Hextrion: The Frontier Architecture Powering Next-Generation AI Compute
Hextrion is a U.S. company building next-generation Semiconductor Interconnect Power and Communications (SIPC) advanced packaging beneath the chip that dramatically improves the economics and performance of AI compute: less energy, less capex, more bandwidth, lower latency. Hextrion helps AI data centers achieve up to:
The Team
Built by the people who built the modern packaging stack
Our leadership team of 20 former C-suite and senior executives brings 400+ years of collective experience from Foxconn, Unimicron, ASE, Lam Research, and TSMC — the global companies that produce the current SIPC advanced packaging stack serving NVIDIA, AMD, Broadcom, Google, Microsoft, Apple, and Amazon. Collectively, the team has built businesses from scratch to over $30B in revenue.
Our AI team consists of physics AI researchers from Harvard and MIT who have developed model architectures that solve complex physics problems up to 100,000× faster than traditional physics-informed neural networks, and AI manufacturing leaders who have deployed production AI systems at TSMC and component manufacturers supplying Apple and BMW.
Advanced Packaging · IC Substrates · HDI · Semiconductor Equipment · Physics AI · Factory Automation · Fab Development
Why Hextrion
Advanced packaging has become AI's hidden bottleneck
For thirty years, progress in computing came from shrinking the transistor. That era is ending — and the next gains are coming from the layer beneath the chip: the packaging that moves data between processors and delivers power to them. It has quietly become the limiting factor in AI — today, roughly two-thirds of a data center's energy never reaches computation, lost instead to moving data, delivering power, and cooling. And it was never designed for this.
Hextrion exists to rebuild that layer. We bring a semiconductor standard to a part of the stack that has long been treated as a commodity, and we pair deep manufacturing experience with physics-AI to design and build it at a pace the old methods can't match.
