This isn't political. This is simple: support the people who live in America. Build here. Hire here. Make things that last.

Great people are everywhere, and of course we sponsor visas—two of our co-founders are here on visas themselves. But our hiring focus is America, with the aim of global impact. Taking care of home first isn't about closing doors—it's about building something strong enough to keep them open.

A strong, honest America doesn't weaken the world—it lifts it. Our stretch goal: manufacture everything in the United States.

Every GPU still runs the first microprocessor's loop: fetch, move, compute, move again. We renamed the traffic jam "parallelism." A trillion dollars of silicon now exists to hide one bottleneck—the distance between memory and math. So we stopped building a chip that loads kernels, and built a chip made of kernels.

Programmable atomic ML math tiles, wired so tightly to the data flow that the computation doesn't run, it unfolds. The layer is the circuit. Operands arrive on wires; the whole window multiplies and reduces in flight: no fetch, no decode, no scheduler, nothing between the model and the math. 10x fewer circuits. The large-scale public demo is in progress.

Every AI chip on earth is starving. The math is fast, but the weights sit in memory that costs more than gold per gigabyte, is sold out into the 2030s, and has to squeeze through a narrow door of wires at the edge of the die. LightFall removes the door: we deliver a model's weights as light, landing directly on the face of the chip.

No HBM. No DRAM. More bandwidth than a flagship GPU's entire memory system at a fraction of the power. Models swap like cartridges.