NVIDIA’s “Quantum Day” Double Hit: Open-Source AI Model Ising Ignites Quantum Stocks, Internal AI Finishes 80 Person-Months of Chip Design in One Night

Author: Claude, Deep Tide TechFlow

Deep Tide Guide: Nvidia released the world’s first open-source quantum AI model family Ising on April 14, “World Quantum Day,” with error correction decoding speeds 2.5 times faster than industry standards and accuracy improved by 3 times.

Quantum concept stocks surged collectively on the same day, IonQ rose 18%, D-Wave rose 15%. On the same day, Chief Scientist William Dally revealed at GTC 2026 that AI has compressed chip standard cell library porting from 8 people over 10 months into a single night on a GPU, with design results surpassing manual efforts.

Nvidia is using AI to accelerate two of the most challenging engineering problems: making quantum computers truly usable, and making GPU design itself faster and better.

On “World Quantum Day” April 14, Nvidia announced the world’s first open-source AI model family for quantum computing, NVIDIA Ising, causing quantum concept stocks to soar collectively. Simultaneously, the company’s Chief Scientist William Dally disclosed the latest progress of AI in Nvidia’s internal chip design process at GTC 2026, with one task’s efficiency increasing by hundreds of times.

Both clues point to the same conclusion: AI is transforming from an “application layer tool” into “the infrastructure of infrastructure,” accelerating downstream industries (quantum computing) and speeding up AI hardware iteration itself.

The world’s first open-source quantum AI model targets two major bottlenecks in quantum computing

According to Nvidia’s April 14 press release, the Ising model family includes two model domains: Ising Calibration and Ising Decoding, addressing the two core bottlenecks in quantum computing deployment.

Quantum bits (qubits) in quantum processors are inherently noisy; the best quantum processors currently have an error rate of about one in a thousand operations. To make quantum computers practically valuable, the error rate must be reduced to below one in a trillion.

Ising Calibration is a 35 billion parameter visual language model that automatically interprets measurement data from quantum processors and makes calibration decisions, reducing calibration time from days to hours. Ising Decoding is a pair of 3D convolutional neural network models (optimized separately for speed and accuracy) used for real-time quantum error correction decoding, outperforming the current open-source industry standard pyMatching by 2.5 times in speed and 3 times in accuracy.

Nvidia Quantum Product Director Sam Stanwyck explained the open-source strategy at the launch: since different quantum hardware manufacturers have distinct noise characteristics, open-source models allow them to fine-tune locally with their own data, improving performance while protecting proprietary data.

Nvidia CEO Jensen Huang’s statement was even more direct. He said AI is becoming the control plane for quantum machines, transforming fragile qubits into scalable, reliable quantum GPU systems.

According to Nvidia, multiple institutions have adopted Ising models early, including Harvard University School of Engineering and Applied Sciences, Fermi National Accelerator Laboratory, IQM Quantum Computers, Lawrence Berkeley National Laboratory, and the UK National Physical Laboratory.

Quantum concept stocks surge collectively, IonQ soars 18% in a single day

On the day of Ising’s release, US-listed quantum concept stocks experienced a collective rally. According to Yahoo Finance data, IonQ rose about 18%, D-Wave Quantum about 15%, and Rigetti Computing about 12%.

This rally was set against a backdrop of deep correction for quantum stocks since the start of the year. As of April 14, IonQ had fallen about 22% year-to-date, D-Wave about 35%, and Rigetti about 23%. The double-digit rebound on that day did not reverse the overall downward trend for the year, but the collective movement was still notable.

It should be pointed out that the driving factors behind this rally were not solely due to the Ising release. On the same day, IonQ announced milestones in quantum networking and a DARPA contract, and Rigetti reported an $8.4 million order from the Indian Center for Development of Advanced Computing (C-DAC). Multiple catalysts combined to amplify the sector effect.

Resonance, an analysis firm, predicts that the global quantum computing market will exceed $11 billion by 2030. The Quantum Economic Development Consortium (QED-C) released a report stating that by 2025, the global quantum market will reach $1.9 billion, with pure quantum companies’ employee growth at 14%.

AI reshapes Nvidia’s chip design process, compressing 80 person-months into a single night

Ising points to external industry acceleration, while internally Nvidia is using AI to reshape its own chip design process.

In a conversation with Google Chief Scientist Jeff Dean at GTC 2026, Nvidia’s Chief Scientist William Dally disclosed several specific cases. The most impactful data involved porting standard cell libraries: whenever Nvidia shifts to a new semiconductor process (e.g., from 7nm to 5nm), about 2,500 to 3,000 standard cells need to be redesigned for the new process. Previously, this required 8 engineers working for about 10 months. Nvidia developed a reinforcement learning tool called NVCell, which can now complete this overnight on a single GPU, with the resulting cells matching or surpassing manual designs in area, power consumption, and latency.

According to Tom’s Hardware, Dally described this process as akin to a “video game fixing design rule errors,” with reinforcement learning excelling at this kind of trial-and-error optimization.

At a higher abstraction level, Nvidia developed internal large language models Chip Nemo and Bug Nemo. These models are fine-tuned based on Nvidia’s 30 years of proprietary data, covering RTL code, hardware design documents, and architecture specifications of all its GPUs. Dally explained that junior engineers can directly ask Chip Nemo questions, saving time from repeatedly bothering senior designers. He described Chip Nemo as “a very patient mentor.”

In circuit optimization, Nvidia also applies reinforcement learning to classic circuit design problems like carry-lookahead chains. Dally stated that AI-generated designs are “completely bizarre solutions that humans wouldn’t think of, but they perform 20% to 30% better than human designs.”

Long way to go before AI independently designs chips

However, Dally also clearly defined the expected boundaries. He said he hopes to achieve end-to-end automation but is still far from that goal.

Currently, Nvidia’s AI chip design remains assistive rather than fully autonomous. AI contributes to standard cell porting, bug classification and summarization, placement and routing prediction, and architecture space exploration, but a complete end-to-end automated process has not yet been realized. Dally envisions a long-term future involving multi-agent models, where different AI systems handle different design stages, similar to human engineering teams’ division of labor.

According to Computer Weekly, Dally and Dean also discussed the impact of AI intelligences on traditional software tools: when AI operates much faster than humans, traditional software tools designed for human users will become performance bottlenecks, necessitating redesigns from programming tools to business applications.