Another huge financing has emerged in the AI computing power race.

Recently, AI computing power company Lightmatter secured a new round of $400 million (approximately RMB 2.845 billion) in Series D funding. This round was led by T. Rowe Price, with follow-on investments from existing shareholders such as GV and Fidelity. Following this round of funding, Lightmatter's valuation has reached $4.4 billion (approximately RMB 31.291 billion).

Lightmatter aims to address the issues of chip computing efficiency and significant energy consumption faced by AI computing centers. It hopes to leverage silicon photonics technology to connect chips for AI deep learning training, accelerating processing speeds while reducing energy consumption. The principle behind this is that Lightmatter's optical interconnect layer enables hundreds of GPUs to work synchronously, simplifying the training and operation of expensive and complex AI models.

Founded in 2017, Lightmatter is headquartered in California, USA, and its founder is Nicholas Harris.

Nicholas obtained his Ph.D. in Electrical Engineering and Computer Science from MIT, specializing in photonics research. His doctoral thesis was titled "Programmable Nanophotonics for Quantum Information Processing and AI." During his doctoral studies, he developed silicon-based integrated photonic chips that transmit and process information using light instead of electricity. Prior to his doctoral studies, Nicholas worked as a research and development engineer at Micron Technology, where he saw the potential of high-performance computing.

Specifically, Lightmatter aims to address the efficiency issues in the interconnect layer of chips (GPUs or CPUs). From a physical perspective, training AI involves connecting thousands of GPUs, but these GPUs are not scattered like solar panels on a hillside; instead, several GPUs are grouped together on a rack, and countless racks are connected through signal switching devices to form a super-large-scale computing center. When running large AI models, there is a time lag in the process of data input and output among these GPUs, with some completing tasks before others. The faster the interconnect speed, the faster the computing center will operate. Lightmatter intends to eliminate this time lag and enable GPUs to run synchronously using silicon photonics technology.

Giants such as Microsoft, Amazon, xAI, and OpenAI have shown significant interest in Lightmatter's solution. Current computing centers are massive structures filled with GPUs and consume enormous amounts of electricity. It has been said that the ultimate competition in AI lies in the competition for electricity. Furthermore, future large AI models may require even more GPUs, and without more efficient solutions, we could see computing centers the size of small towns filled with countless GPUs. Imagine the number of GPUs and electricity required for such centers, and the prospect of building entire "computing cities" in the future. Some predict that by 2040, approximately 80% of the world's energy will be used for data centers and computing.

Solutions that can substantially save electricity and accelerate computing efficiency are highly coveted by these giants. As Moore's Law slows down and traditional silicon-based computing technology approaches its physical limits, photonics technology has emerged as a potential savior for the continued progress of computing technology. Lightmatter's solution replaces standard electrical connections with optical fibers, achieving bandwidths of hundreds of gigabits, far exceeding current computing center capabilities.

Lightmatter's flagship product is called Passage (pictured above), which utilizes three-dimensional stacked photonic chips to move data between processors. This approach significantly enhances the bandwidth and performance of AI clusters while reducing power consumption. By integrating thousands to millions of advanced chips into a cohesive system, Lightmatter is currently collaborating with chip manufacturers and cloud service providers to enable large-scale deployments. Nicholas notes that since Lightmatter's devices operate on silicon, they can be produced by existing semiconductor manufacturing facilities without requiring significant process changes.

Nicholas also hinted, "This may be our last round of private funding," which could suggest that Lightmatter is preparing for an IPO.

Earlier this year, Lightmatter appointed Simona Jankowski, a former NVIDIA executive, as CFO, further demonstrating its ambition to go public.

Last week, Xscape Photonics in New York also secured $44 million in Series A funding, also leveraging photonics technology to address the challenges of energy, performance, and scalability in AI data centers. The round was led by IAG Capital Partners, with follow-on investments from Cisco Investments and NVIDIA, among others.

This article does not constitute investment advice.