Industry experts participated in a forum for COMPUTEX 2026 where they looked at AI’s foundation, data, and how companies optimize its storage, security, and governance.
Synology Managing Director Mike Chen highlighted how his company provides data sovereignty and backup solutions for more than 350,000 customers across the private and public sectors. As AI use expands, users feed models critical information that organizations and government may not want to lose control over.
“Don’t fear the tool. Control the tool,” Chen said. Synology places its technology at the data’s source, from space stations to race cars, to lower latency and costs, while offering users its full control. Synology also centralizes data to define and identify issues, from ransomware attacks to restoring backups.
Ahmed Shihab, CPO at WD, said AI models generate training data, inference data, and data that influences users. Using an example of generating an eight-second, four-megabyte video, he estimated a model may create more than 40 gigabytes in inference data. He put forward WD’s technology as effective for how to meet these growing demands, keeping more critical data on SSDs while using bulk storage HDDs for most of it.
“The more bulk storage, the more embeddings, the more embeddings, the better the AI, the better the AI, the more data and the cycle continues,” Shihab said.
“How much do I save? Where do I save it? Do I save all of it? Do I save only a little bit of it? All those decisions translate to data growth,” Solidigm Vice President of AI Ecosystem and Market Enablement Avi Shetty said. “Storage is the foundation of AI inference,” he added.
Before large language models and AI answer a user’s question, they process thousands of prior interactions and background information to understand instructions, he said. That translates to incredible amounts of storage required, which is only increasing, while memory suppliers are already struggling to meet global demand.
Shetty presented Solidigm’s solution, highly-efficient data centers with different tiers made up of different storage types, lowering memory footprint per token to ensure maximum return on investment for its partners.
Kioxia’s session, led by Koichi Fukuda, went more in-depth on the specific role played by SSDs in AI infrastructure. “Ai performance depends on data, and data is enabled by storage,” Fukuda said, “that is why SSDs are not just components, but the key enabler of modern AI.”
As the data center market continues to grow, SSD manufacturers like Kioxia play a key role in delivering high-speed, high-bandwidth memory. Fukada compared the benefits of different types of memory, SSDs, DRAM, and HBM and said that for many, SSDs deliver the best balance of performance, capacity, scalability, and cost.
“The next phase of AI will not be defined by compute alone. It will be defined by how effectively we connect compute to data,” Fukada said.
Fukuda stressed that given the massive quantity of data now available, leading companies will need to be able to leverage it, rather than focusing only on generating tokens. Kioxia is building software alongside its hardware to meet that need. Fukuda specifically mentioned AiSAQ, an open-source software that optimizes memory usage by storing data in flash storage rather than volatile memory.
