GCRI · Foundations
The technical architecture: where frontier technologies meet resilience evidence
How the twenty documented Nexus systems are meant to bring AI, compute, Earth observation, sensors, networks and simulation together on one evidence path, and why each frontier capability enters with a record and a limit.
- As of
- 21 October 2026
- Reading
- 10 minutes
- Evidence
- 15 sources, cited in order
Methods and evidence
- Evidence
- 15 sources, cited in order and listed under References
- Figures
- 5 figures
- Data availability
- Every chart's data is published as a table on this page.
01What the technical architecture is
What the technical architecture is
The technical architecture of the Nexus programme is the arrangement of twenty systems, each named by its Nexus Standards display code and registered descriptor, through which data, models, compute, observation and decision support are meant to connect 1. It is a convergence architecture: its purpose is not to replace the technologies that governments, operators, universities and providers already run, but to give them a common evidence path, common records and common limits, so that what one produces another can check.
The architecture is stewarded within the Nexus Consortium with The Global Risks Forum (GRF), which convenes the annual programme where capabilities are tested, and The Global Risks Alliance (GRA), which reads the resulting evidence for finance and insurance 1. Technology, infrastructure and service providers work with the Consortium at arm's length under written agreements and are described by role only; a provider does not operate the programme or direct its decisions 1.
The earlier version of this page centred the architecture on a branded annual compute core and a passport for every component. Neither is on the current record. The architecture documented today is the one on the systems pages, and this page follows it.
02Why convergence is the problem to solve
Why convergence is the problem to solve
Systemic risk is a convergence problem. In coupled networks, the failure of nodes in one network can bring down dependent nodes in another, so that the behaviour of the whole cannot be read from any part (Buldyrev et al., 2010) 5. The economic consequence is measured: the Global Assessment Report 2025 estimates that disaster costs of about $202 billion in direct losses exceed $2.3 trillion once cascading and ecosystem costs are counted 6.
A climate model needs infrastructure data; a cyber exercise needs assumptions about financial continuity; a digital twin needs data governance; an AI system needs a data boundary and human oversight. No single provider, dataset, dashboard or model can carry all of that. What is missing is rarely another platform. It is a trusted way for existing platforms to converge on one question with their assumptions visible.
Hazards themselves are changing faster than systems built for one of them. The World Meteorological Organization reports that 119 countries now have multi-hazard early warning systems, while singling out extreme heat, wildfires and glacial lake outburst floods as emerging hazards that many systems were not designed for 7. An architecture that can add a hazard without rebuilding its foundations is the practical answer.
03One evidence path through the systems
One evidence path through the systems
The architecture is easiest to read as a path. A signal is captured by remote sensing, an in-situ instrument or an edge device, in NXS-OPS (instruments, sensors and calibration), NXS-EOP (earth observation and digital twins) or NXS-SIG (signal, spectrum, timing and PNT). It is structured as an observation under Nexus Observatory Protocol, owned by NXS-OPS, and held as governed evidence with quality and provenance in NXS-GRX (canonical data, provenance and risk) 1.
After review, NXS-EWS (multi-hazard monitoring and warning) may prepare a warning-support candidate and NXS-AAP (anticipatory action and triggers) a trigger check; NXS-DSS (decision cases and deliberation) assembles a decision-support package. The NXS-NSF (trust, authorization and assurance) policy gate releases it, and NXS-RLS (participants, agreements and handoffs) hands it off under written agreements. The competent authority alone decides and issues any warning 1.
Two rules hold the path together. Every value carries one of nine evidence labels, and simulated results are never reported as observed impact 2. Data is held so that it is findable, accessible, interoperable and reusable for machines and people, the test set by the FAIR principles 8.
Fig. 01From a signal to a decision the programme does not take
Figure 1 in words
Eight boxes joined by arrows. Row one: signal captured by remote sensing, in-situ instruments or edge devices (NXS-OPS, NXS-EOP, NXS-SIG); observation structured under Nexus Observatory Protocol; governed evidence with quality, provenance and an evidence label (NXS-GRX); evidence review before any use. Row two: warning-support candidate or trigger check, never an alert (NXS-EWS, NXS-AAP); decision package with options, uncertainty and limits (NXS-DSS); policy gate and handoff under a written agreement (NXS-NSF, NXS-RLS); the authority decides outside the programme and alone issues any warning. A note: corrections return to the start as new recorded versions.
04Frontier computing has a physical footprint
Frontier computing has a physical footprint
Frontier capability is not weightless. The U.S. Department of Energy reports, from a Lawrence Berkeley National Laboratory study, that data centre electricity use in the United States rose from 58 TWh in 2014 to 176 TWh in 2023, about 4.4% of national electricity, and could reach 325 to 580 TWh by 2028, between 6.7% and 12% of the total 3. The range is wide because it spans scenarios of AI adoption and efficiency; it is a range, not a forecast of one value.
