HARDWARE

We give away the scanner because the real business is every crown, bridge, and veneer that flows through our platform for the next decade. The scanner is the door. The lab is the house.

We built a chip with 2.6 trillion transistors. The next largest has 80 billion. We're not incrementally better. We're architecturally different.

Moore's Law said chips get smaller. We said: what if they get bigger? Everyone laughed. Then they saw the benchmark results.

NVIDIA has the market. We have the physics. A single wafer-scale chip replaces thousands of GPUs. The math works. It just takes time for the market to catch up.

Everyone in AI is trying to make software smarter. We said: make the chip bigger. Literally. Our chip is the size of a dinner plate. That's not a metaphor.

NVIDIA builds GPUs for everything. We built a chip for one thing: running language models as fast as physically possible. Specialization always beats generalization.

I invented the TPU at Google. The Tensor Processing Unit. Then I left because I knew I could build something faster. That chip is the LPU.

We raised $930 million. Nearly a billion dollars. For a hardware company that never figured out how to make a profit on a single product. That's a special kind of failure.

We made the Jambox — a Bluetooth speaker so beautifully designed that people put it on their coffee tables like art. We were a hardware company that cared about aesthetics. That was rare in 2010.

I wanted to build the Keurig of juice. A pod system where you get fresh, cold-pressed juice at home with the push of a button. That was the vision. The execution got complicated.

The press has 8 tons of force. Eight tons. More than a Tesla Model S weighs. That engineering is real. That engineering cost $120 million.

We raised $1.1 billion. For a chip company. In a market where NVIDIA has 90% share. Either we're brilliant or insane. The benchmark results suggest brilliant. The market cap gap suggests we have work to do.

We built a chip from scratch specifically for AI. Not a GPU repurposed for AI — a chip designed from the ground up for dataflow computing. That distinction matters when you're running trillion-parameter models.

The internet model for hardware is simple: sell the device at cost, build a relationship with the user, and make money on services for the next ten years. The phone is not the product — the user is the product.

A Lime scooter lasts 3-4 years now. The first generation lasted 3 months. That improvement in hardware is why we are profitable and everyone else is dead.

Move fast and break things works for software. It does not work when the things you are breaking are actual physical scooters.

a16z invested because they believe AI-powered hardware will define the next decade. We are the proof point.

Content is our real product. The hardware is how you access it.

Quality issues are real. We're fixing them. Building hardware at scale is harder than building an app.

Hardware is hard. That is why it is worth doing. Software startups can be copied in a weekend. A chip takes three years and a billion dollars.