By Phillip Merrick, Co-founder and Chief Executive Officer, pgEdge
Data sovereignty is set to become the most significant issue in corporate AI and cloud computing in 2027, owing to multiple converging factors. Concerns about token costs, security, maintaining control over corporate data, and the possibility of such data being used in training of AI models — unbeknownst to the data owner — are all driving enterprises to host and run open-weight models on-premises or in their own cloud accounts.
Token costs remain an understandable concern for enterprises everywhere, especially as agentic applications consume tokens at a rate chatbots never did. Beyond that, the internal data needed for supplying agents with information is often co-located with the models. Many enterprises have sensitive data that cannot be uploaded into the cloud, except in some cases where they own and manage the cloud accounts themselves.
Organizations moving to self-hosted open-weight models with co-located data infrastructure clearly see benefits in terms of reduced AI bills and lower latency with their agentic AI applications. These smaller open models give organizations the power to fine-tune exactly what information is provided, with no token costs and no information shared with third-party technology companies … and the result is a model that is hyperspecific for purposes like prediction, auditing for anomalies, knowledgebases, chat bots, or similar highly targeted uses.
The transition is happening quickly, and it appears as though decisions based on data sovereignty are the primary driver behind the change. In Futurum’s September survey, 42% of enterprises were running open-weight models in production, up from 31% just two months earlier. It was most common to run these models in self-managed cloud or on-premises infrastructure. Notably, for 47% of those production users, a model’s country of origin is now a major or deciding factor, up from 35%.
None of this means the frontier models are going away. Models themselves aren’t becoming commodities, and for some workloads, a frontier model will remain the right call. What’s changed is that enterprises now have far more choice than they used to, and choice is what makes sovereignty possible.
However, rather than moving workloads wholesale back to on-premises infrastructure, enterprises need the flexibility to locate them where it makes most sense — based on the overall assessment of risk and data sensitivity against costs. This calls for data infrastructure platforms that offer the flexibility for deployment across the spectrum of vendor cloud, enterprise-owned cloud, on-premises and even air-gapped. The need for data infrastructure platforms with these kinds of capabilities is exactly why we built pgEdge Starfleet, our managed Postgres database-as-a-service platform that offers full deployment flexibility. Teams can start in our cloud, then deploy to their own cloud account, on-premises, or fully air-gapped, without changing the database underneath.
Take steps to internally publish a list of approved AI coding and database platforms that let your teams move fast while still meeting your security, compliance and sovereignty requirements. Without one, people will choose their own tools, and you’ll have both a shadow AI and a shadow IT problem. Moreover, the concept of using open source tooling to escape lock-in applies directly here as well: Designing your data infrastructure and AI architecture around proprietary models means one vendor decides where your most valuable asset lives and what it costs you to use it.
The enterprises that come out ahead in 2027 won’t be the ones with the biggest model budgets; they’ll be the ones that embrace flexible data infrastructure platforms in combination with open-weight models to address data sovereignty concerns. These companies will be best positioned to realize returns on their agentic AI investments while meeting requirements for security, compliance and reduced AI token costs.
##
ABOUT THE AUTHOR
Phillip Merrick is the co-founder and CEO of pgEdge. A developer and entrepreneur, he has more than three decades of experience in the enterprise software industry. Companies he has co-founded and/or led include webMethods, EnterpriseDB (EDB), SparkPost and Fugue, all of which achieved nine- or 10-figure exits. At webMethods, Phillip co-invented web services and web APIs and, under his leadership, completed a highly successful IPO, becoming the fastest-growing U.S. software company in that era. At EDB, he co-founded the first successful company to commercialize the open source Postgres database. Founded in 2023, pgEdge delivers open source, 100% Postgres infrastructure for agentic AI and other enterprise applications that demand high availability, reliability and/or data sovereignty. Visit www.pgedge.com and follow the company on LinkedIn.