For a resilience architecture, the consequence is direct. Compute borrowed for a year's missions borrows power, cooling and grid capacity with it, and those are themselves exposed to heat, drought and grid failure. NXS-COR (compute, nodes and runtimes) therefore records placement, runtime and measured use for each workload, and the Network and Core page records offered capacity through separate states, none of which implies the next 9.
Fig. 02Frontier computing has a physical footprint
Figure 2 in words, and its data
Three bars on one TWh scale: 2014, 58 TWh; 2023, 176 TWh; 2028, a dashed range bar from 325 to 580 TWh. A side panel: data centres consumed about 4.4 per cent of total US electricity in 2023 and are expected to consume about 6.7 to 12 per cent by 2028; a testbed that borrows compute also borrows power, cooling and grid capacity.
05Where frontier technologies meet the architecture
Where frontier technologies meet the architecture
Each frontier capability has a home system and a limit. AI models and agents are held in NXS-NXQ (software, models and capability lifecycle) with their version, role, data boundary and tool permissions; an AI output is never evidence on its own. Earth observation and twins sit in NXS-EOP, drawing on public programmes such as Copernicus, whose data is largely available worldwide on a free, full and open basis 10. Simulations and frontier evaluation run in NXS-LAB (research, experiments and simulations), where a simulated capability is not a validated result 1.
The external frameworks point the same way. The NIST AI Risk Management Framework, released on 26 January 2023 and extended by a Generative AI Profile on 26 July 2024, gives organizations a voluntary, consensus-built method for managing AI risk 11. The OECD AI Principles, updated in May 2024, set the intergovernmental baseline 12. The EU AI Act requires high-risk systems to carry risk mitigation, data quality, logging for traceability, documentation, human oversight and robustness, with rules for uses such as critical infrastructure applying from 2 December 2027 4. Builds of the architecture are signed and tested before a gate admits them, with secure development practice of the kind set out in NIST SP 800-218 as the reference 13.
Fig. 03Where frontier technologies meet the architecture
Figure 3 in words
Six rows. AI models and agents, NXS-NXQ: version, role, data boundary and tool permissions; an AI output is never evidence on its own. High-performance and edge compute, NXS-COR: placement, runtime and measured use; requested capacity is not allocated capacity. Earth observation and digital twins, NXS-EOP: imagery, terrain and twins with source and licence; a twin is a model, not the place. Sensors and instruments, NXS-OPS: calibration and primary observation evidence. Networks and timing, NXS-NET and NXS-SIG: topology, continuity, authorized signal and PNT integrity; restricted topology never enters a public view. Simulation and frontier evaluation, NXS-LAB: method, assumptions, run conditions and failure cases; a simulated capability is not a validated result. Every row passes the NXS-NSF policy gate.
06One protocol for every test
One protocol for every test
Technologies enter the architecture through testbeds, not announcements. The programme's Responsible Innovation and Interoperability Testbeds, offering O04, ask whether a candidate technology solves the intended problem under representative operating conditions 1. Every test on the Labs gateway follows the same protocol: a use case and comparator, data and metrics, a reproducible build, a test envelope and findings 14. Interoperability, failure behaviour, rights, security, resource cost and human use are all tested, and the result is a measured evaluation with failure cases and a transition recommendation 1.
The protocol keeps its own boundary. Sponsorship never influences a result, evaluation never depends on a high contribution, and testbed participation is not a regulatory sandbox, a certification or a procurement status 1. The Nexus Standards for digital systems and AI are proposed, not adopted, and each jurisdiction decides whether to use them; the develop a standard route explains how a tested method can become a change proposal.
Fig. 04Every candidate meets the same protocol
Figure 4 in words
Six boxes joined by arrows: 1 use case, the problem and a comparator to beat; 2 data and metrics, what is measured on which data; 3 reproducible build, a versioned candidate anyone can rebuild; 4 test envelope, conditions, limits and stop rules; 5 findings, a measured evaluation with failure cases; 6 transition, a recommendation, never a certificate. A note: interoperability, failure behaviour, rights, security, resource cost and human use are tested; sponsorship never influences a result; testbed participation is not a regulatory sandbox, certification or procurement status.
07Assembled, tested and released each year
Assembled, tested and released each year
The architecture is not run as a permanent machine. It is assembled for each programme year in the Foundry pillar, tested in Nexus Universe and released, with each move recorded at a gate. Planning opens the cycle projects at G0; readiness preparation qualifies sources and resource offers at G2; a one-month controlled build integrates software and data, signs builds and tests security and interoperability at G3; Action Week runs simulations and benchmarks at G4; closeout seals evidence at G5; reporting supplies permitted data bundles at G6; and renewal carries open work into the next cycle at G7 1,15.
