Investment in Research Infrastructure — Written Evidence
House of Commons Public Accounts Committee
Date: 12 May 2026
Submitted by: Vsevolod Shabad, Independent Researcher; Principal Enterprise Architect at a FTSE 100 company. This submission is made in a personal capacity and does not represent the views of any employer, institution, or organisation.
The National Audit Office's March 2026 report on DSIT's investment in research infrastructure concludes that "the lack of a consistent approach makes it harder for DSIT to understand the entire research infrastructure landscape and report publicly on the impact of research infrastructure" (National Audit Office, 2026, para 15). This finding is not incidental to the inquiry — it identifies the primary governance constraint: DSIT cannot exercise effective portfolio oversight of a landscape it cannot coherently observe.
The fragmentation documented by the NAO has structural roots. UKRI's research infrastructure funding is split across multiple streams — World Class Labs (£628 million spending in 2024-25), the Infrastructure Fund (£248 million), Research England allocations, and individual research council funding — without a unified portfolio view enabling DSIT to assess their aggregate performance or strategic alignment (National Audit Office, 2026, para 5). UKRI's 2020 Landscape Analysis identified over 527 nationally and internationally significant infrastructures across six sectors, employing nearly 25,000 full-time equivalent staff, with 92 per cent of infrastructures working across multiple research domains (UK Research and Innovation, 2020a). The size and diversity of this estate make the absence of consolidated portfolio-level data a severe governance gap.
The governance information problem is not new. Both the House of Lords Science and Technology Select Committee and the NAO previously identified the need for better landscape understanding (National Audit Office, 2016), prompting the 2020 Roadmap Programme that produced the Landscape Analysis and the Opportunities to Grow report (UK Research and Innovation, 2020a, 2020b). Six years after those documents were published, the NAO's 2026 findings confirm that a consistent monitoring framework does not yet exist. The Landscape Analysis itself documented that "only 41% of respondents feel that they are able to plan beyond three years ahead, highlighting a mismatch between funding cycles for research and innovation infrastructure and planning requirements" — and that this occurs despite 60 per cent of infrastructures having operational lifespans exceeding 25 years and over three quarters facing major decisions within the next five years (UK Research and Innovation, 2020a). The mismatch between governance horizon and asset horizon has been known since 2020; the current inquiry provides the opportunity to establish the governance architecture that resolves it.
DSIT's October 2025 R&D plans to 2029/30 describe an intention to manage R&D funding "in a more agile way", including the ability to "move some funding across financial years", "reallocate funding where a programme underspends", and "deprioritise investment that is not delivering the outcomes expected" (DSIT, 2025). These are the right governance aspirations. Their realisation depends, however, on DSIT having access to the comparative portfolio-level performance data that the NAO confirms is currently absent.
The NAO's March 2026 report provides two findings on funding determination that deserve sustained attention.
First, the primary appraisal instrument is the wrong tool for this class of investment — and decision-makers already know it. The NAO concludes that "information on cost-benefit analysis was less likely to be useful for decision-makers as they do not primarily rely on benefit-cost ratios to compare different projects. This is because it is particularly challenging to quantify and monetise the benefits that an individual proposal would be expected to deliver" (National Audit Office, 2026, para 9). The correct interpretation of this finding is not that benefit-cost ratios (BCRs) are being applied poorly, but that they are structurally misaligned with the investment objective. BCRs optimise for discounted expected returns over a defined period. For frontier research infrastructure — where the relevant benefit is strategic capability, geostrategic positioning, and optionality that may not materialise for decades and cannot be monetised in advance — this is the wrong objective function. Decision-makers are right not to rely on it. The UKRI 2020 Opportunities to Grow report acknowledged the same constraint, explicitly stating that "we have not attempted to prioritise the opportunities described here" and that the report "does not represent a funding commitment or seek to prioritise the funding of particular infrastructures" (UK Research and Innovation, 2020b) — in part because BCR-based ranking was not credible for the benefits involved.
