The Access Paradox
When NVIDIA, Google DeepMind, and Disney Research released Newton in September 2024, the open-source physics engine promised to democratise robotics simulation. Built on NVIDIA Warp and OpenUSD, it cuts simulation time from days to minutes through GPU acceleration. Genesis, released three months later, delivers 43 million frames per second on a single RTX 4090 GPU - 430,000 times faster than real-time.
The software is free. The barrier is elsewhere.
Data from the Zindi network shows only 5% of Africa's 11,000 data scientists have access to computational power for AI research. When they do get GPU access, it's typically after US researchers finish their workday. The tools are open-source. The infrastructure isn't.
The Citation Gap
DeepMind's AlphaFold demonstrates the pattern. The protein-folding AI, open-sourced in 2020, now covers 98% of the human proteome - up from 17% before its release. It's been cited over 4,000 times.
But a 2022 Nature analysis of 20 million papers across 35 years found leading countries receive disproportionately more citations for comparable research. Developed and developing nations study similar phenomena. The citation counts diverge based on institutional affiliation and geography. That gap directly affects funding, collaboration, and career progression.
Infrastructure Reality
At a UN meeting in October 2023, delegates warned the digital divide is widening. Only 36% of populations in least developed countries use the internet, versus 66% globally. UNESCO's Science Report found four in five countries spend under 1% of GDP on R&D.
Project Chrono, an open-source physics simulator from University of Parma and University of Wisconsin-Madison, is used at dozens of universities and federal labs. Hugging Face hosts thousands of pre-trained models, including IBM and NASA's solar observation foundation model. The LA-CoNGA Physics project has built computational capacity across Latin American universities since 2020, with Mexico's installed servers up 39.6% in 2024.
The software ecosystem exists. The surrounding infrastructure determines whether researchers can actually use it.
What This Means
Open-source physics AI solves the licensing problem. It doesn't solve the connectivity problem, the power supply problem, or the hardware access problem. For APAC organisations evaluating simulation tools, the calculation isn't just software capabilities - it's whether your teams and partners can actually run them.
The code is democratic. The compute remains geographically concentrated.