Stand-down matters as much as build. Temporary access is revoked and temporary capacity released after Action Week, and measured capacity results are reviewed in the fourth quarter 1. The programme year page gives every phase, and GRF publishes the annual cycle of Nexus Universe on its site and on universe.nexus.
Fig. 05The architecture is assembled, tested and released each year
Figure 5 in words
A timeline grouped by quarter. Q1: planning, gate G0, cycle projects open and portfolio baseline. Q2: readiness preparation, gate G2, sources and resource offers qualified; controlled build, gate G3, one month to integrate, sign and test. Q3: Action Week, gate G4, simulations and benchmarks run; stand-down, temporary access revoked; closeout, gate G5, evidence sealed and services verified. Q4: reporting, gate G6, permitted data bundles supplied; renewal, gate G7, open work carried to the next cycle.
08What is changing
What is changing
- Compute meets the grid. Data centre demand of 325 to 580 TWh projected for the United States by 2028 makes energy a design constraint of any compute-heavy architecture 3.
- AI records become obligations. From 2 December 2027, EU high-risk AI in critical infrastructure must keep logs, documentation and human oversight 4.
- Generative AI gets its own profile. NIST's Generative AI Profile of July 2024 extends AI risk management to models whose outputs are open-ended 11.
- New hazards, same path. Extreme heat, wildfire and glacial lake outburst floods test whether warning chains can add a hazard without rebuilding 7.
- Open Earth data as a baseline. Free, full and open satellite data lets a modest testbed work with the same imagery as a large agency 10.
09Three institutions around one architecture
Three institutions around one architecture
GCRI brings the science and technical platforms the Foundry builds on and runs the experiments of the Labs pillar 1. GRF admits missions and accepts the build for exercise at a recorded gate, so that work enters the programme on the record, through Nexus Universe 15. GRA brings finance-readiness questions so that builds produce evidence financial institutions can read, through finance readiness 1.
All three follow the Nexus cycle, in which Decide belongs to the competent authority, and teach the systems through Risk Academy, where each system's rooms are designed to become learning-by-doing courses.
What this page does not claim: the technical architecture is a documented design; no system is recorded as deployed, in production use or offered as a service, and the programme is at service level R0. Naming a technology, framework, data programme or law is not a claim of partnership, endorsement or conformity. GCRI does not regulate or certify, GRF does not regulate, finance or procure, and GRA does not lend, underwrite, rate or invest. Testing is never certification, and any decision informed by programme evidence is taken by the competent authority, outside the programme, as the lawful handoffs page explains.
Questions
Questions and answers
Does GCRI build or own all twenty systems?
No. The systems are documented as a target architecture, provided under written agreements, with technology providers at arm's length and described by role only. No Nexus Standard depends on them, and no system is recorded as deployed.
How does a new technology get into the architecture?
Through a testbed or mission call. The candidate is tested against a use case and comparator, with data, metrics, a reproducible build and a test envelope, and receives a measured evaluation with failure cases, never a certificate.
Can an AI system produce evidence in the programme?
An AI output is never evidence on its own. Its model role, version, data boundary and tool permissions are recorded, and its results carry an evidence label such as modelled or estimated, with human review before any use.
Why does the architecture care about electricity?
Because compute borrowed for missions also borrows power, cooling and grid capacity, which are themselves exposed to heat, drought and outages. Recording placement and measured use keeps that dependency visible.
Is the architecture running today?
No. The programme is in formation at service level R0, information and enquiries. Calls, allocations and registrations open only when their service team, terms and review capacity are in place.
Terms
Glossary
- Technical architecture
- The arrangement of twenty documented systems through which data, models, compute, observation and decision support connect on one evidence path.
- Convergence architecture
- A design whose purpose is to let existing and emerging technologies meet on common records and limits rather than to replace them.
- Testbed
- A bounded test of a candidate technology under representative conditions, with a comparator, metrics, a reproducible build and failure cases.
- Nexus Observatory Protocol
- The observation protocol owned by NXS-OPS, with NXS-EOP and NXS-SIG, that structures a captured signal as an observation.
- Policy gate
- The NXS-NSF decision that releases a record or package to a recipient under the applicable agreement.
- Stand-down
- The phase after Action Week when temporary access is revoked and temporary capacity released.
Reuse and cite
Text CC BY 4.0. Cite the page with its date; data tables are linked under each figure.
Sources
References
15 sources, numbered in the order they are first cited
- The Global Centre for Risk and Innovation (2026-10). How Nexus Consortium works. therisk.global. Accessed 2026-10-21.
