Secure, Air‑Gapped AI Moves

Google and NetApp are moving the same way: putting artificial intelligence inside systems that can run without an internet connection. (businesstoday.in, blocksandfiles.com) On April 17, Business Today reported that Google is discussing a deal with the U.S. Department of Defense to let the Pentagon use Gemini in classified environments if legal and regulatory terms are satisfied. Reuters, cited in follow-up coverage on April 16 and April 18, said the talks include limits Google wants on uses such as domestic mass surveillance and fully autonomous lethal weapons. (businesstoday.in, money.usnews.com, winbuzzer.com) A day earlier, NetApp said it signed a four-year enterprise agreement with Google Cloud to speed deployment of NetApp storage on Google Distributed Cloud. NetApp said the joint setup is aimed at sovereign, air-gapped, and disconnected sites, and Google said those sites increasingly want cloud services and AI without external network connectivity. (netapp.com, cloud.google.com) An air gap is a physical or logical break from the public internet, used by militaries, intelligence agencies, and critical infrastructure operators that cannot leave sensitive systems online. Google says its air-gapped Distributed Cloud is built for isolated environments and lists support for standards including NIST, NATO D48, ISO 27001 and ISO 27017, and SOC 2 technical requirements. (cloud.google.com) That is the immediate problem both announcements are trying to solve: large AI models usually depend on centralized cloud computing, but classified and sovereignty-bound workloads often have to stay on local hardware under tighter control. NetApp’s role is the storage layer for those environments, while Google’s reported Pentagon talks point to the model layer moving into the same kind of sealed systems. (blocksandfiles.com, cloud.google.com, businesstoday.in) Google has been building toward this for several years. Data Center Dynamics reported that Google first announced Distributed Cloud in 2021 and launched an air-gapped version in 2024, extending its cloud stack into sites that operate autonomously from the internet. (datacenterdynamics.com) The Pentagon talks also reopen a question Google has faced before about military AI work. In 2018, employee protests pushed Google to step away from Project Maven, a Defense Department program that used machine learning to analyze drone imagery; the company later published AI principles that limited some uses. (reuters.com, money.usnews.com) The current negotiations suggest Google is trying to write those limits into contract language rather than avoid the work entirely. If a deal is reached, Gemini would not just be a public-cloud service; it would become software the government could run inside classified systems built to stay cut off. (money.usnews.com, businesstoday.in, cloud.google.com)