SAN DIEGO, May 12, 2026 (GLOBE NEWSWIRE) — Kneron, the San Diego based edge AI company developing full stack inference infrastructure, says the artificial intelligence industry may be vastly underestimating the next major bottleneck of AI and it has nothing to do with training larger models.
As global tech giants continue pouring billions into GPUs, hyperscale data centers, and increasingly massive foundation models, Kneron Founder and CEO, Dr. Albert Liu believes the industry is approaching a far more difficult challenge: how to operate AI continuously in the real world at scale.
“The market has been obsessed with training intelligence,” Liu said. “But training is only one chapter of the AI story. The real challenge begins when intelligence must run continuously across billions of devices, factories, hospitals, vehicles, and enterprise systems in real time.”
Unlike training workloads, which happen periodically inside centralized compute clusters, inference workloads operate continuously. Every AI assistant response, autonomous decision, robotics deployment, smart city system, and real time video analysis depends on inference running persistently in the background.
According to Liu, that shift is beginning to expose an entirely different set of infrastructure pressures around power consumption, cooling, deployment cost, latency, and long term sustainability.
The warning comes as concerns around AI infrastructure power demands intensify globally. The International Energy Agency recently projected that electricity demand from data centers could nearly double by 2030 as AI adoption accelerates worldwide, while a recent McKinsey analysis described the AI infrastructure buildout as a potential multi trillion dollar race increasingly constrained by energy availability, cooling systems, and deployment logistics.
“Inference changes everything,” Liu said. “Once AI becomes an always-on operational layer, the conversation shifts from raw compute performance to efficiency, latency, cooling, deployment cost, privacy, and sustainability.”
Researchers and enterprise leaders have increasingly pointed toward edge AI and distributed inference systems as critical components of the next generation AI stack, particularly as organizations seek lower latency, lower operating costs, and greater control over sensitive data.
Founded in 2015, Kneron has spent nearly a decade building what it describes as a full stack inference ecosystem focused on deployable AI infrastructure. Its platform includes reconfigurable neural processing units, edge AI servers, orchestration software, runtime systems, and localized AI platforms designed to operate without continuous cloud reliance.
Unlike traditional cloud centric AI systems, Kneron’s architecture focuses on low power, privacy preserving AI processing directly on devices and localized infrastructure. “The future of AI cannot exist entirely inside hyperscale data centers,” Liu said. “AI must become deployable, persistent, private, and capable of operating directly where data is created.”
Kneron describes this shift as the emergence of an “inference infrastructure era,” where success may depend less on building larger models and more on creating systems capable of running intelligence efficiently and continuously across real world environments. “The first era of AI was defined by training,” Liu said. “The next era will be defined by how intelligence operates everywhere.”
Kneron is expected to showcase its latest edge AI infrastructure and inference focused technologies at COMPUTEX Taipei from June 2 through June 5, where AI infrastructure, robotics, edge computing, and next generation deployment systems are expected to dominate industry discussions this year.