The practical gap is not a missing BCR calculation but a missing portfolio-level objective function. DSIT's October 2025 R&D plans explicitly signal the direction: DSIT is "taking a new approach to managing its R&D funding in a more agile way", including the ability to "deprioritise investment that is not delivering the outcomes expected" (DSIT, 2025). Agile portfolio management provides precisely the objective function BCRs cannot: Cost of Delay — the per-period scientific and strategic value foregone by not having a capability — ranked against delivery duration. This does not require multi-decade demand projections. It requires a defensible near-term judgement about which capability gap is most costly to leave open. Under deep uncertainty, this is both a more honest and a more robust basis for portfolio prioritisation than point-estimate optimisation against a thirty-year forecast (Reinertsen, 2009; Shabad et al., 2026).
Second, commitments are fixed before uncertainty is sufficiently resolved. The NAO found that UKRI "fixed budgets too early before costs could have been well-understood" and "did not initially put in place appropriate safeguards to ensure it challenged over-optimism from individual research councils — who were naturally keen to ensure that their priorities were funded" (National Audit Office, 2026, para 10). The over-optimism at early stages is not a failure of individual diligence. It is a structural consequence of a governance architecture in which binary commit-or-cancel decisions are made before adequate cost and risk information is available, generating institutional pressure toward optimistic forecasts as the price of inclusion in the portfolio (Flyvbjerg et al., 2003; Staw, 1976).
A further structural distortion operates through accounting classification. The NAO's recommendation on financial management explicitly notes that "all UKRI spending is classed as capital spending" and recommends that UKRI develop frameworks to understand "the relative merits of spending on new infrastructure, maintaining existing infrastructure or directly funding research" (National Audit Office, 2026, recommendation d). This is the CAPEX-bias mechanism documented in regulatory economics: where governance frameworks treat capital and operational expenditure differently, institutions systematically favour capital solutions over operationally flexible alternatives, regardless of their comparative scientific value (Averch & Johnson, 1962). For UKRI, classifying all spending as capital creates pressure toward new physical construction over maintenance optimisation and operational access arrangements — a dynamic directly visible in the maintenance backlog discussed below.
The Infrastructure Fund, established in 2022, represents genuine progress: structured assessment, UKRI-wide governance, and 30 approved projects with an expected total cost of £2.04 billion. However, the NAO observes that UKRI "is still not managing all its research infrastructure as an integrated portfolio" and that its ability to do so "is limited by the inflexibility of the funding arrangements" (National Audit Office, 2026, para 8). The NAO is right to identify this as the central remaining gap, and recommendation (b) correctly calls for UKRI to "take a broader view of portfolio management" and "introduce more flexibility into how it manages projects such as staged funding" (National Audit Office, 2026, recommendation b). The Fund addresses project-level governance; portfolio-level governance — the ability to trade off between projects dynamically — is what recommendation (b) targets and what this submission proposes to specify.
The NAO documents three delivery failures that expose systemic governance constraints.
Portfolio inflexibility is producing avoidable underspend. Between November 2023 and July 2025, UKRI "did not allow any projects to bring spending forward — despite knowing other projects were proceeding more slowly than planned because UKRI needed to restrict spending to remain within its overall budget. This resulted in an £81 million (18%) underspend" (National Audit Office, 2026, para 8). The NAO also found "no evidence of UKRI slowing or stopping research infrastructure projects to enable more strategically important projects to proceed until December 2025." A portfolio with projects advancing at different speeds could not reallocate resources between them because the funding architecture did not permit it. This is the governance failure the Annual Portfolio Model's annual envelope reset is designed to prevent: within an annually revised envelope, the portfolio manager can bring projects forward or defer them based on observed performance without requiring special authorisation (Shabad et al., 2026).
