From USB Storage to AI Data Center: How Phison is Solving the Memory Gap
The company is reinventing itself to solve the AI era’s most expensive problem.
For over two decades, Phison Electronics, the company founded by USB (thumb drive) inventor K. S. Pua, has been the silent engine behind the world’s flash storage. But after 26 years of dominating the NAND Flash market, Phison is orchestrating a pivot that moves it beyond “storage.” Their silicon isn’t just in your laptop; it is currently powering NASA’s missions on Mars and being integrated into the next generation of AI data centers.
In a recent keynote, Phison CTO Wei Lin outlined how the company is reinventing itself to solve the AI era’s most expensive problem: The Memory Bottleneck.
The “Space-Grade” Foundation
Phison’s technical pedigree is best illustrated by its presence on the frontiers of exploration.
Mars and Beyond: When NASA’s Perseverance rover captures high-resolution images of the Martian surface, that data is processed and stored on Phison SSDs. “On Mars, we have 100% market share,” Phison leadership noted. This isn’t just a marketing flex—it’s proof that their hardware can survive extreme radiation and conditions that would destroy standard consumer electronics.
The Lunar Shortage: Phison hardware is already adopted by the Artemis program. Ironically, the demand for AI-grade storage on Earth is so high that further lunar data center expansions have been put on hold. Even for space missions, the terrestrial AI boom is the priority.
Disrupting the Enterprise Status Quo
For years, the high-end Enterprise SSD market was an exclusive club controlled by the giant memory manufacturers. In 2023, Phison disrupted this hierarchy by launching its own Enterprise-grade PCIe Gen4 SSDs.
The move was a survival necessity. As traditional NAND prices drop and the industry matures, Phison realized that the future isn’t in just “warehousing” data, but in accelerating it for the AI age.
The “Flash-as-Memory” Revolution
The biggest hurdle for Generative AI today is the cost of “Large Language Models” (LLMs). These models require massive amounts of DRAM to be made into HBMs, which is in short supply and prohibitively expensive. Phison’s breakthrough technology, aiDAPTIV+, allows systems to use NAND Flash as a functional extension of high-speed memory.
Eliminating AI Latency: Standard AI setups often experience “lag” because they must fetch data from the cloud. Phison’s tech allows AI to “cache” massive datasets directly, enabling instant, real-time responses.
Democratizing AI Power: While a top-tier NVIDIA-based AI rack can cost upwards of $150 million, Phison is enabling mid-sized enterprises to run powerful models on standard servers and high-end workstations.
“If you have $150 million, go get NVIDIA gears. If you don’t but also want high-performance, cost-effective computing power, you come to Phison,” said Lin.
Hybrid AI and Data Sovereignty
A standout example of this tech in action is the “AI Edge Box for OpenClaw.” These localized units allow businesses to run a “Hybrid” AI model:
The Cloud handles massive, general queries.
The Local Box handles day-to-day tasks like internal translations or proprietary data processing.
This approach slashes cloud fees by up to 70%. More importantly, it addresses Data Sovereignty. For hospitals, banks, and government agencies that cannot risk uploading sensitive information to a public cloud, Phison’s “on-premise” AI allows them to keep their data local, secure, and under their own control.
Conclusion
As industry leaders like NVIDIA begin integrating NAND Flash directly into their next-generation server architectures, Phison’s long-term bet is paying off. Lin said Phison is also collaborating with Qualcomm, AMD, Intel, and all other PC ODMs. By transforming the humble SSD into a vital component of the AI compute engine, Phison is proving that the next leap in artificial intelligence won’t just happen in the cloud—it will happen in the edge devices, such as desktops, laptops, etc., that we use every day.
The Chinese version:
Editor’s note: This article is a collaboration between the author and Gemini.