant, but comes in a much nicer aesthetic.
Listen, transcribe, and then provide insights that are specific to and for you.
The Numbers:
Funding: $1.9M seed
Cost: $600
Subscription: $50/month
Total revenue: $100,000 in orders so far
Investors: Caffeinated Capital, Rief Capital, Cory Levy, and other notable investors such as Perplexity AI's Aravind Srinivas.
Key Takeaways
We’re in the early innings of AI Wearables. There will be much more to come.
Startups are using new devices to weave AI into our daily lives and make the tech behind it more defensive.
Hardware is more defensible for many reasons; optionality, additional complexity to copy, and more expensive switching costs.
AI is enabling new types of devices and services.
AI wearables are selling, but they still have a long way to go before they reach the mass market
What I’m Reading
Defensibility in the Age of AI
How to Build a Defensible AI Startup - CRV
Rabbit sells out two batches of 10,000 R1 pocket Ai companions over two days - The Verge
YouTube: Introducing r1 - Rabbit
GTM Foundry is for anyone bringing a new product or technology to market. We cover strategies, frameworks, and trends that shape the technology & GTM frontier. Subscribe for semi-weekly posts.