Slow response to infrastructure obsolescence. DSIT and UKRI "did not act quickly enough to replace the ageing ARCHER2 supercomputer and did not have an overarching strategy for computing until recently" (National Audit Office, 2026, para 11). ARCHER2 is expected to close 13 months before its replacement becomes operational, meaning a period of reduced supercomputing capability for UK research. By contrast, the National Quantum Computing Centre — where DSIT and UKRI "adapted their processes appropriately to reflect the importance of quantum computing and the need for speed and agility" — demonstrates that adaptive governance is possible when it receives explicit senior attention. The asymmetry is instructive: adaptive capacity currently depends on political salience rather than being a structural property of the governance architecture.
Irreversible commitment before adequate evidence. One project reviewed by the NAO "will not now deliver all the benefits anticipated in its business case". A second — an aircraft upgrade — "would not have delivered all the planned benefits even before UKRI decided it would decommission the aircraft instead (after this project had spent £46 million of its £49 million budget)" (National Audit Office, 2026, para 10). The aircraft case illustrates precisely the governance gap that a formal Standby mechanism addresses: £46 million was committed before the case for decommissioning became clear. A three-state commitment architecture — Active, Standby, Exit — with annual review triggers would have provided the institutional process to pause the programme at an intermediate evidence point before irreversible capital was consumed (Shabad et al., 2026). The NAO's recommendation on portfolio management calls for "staged funding" — the Annual Portfolio Model provides the specific institutional design that staging requires.
Accumulated maintenance backlog. The NAO finds that the Science and Technology Facilities Council "has not carried out sufficient maintenance to ensure its estate continues to meet its standards" and estimates it would need £360 million to restore its condition to an acceptable standard (National Audit Office, 2026, para. 13). More broadly, Research England estimates that £5.6 billion would be required to restore university-owned research infrastructure to a fully operational condition, with English universities currently spending £758 million annually on maintenance — of which £59 million in UKRI support is ring-fenced for existing infrastructure (National Audit Office, 2026, para 13). This backlog is the long-run consequence of the CAPEX bias identified above: classifying all UKRI spending as capital creates systematic pressure toward new construction over maintenance investment, even where the marginal research value of maintenance would exceed the marginal value of new capability.
The NAO's recommendation (d) directly addresses this: UKRI should develop frameworks "to better understand the relative merits of spending on new infrastructure, maintaining existing infrastructure or directly funding research", and should "explore whether there are any internal financial management arrangements which hinder this, noting that all UKRI spending is classed as capital spending." Achieving total expenditure indifference — assessing capital and operational solutions to research infrastructure outcomes against equivalent governance criteria — is the institutional reform that this recommendation implies. Without it, new capital will continue to be favoured over maintenance in ways that produce the £5.6 billion backlog this inquiry has documented.
Inconsistent monitoring preventing evidence-based reallocation. The NAO finds that "UKRI has a framework for monitoring and evaluation but has not issued specific guidance about how the performance of research infrastructure should be monitored" and documents "a variety of approaches to monitoring inputs, outputs, outcomes, and impacts" across the case studies examined (National Audit Office, 2026, para 15). Without a consistent cross-portfolio monitoring framework, DSIT cannot use utilisation and performance data to inform funding reallocation decisions. The Landscape Analysis identified the same problem in 2020: "Measuring usage and capacity is complex given the diversity of infrastructures and range of access models" (UK Research and Innovation, 2020a). Five years later, the NAO confirms that this complexity has not been resolved through governance design — leaving the £81 million underspend result, the ARCHER2 latency, and the aircraft decommissioning decision all as reactive responses rather than outputs of systematic portfolio monitoring.
The NAO's March 2026 report and the 2020 Roadmap Programme documents together identify a consistent set of governance failures: fragmented landscape intelligence, appraisal tools that decision-makers do not rely on, funding inflexibility that prevents portfolio-level trade-offs, over-optimism in early-stage business cases, and a maintenance backlog reflecting structural CAPEX bias. These are features of a governance architecture designed for a different uncertainty environment than the research infrastructure actually operates in. DSIT's October 2025 R&D plans signal the direction of reform; the Committee's inquiry can specify the governance mechanisms that make that direction achievable.
