# The SAVRN Standard: Seven Principles for Controlled AI Source: https://savrn.com/ai-index/standard Markdown alternate of the page above; the site index is https://savrn.com/llms.txt --- [SAVRN Index.](https://savrn.com/ai-index) Explore the IndexSAVRN Standard Assessments ## 7 offerings, measured against the Standard None of the 6 documented offerings is rated Supported on any principle. Power provision is not established for any of them. Workforce reskilling has not been assessed for them yet. SAVRN wrote the Standard and builds to every principle. - Supported - Partial - Not established - Not yet assessed SThe SAVRN Standard seal goes to an offering rated Supported on every principle. Awarded to SAVRN controlled AI environment. Offering 01Workforce 02Operations 03Data 04Models 05Supply 06Infrastructure 07Power Supported SAVRN controlled AI environment SAVRN · Built to the Standard 7of 7 Azure Local disconnected operations Microsoft · Documented offering 4of 74 partial Google Distributed Cloud air-gapped Google · Documented offering 4of 74 partial HPE Private Cloud AI air-gapped edition HPE · Documented offering 5of 75 partial AWS AI Factories AWS · Documented offering 5of 75 partial OCI Dedicated Region Oracle · Documented offering 4of 74 partial Dell AI Factory with NVIDIA / Palantir reference architecture Dell · Documented offering 5of 75 partial Supported, by principle 1 of 7others not yet assessed 6 of 75 partial 7 of 76 partial 5 of 74 partial 7 of 76 partial 1 of 7none partial Read the source ↗ The principles in full ## What each principle requires, and why it's there 1. 01 ### Workforce reskilling 1. The Standard requiresA training institute reskills the American workforce to operate, maintain and use the environment. That is where the value is delivered. 1. Documented withName the programs, hands-on training hours, credentials and the roles they prepare people for. 1. Why it mattersSAVRN treats workforce reskilling as the first principle, not the last, because every other principle depends on people who can operate it. A campus that generates its own power, controls its own supply chain and keeps data on premises still needs technicians, operators and engineers who can run it, day to day, for decades. The SAVRN Training Institute sits on every campus for that reason: it teaches the American workforce to operate, maintain and use the environment, and it feeds SAVRN's own hiring pipeline directly from the community around each site. Building the workforce first is what makes every other principle sustainable, rather than a one-time capital project. 2. 02 ### Independent operations 2. The Standard requiresThe environment operates with no outside connection, running its own identity, management, inference, monitoring and recovery. 2. Documented withProven with a proposed common 72-hour disconnected test. Specify workload, thresholds, updates and support dependencies. 2. Why it mattersAn environment that depends on an outside connection to keep running is not actually independent, no matter how well it is otherwise built. SAVRN's environments run their own identity, management, inference, monitoring and recovery, with no outside connection required to operate day to day. That independence is proven, not assumed: SAVRN proposes a common 72-hour disconnected test as the standard for verifying it, with the workload, thresholds, updates and support dependencies specified in advance, so the test means the same thing for every offering measured against it. 3. 03 ### Data custody 3. The Standard requiresOn-premises data custody: customer data, storage, model artifacts, logs and backups remain inside private, controlled, sovereign boundaries. 3. Documented withDefine key authority, retention, deletion and transfer controls. 3. Why it mattersOn-premises data custody is both a control principle and a cost principle. Keeping customer data, storage, model artifacts, logs and backups inside private, sovereign boundaries means the customer, not a third party, holds the keys, the retention policy and the deletion rights. It also means the customer is not paying an outside provider's storage and egress economics indefinitely. Managing data on premises lets SAVRN's customers extend and maintain their own data environment on their own terms, with the savings that come from owning the infrastructure instead of renting access to it. 4. 04 ### Customer model capability 4. The Standard requiresThe value is in the customer's data. That data trains a local model built for the customer's specific use, not a frontier model that is not trained for it. 4. Documented withDefine data rights, evaluation, promotion and rollback. Identify retrieval-only services separately. 