China’s Photonic Quantum AI Chip Beats NVIDIA by 1000x

There has been a big breakthrough in the world of AI hardware. A group of Chinese researchers made a photonic quantum chip that is already being used in data centers. This new chip says it can speed up some AI tasks by up to 1,000 times compared to NVIDIA GPUs that are already on the market. This is not a technology that will be around in the future. It is now being used in important fields like aerospace, biomedicine, and finance. This development changes the race for AI computing that is very powerful.

A Big Step Forward in Quantum Hardware

CHIPX, which stands for Chip Hub for Integrated Photonix Explorer, and Turing Quantum worked together to make this new chip. They worked together to make what they call the world’s first truly scalable optical quantum chip for use in industry. This is a big step because it makes quantum computing useful in the real world.

The chip itself is tiny. It has a 6-inch thin-film lithium niobate wafer. This wafer has more than a thousand optical parts on one piece of silicon. Photonic integration makes it possible for this small size. It keeps everything small but still strong enough to use both quantum and classical computing methods.

How Light Affects Computers

The way this chip works is what makes it special. It doesn’t use electric signals; instead, it uses light to move qubits and do math. There are a lot of reasons why this is a big deal:

• Less Heat: Light doesn’t make heat the same way electricity does. Data centers have a hard time with heat and using a lot of power. A light-powered chip can help fix these issues.

• More Space: Light takes up less physical space than electrical parts. This makes it possible to fit more parts into a smaller space.

• Photons, or light particles, move faster and lose less energy, which makes them faster and more efficient. This means that data moves more easily and quickly.

A lot of power is used by modern data centers. When training AI models, they also make a lot of heat. A chip that moves information with light directly solves these big problems.

The Claim of 1,000x Speed

People are most interested in the claim that the chip is 1,000 times faster than NVIDIA GPUs for some AI tasks. The South China Morning Post and the developers’ statements say this number. They say the chips make it this much faster to solve hard problems. They also say that the chip is already in data centers.

You should know that this speed boost doesn’t work for every task. But for certain AI problems, especially those that need quantum-like parallelism or very fast light pathways, this huge speed boost makes sense. This means that for these specific tasks, the chip can do much better than regular hardware.

Making it Bigger for Real-World Use

One big problem with quantum systems is that they are too big and complicated. They are often very big, breakable, and take a long time to build and set up. This new photonic chip fixes that.

CHIPX says that systems that used to take six months to build can now be set up in just two weeks. It’s amazing how quickly a quantum-like system can be set up.

This is not a complete universal quantum computer. It won’t be able to break all encryption or send emails through space. It is instead photonic quantum hardware that is made for fast AI and data processing. Chip combines quantum and classical logic. The chip uses light to work like a programmable accelerator. It uses the properties of light to process data, such as:

• Timing

• Phase

• Color

• Distribution

It puts together thousands of optical parts into one design. Many researchers around the world have been impressed by this “monolithic” or “integrated” design. It puts everything on the 6-inch wafer. It doesn’t need any extra parts or modules.

At the 2025 World Internet Conference Wuhan Summit, the team that made this chip won the Leading Technology Award. There were more than 400 entries from 34 countries, so this was a tough competition. Winning this award shows that China is ahead of the rest of the world in at least one area of photonic quantum hardware, especially when it comes to making these chips.

Making a Production Line

CHIPX started a pilot production line to show that they are serious about making this technology available to everyone. This line started up at the beginning of this year. It can make 12,000 wafers every year. About 350 chips can be made from each wafer.

This isn’t a lot by chip-making standards, but it’s a lot for quantum-classical photonic chips. Experts thought China would be able to do this much later.

The developers say that making the game is still hard. The materials, such as thin-film lithium niobate, are fragile. They need to make sure that all of the wafers are the same and that more good chips come from each one.

Despite these problems, they have made a full system. This includes designing the chips, making them, packaging them, testing them, and putting the system together. All of these things happen in the same place. This is a big step.

