Vertical Labs

More Intelligence from Every Watt

AI is advancing faster than our ability to supply the compute and power it demands. Today’s infrastructure wasn’t designed for the AGI era. We believe the answer isn’t simply more compute. It’s getting more intelligence from every watt.

A voice agent, a coding agent and a long horizon agent each demand different tradeoffs in quality, speed and cost. Yet much of today’s infrastructure treats them the same. Intelligence is more nebulous than traditional compute.

Models are judged by generic benchmarks while inference is optimized for tokens, throughput and latency. These metrics matter but they tell us little about intelligence delivered on real workloads.

A Frontier for Every Workload

We believe the workload is the unit of AI. Each workload has its own Pareto frontier across intelligence, speed, cost and power.

Finding and advancing it requires optimizing the entire stack: measuring intelligence in production, post training, model selection, inference, scheduling, memory, networking, silicon and power. These layers are deeply interconnected. A quick fix that works for one workload can break another. A speedup means nothing if quality suffers. Every new model launch and chip shifts the frontier but the infrastructure is the same, making continuous optimization essential.

Infrastructure That Learns

We believe AI infrastructure must become adaptive, learning from the workloads it serves and continuously improving how intelligence is delivered. While recursive self improvement has focused on making models smarter, we’re applying the same principle across the entire infrastructure stack.

We’re already seeing 10 to 15× efficiency gains on select workloads. We see a path toward 1,000× gains, redefining the frontier of what AI can achieve. If we can achieve this, we could fundamentally expand the frontier of what future AI systems can accomplish.

Why Us?

We are a research lab focused on the efficiency of intelligence. Our mission is to deliver the most useful intelligence per watt and per dollar, tailored to every company’s workloads.

Our founding team brings together AI researchers and hardware scientists from Amazon, Google, Meta, Anthropic and Microsoft. We’ve built and deployed AI products and systems serving billions of people.

Our north star is AI infrastructure that knows when to think harder, how much compute each task deserves and how to continually learn from every outcome, much like human intelligence.