4. Why it mattersThe value in a customer's AI environment is in the customer's own data, not the frontier model. SAVRN's approach trains a model on that data, inside the customer's own environment, until it performs better for that customer's specific use than a stock frontier model can, because the frontier model was never trained on it. Frontier model access from a lab is a starting point, not the end state; a model shaped by a customer's own operations, history and edge cases is worth more to that customer than general access to someone else's model. 5. 05 ### Supply assurance 5. The Standard requiresThe global supply chain behind the infrastructure is controlled, with traceable provenance, controlled procurement, maintenance rights and a documented replacement strategy for critical hardware, software and models. 5. Documented withDisclose dependencies. 5. Why it mattersA controlled AI environment is only as reliable as the supply chain behind it. SAVRN keeps every component of that chain, hardware, software and the models themselves, under its own control, with traceable provenance and a documented replacement strategy for anything critical. The alternative is what most of the market lives with today: dependencies split across vendors with no visibility into where a part comes from or how long a replacement will take. SAVRN's approach is built specifically to avoid the long delays that come from an uncontrolled or opaque supply chain. 6. 06 ### Infrastructure control 6. The Standard requiresDedicated compute, networking and facility infrastructure have documented customer rights and a named accountable operator. 6. Documented withOwnership or enforceable control rights may qualify. 6. Why it mattersDedicated compute, networking and facility infrastructure only count toward the Standard when a named operator holds documented rights over all three, not just the servers. SAVRN's partnership with IntelliFlex is what makes that control real: IntelliFlex brings the operating discipline and facility management that lets SAVRN name one accountable operator for the full stack, from the network fabric to the building envelope, instead of splitting responsibility across a landlord, a colocation provider and a separate managed-services vendor. That single line of accountability is what a customer is actually buying when they buy infrastructure control. 7. 07 ### Power provision 7. The Standard requiresDedicated on-site generation and defined resilience are included in the accountable service scope. 7. Documented withState ownership or contractual rights, fuel dependencies, supported critical load and commissioning evidence. 7. Why it mattersSAVRN campuses are not connected to the surrounding grid. Power comes from on-site fuel cells, generated and controlled on the same campus that uses it, so there is no utility interconnection queue and no dependency on shared grid capacity. On-site generation is also cleaner than drawing from a regional grid that still runs on a mixed fuel stack, though SAVRN is careful not to overstate that difference. What can be said plainly is that dedicated on-site generation gives the customer defined resilience and a supported critical load inside the accountable service scope, backed by commissioning evidence, rather than a shared claim on someone else's grid. The SAVRN reference set ## Seven references, one system Capital, delay, policy, water, grid, permits and scarcity are one system seen from seven sides. Each reference is primary-sourced, refreshed on a schedule, and links into the others, so a project can be followed from the money committed to the docket that stopped it. Capital Atlas Who is funding AI data centers: capital providers, named sites, financing transactions, and the blocked-and-delayed record, state by state and company by company. [Open the atlas →](https://savrn.com/data-center-capital-atlas) Delay Watchlist Project-level slippage: named projects, jurisdictional actions, utility and federal dockets, SEC disclosures, and the interconnection queue. Open the watchlist → Moratorium Tracker The policy layer: every state measure and local moratorium mapped and cited, with one page per state. Open the tracker → Water Tracker Every federal, state, local and interstate water measure that touches a data center, plus a fifty-state water ledger. Open the water tracker → Grid Operator Watchlist ERCOT, PJM, MISO, CAISO, SPP, NYISO, ISO-NE and FERC: the large-load rules, filings and queue positions that decide whether a site can energize. Open the watchlist → Permits and Power Development Data center and substation permits, large loads in the interconnection queues, and the power plants planned to serve them, refreshed hourly from public records. Open the permits tracker → SAVRN Scarcity Tracker Eleven primary-sourced indicators of AI infrastructure scarcity, one composite index, read against the token index, SAVRN's own cost basis and the frontier ceiling. Open the scarcity tracker → Have a specific address? Site Intelligence reads all seven references against one location, county, utility and grid operator included, and cites every finding. [Check an address →](https://savrn.com/site-intelligence)