Right now, most photonic chips in Europe and the US are either prototypes or small research batches. Smart Photonics in the Netherlands, for instance, recently switched to 4-inch indium phosphide wafers. Cquantum in California is changing 300 mm silicon photonics to fit its quantum plans. China has already built a working pilot line for 6-inch thin-film lithium niobate, though. This shows that the competition around the world has changed.

The Race for Photonics Around the World

NVIDIA is also putting a lot of money into quantum and optical technology. They are not behind. Quantum optics is a global focus because data centers can’t keep growing by using only electric signals. The amount of power being used is already at its limit. Companies that train big AI models are spending more on electricity than on hardware. For these reasons, both sides of the world are working hard on photonics.

It’s easy to connect these new chips. Researchers say that the design can handle up to 1 million qubits of quantum processing power. These qubits are not the same as those in super-cooled quantum computers, but it’s important that you can stack these chips like GPU clusters. It means that the system can work like a huge processing grid that is based on quantum computing. It works at many levels at the same time: photons, pathways, and wafer integration.

The chip does more than just speed. It also promises to be stable and use little power. Photons don’t get hot like electrons do. This keeps the chip cooler and stops the problems with overheating that data centers always have. Light also sends more information with less energy. This means that more data can be compressed, there is more bandwidth, and there are more ways to send signals without problems. This technology is great for training larger AI models or speeding up the work done in data centers.

Uses in the Real World

Some claims need to be checked by someone else, but the potential is clear. China is already using these chips in some places:

• Aerospace: For complicated design and simulations.

• Biomedicine: For modeling molecules and finding new drugs.

• Finance: For algorithmic trading and finding fraud.

These fields need to be able to solve problems quickly the most. They are the first to use these chips. Photonic chips are easier to add to existing data center setups because they don’t need special cooling.

Advanced Co-Packaging Technology

The new co-packaging technology is a very interesting technical detail. It puts electronic and photonic parts next to each other on the same wafer. This closeness cuts down on delays, interference, and the amount of data that can be moved by a lot. The single design packs everything in tightly. Photons don’t have to go very far. Fewer errors and more data flow happen over shorter distances. This is a higher level of optimization than what is found in high-end traditional chips, but it uses different physics.

A New Age for Quantum Computing in Business

China’s announcement comes at the same time that the US, Europe, and other countries are also stepping up their work in photonics. Quantum computing is in a special place right now where no one company is in charge. Different areas are putting their money on different technologies:

• IBM and Google are both interested in superconducting qubits.

• Silicon photonics is what Cquantum uses.

• Indium phosphide platforms are common in Europe.

• China is now making progress with thin-film lithium niobate.

This time, China isn’t just taking a chance. They have already built the factory line, made the chips, and started using them. Researchers say that this is the first time that optical quantum computers can be thought of as products that can be used in business. This one statement shows a big change. They don’t see it as a lab project anymore. They think it should be in data centers with GPUs and other AI accelerators.

If this idea turns out to be true, the next few years could be very different. We could live in a world where light-based systems do the heavy lifting and silicon-based systems do everything else. NVIDIA GPUs will still be useful. But they may not be the only main choice for a long time, especially since power use keeps going up and businesses look for faster, cooler ways to meet their computing needs.

You can read more about China’s work on photonic chips and future technologies in MERICS’s article on China’s position and Manufacturing Asia’s article on how to increase photonic chip production. The Quantum Insider also has information about technical and production lines.

Conclusion

The launch of China’s new photonic quantum AI chip is a big step forward for AI hardware. This technology solves important problems like power use and speed by bringing quantum-like capabilities from labs to active data centers. The move toward optical quantum products that are good enough for use in industry shows that the global AI competition is changing.

As this technology gets better, it might change the way we handle complicated AI tasks. It is a strong, useful alternative to older systems. The race to build the next generation of AI infrastructure is picking up speed, and light-based computing is clearly the best way to do important tasks.