Oracle Corporation and software developer Classiq have successfully tested a workflow for generating and simulating quantum code within the OCI cloud infrastructure.
As part of a "proof of concept," the partners implemented a portfolio optimization task based on the Markowitz model for 12 assets. A key feature of the experiment was the use of Classiq's AI agent, which generated executable code from a natural language text prompt in under 15 minutes.
To validate the algorithm, a node from the NVIDIA DGX A100 supercomputer was utilized within Oracle Cloud Infrastructure. Using the NVIDIA cuQuantum library, engineers simulated a 36-qubit circuit, significantly exceeding Classiq's standard limit of 29 qubits.
The calculations were performed on eight A100 GPUs. According to project representatives, the full state vector model for 36 qubits requires approximately 512 GB of video memory.
The application of a hybrid QAOA algorithm with a chain depth of 730 allowed for the processing of a space of 68.7 billion combinations. The entire simulation cycle took five hours, with the final result deviating from the classical market benchmark by only 4.63%.
Classiq emphasized that the project did not aim for quantum supremacy. Its goal was to demonstrate the engineering link between AI code generation and scalable cloud computing for the financial sector and logistics.
There is no evidence of superiority over the best classical methods. This is a demonstration of a technological process, not a breakthrough in computational efficiency.
Recall that in May, researchers from Multiverse Computing announced a quantum improvement of a large language model on IBM hardware.