- The Global Centre for Risk and Innovation (2026-10). Programme status. therisk.global. Accessed 2026-10-21.
- U.S. Department of Energy and Lawrence Berkeley National Laboratory (2024-12). DOE releases new report evaluating increase in electricity demand from data centers. www.energy.gov. Accessed 2026-10-21.
- European Commission (2026-10). AI Act: regulatory framework for artificial intelligence. digital-strategy.ec.europa.eu. Accessed 2026-10-21.
- Buldyrev et al., Nature 464, 1025 (2010-04-15). Catastrophic cascade of failures in interdependent networks. doi.org. Accessed 2026-10-21.
- United Nations Office for Disaster Risk Reduction (UNDRR) (2025-05-27). Global Assessment Report on Disaster Risk Reduction 2025 (launch release). reliefweb.int. Accessed 2026-10-21.
- World Meteorological Organization (WMO) with UNDRR (2025-11-12). Early warning systems reach new heights, but critical gaps jeopardize global progress. wmo.int. Accessed 2026-10-21.
- Wilkinson et al., Scientific Data 3, 160018 (2016-03-15). The FAIR Guiding Principles for scientific data management and stewardship. doi.org. Accessed 2026-10-21.
- The Global Centre for Risk and Innovation (2026-10). Network and Core. therisk.global. Accessed 2026-10-21.
- Copernicus, the Earth observation component of the EU Space Programme (2026). Access to data: the Copernicus free, full and open data policy. www.copernicus.eu. Accessed 2026-10-21.
- U.S. National Institute of Standards and Technology (NIST) (2024-07-26). AI Risk Management Framework. www.nist.gov. Accessed 2026-10-21.
- Organisation for Economic Co-operation and Development (OECD) (2024-05). OECD AI Principles. oecd.ai. Accessed 2026-10-21.
- U.S. National Institute of Standards and Technology (NIST) (2022-02). SP 800-218, Secure Software Development Framework (SSDF) version 1.1. csrc.nist.gov. Accessed 2026-10-21.
- The Global Centre for Risk and Innovation (2026-10). Labs gateway. therisk.global. Accessed 2026-10-21.
- The Global Risks Forum (2026-10). Nexus Universe: the annual cycle. globalriskforum.com. Accessed 2026-10-21.
Record
Editorial record
Facts and programme states are given as of 21 October 2026. This page is reviewed at least every 90 days and whenever a cited figure, instrument date or programme state changes. It replaces an earlier version of the same page, whose branded annual compute core, component passports and participation protocol for that core are not on the programme's current record; those structures have been removed and the architecture is described from the systems documentation.
Published by the GCRI editorial desk under the editorial integrity policy. To report an error or request a correction, use the corrections route; corrections are dated and listed with the page. Participation is recorded in the programme's participation records.
Cite this article
GCRI editorial desk (2026, June 8). Technical architecture for frontier technology and resilience. The Global Centre for Risk and Innovation. https://therisk.global/the-technical-architecture-a-nexus-for-frontier-technologies-systems-providers-protocols-and-programmatic-resilience-infrastructure/
@online{gcri2026thetechnicalarchitectu,
author = {{GCRI editorial desk}},
title = {{Technical architecture for frontier technology and resilience}},
publisher = {The Global Centre for Risk and Innovation},
year = {2026},
month = {6},
url = {https://therisk.global/the-technical-architecture-a-nexus-for-frontier-technologies-systems-providers-protocols-and-programmatic-resilience-infrastructure/},
language = {en}
} TY - ELEC AU - GCRI editorial desk TI - Technical architecture for frontier technology and resilience PB - The Global Centre for Risk and Innovation PY - 2026 DA - 2026/06/08 UR - https://therisk.global/the-technical-architecture-a-nexus-for-frontier-technologies-systems-providers-protocols-and-programmatic-resilience-infrastructure/ LA - en ER -
Republish
Republish or translate this article free of charge, in print or online, under CC BY 4.0: keep the byline and The Global Centre for Risk and Innovation, link to the original and note any change. Figures, maps and twins made by the institutions carry the same licence; third-party data keeps its own credit line.
Attribution to paste
<p>The technical architecture: where frontier technologies meet resilience evidence — GCRI editorial desk, The Global Centre for Risk and Innovation. First published 8 June 2026 at <a href="https://therisk.global/the-technical-architecture-a-nexus-for-frontier-technologies-systems-providers-protocols-and-programmatic-resilience-infrastructure/">https://therisk.global/the-technical-architecture-a-nexus-for-frontier-technologies-systems-providers-protocols-and-programmatic-resilience-infrastructure/</a>. Republished under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/).</p>