The Signal
Zapata CTO calls for faster quantum use-case screening
Yudong Cao proposes using a predictive model to screen quantum applications before committing to detailed resource estimates. In a QuEra post, he argues that a three-year benchmarking effort produced about ten complete estimates from 150 to 200 initial candidates.

A three-year quantum benchmarking effort produced detailed resource estimates for roughly ten applications from an initial pool of 150 to 200 candidates, prompting a call from Zapata CTO Yudong Cao for a faster way to assess use cases. In an opinion post published by QuEra, Cao proposes screening a wider range of problems with a predictive model before spending years on detailed estimates.
QuEra published the argument on November 12, drawing on Cao's talk at the QuEra Quantum Alliance 2026 Annual Meeting. The company describes the critique as the speaker's assessment of an industry-wide practice, not as a QuEra position.
From interviews to estimates
Cao worked across the DARPA Quantum Benchmarking program, from identifying candidate applications to estimating the resources they would need on a fault-tolerant machine. Its starting question was whether the US government should procure quantum computing capability and whether the cost would be worthwhile.
Zapata, L3Harris and other enterprises interviewed domain experts across seven to nine fields about where quantum computers might help. Those conversations yielded 150 to 200 candidate application instances, which Cao described as initial explorations rather than scientifically established results.
Chemistry and differential equations emerged as areas worth taking further, according to the account. The roughly ten instances that received end-to-end estimates were assessed down to physical qubit counts and wall-clock runtime.
A simplified ship hull
One example in Cao's critique concerns computational fluid dynamics and the problem of estimating drag on a ship's hull. The account describes reformulating the problem into a form a quantum algorithm can approach as a research project in itself, before resource estimation can begin.
The resulting end-to-end estimate addresses a simplified version in which the hull is a sphere, rather than the original ship-hull problem. The associated paper was still being updated two years into the work, while independent improvements were appearing in the literature, according to the post.
Cao also questioned how the resulting resource-cost plot should be read. The work behind each point stopped when the program ended, he said, rather than when the analysis had been fully optimized.
The plot was a snapshot as of March 2025, according to the post. Cao argued that more work could produce improvements of orders of magnitude, but that the team had to deliver its results when the program ended.
Estimates and decisions
In later advisory work with enterprise clients, Cao encountered a communication problem: resource estimates expressed in qubits and hours did not necessarily mean much to people deciding whether to fund a program. His example was a logistics executive faced with an application estimate of 5 million qubits and five hours.
The post also argues that an estimate can change by orders of magnitude following a new result in error correction, compilation or algorithm design. That makes the plotted figure a reading from a particular point in time, rather than a finished assessment.
A broader screen
Cao's proposed model would use the detailed data already produced by benchmarking programs to generalize across a larger search space. The goal is to identify problems that are classically hard and quantum easy in weeks rather than years, then concentrate expensive analysis on the more promising candidates.
Estimates should also be updated as hardware roadmaps and error-correction approaches change, according to the proposal. The approach draws on an analogy with pharmaceutical screening, where a broad initial search precedes more intensive testing of selected candidates.
QuEra's roadmap puts its planned Libra system on Amazon Braket in 2028, with 256 logical qubits built from over 10,000 physical qubits, according to the company. The post says its co-design work with customers includes identifying quantum-relevant workloads and testing them against classical baselines before committing to a larger program.