1. Implement NAO recommendation (b) through a specified three-state portfolio architecture. The NAO calls for staged funding and broader portfolio management. The Annual Portfolio Model provides the mechanism: a formal Active, Standby, and Exit classification for all Infrastructure Fund commitments, with defined annual review triggers and explicit transition criteria. This is not a new policy objective; it is the institutional design that makes the existing objective achievable within the current Spending Review framework.
2. Operationalise DSIT's "agile" direction with a portfolio-level objective function. DSIT's stated intent to manage R&D funding in a more agile way requires a replacement objective function, not just a relaxation of BCR requirements. DSIT should update business case guidance for research infrastructure to adopt Cost of Delay as the primary prioritisation metric: per-period scientific and strategic value foregone by deferring each investment, divided by delivery duration. This aligns with the logic of agile portfolio management, demands only near-term capability assessments rather than multi-decade demand projections, and makes DSIT's stated goal of deprioritising underperformers and doubling down on high-impact investments operationally tractable.
3. Establish a formal Standby governance category for the Infrastructure Fund. The aircraft case — £46 million spent before a decommissioning decision — illustrates the cost of binary governance. A Standby category preserves option value while halting new capital commitment, allowing evidence to accumulate before a full Exit decision is required.
4. Act on NAO recommendation (d) to remove the CAPEX-bias created by classifying all UKRI spending as capital. DSIT and UKRI should introduce governance frameworks that assess new construction, maintenance, and operational access against equivalent criteria. Without this reform, the £5.6 billion maintenance backlog will continue to grow as capital programmes crowd out maintenance investment.
5. Mandate a consolidated annual portfolio utilisation report. DSIT should require UKRI to publish a comparative annual utilisation report across all national research infrastructure — enabling portfolio-level reallocation decisions. The NAO explicitly identifies the absence of consistent monitoring as a constraint on DSIT's landscape understanding; this recommendation makes the constraint visible and accountable.
Averch, H., & Johnson, L. L. (1962). Behavior of the Firm Under Regulatory Constraint. The American Economic Review, 52(5), 1052–1069.
DSIT. (2025). DSIT Research and Development (R&D) plans to 2029/2030.
Flyvbjerg, B., Bruzelius, N., & Rothengatter, W. (2003). Megaprojects and Risk: An Anatomy of Ambition. Cambridge University Press. https://doi.org/10.1017/CBO9781107050891
National Audit Office. (2016). BIS’s capital investment in science projects (HC 885). NAO. https://www.nao.org.uk/wp-content/uploads/2016/03/Capital-investment-in-science-projects.pdf
National Audit Office. (2026). DSIT’s investment in research infrastructure (HC 1735). NAO. https://www.nao.org.uk/wp-content/uploads/2026/03/DSITs-investment-in-research-infrastructure.pdf
Reinertsen, D. G. (2009). The Principles of Product Development Flow: Second Generation Lean Product Development. Celeritas Publishing.
Shabad, V., Yusupov, R., & Ivanov, K. (2026). Technology Investment Governance Under Deep Uncertainty: The Case for Annual Portfolio Models in Public Infrastructure (6469862). SSRN.
Staw, B. M. (1976). Knee-deep in the big muddy: a study of escalating commitment to a chosen course of action. Organizational Behavior and Human Performance, 16(1), 27–44. https://doi.org/10.1016/0030-5073(76)90005-2
UK Research and Innovation. (2020a). The UK’s research and innovation infrastructure: Landscape Analysis. UKRI. https://www.ukri.org/wp-content/uploads/2020/10/UKRI-201020-LandscapeAnalysis-FINAL.pdf
UK Research and Innovation. (2020b). The UK’s research and innovation infrastructure: opportunities to grow our capability. UKRI. https://www.ukri.org/wp-content/uploads/2020/10/UKRI-201020-UKinfrastructure-opportunities-to-grow-our-capacity-FINAL.pdf
AI use: During the preparation of this submission, the author used Claude (Anthropic) to improve readability and language quality as a non-native English speaker. After using this tool, the author reviewed and edited the content and takes full responsibility for the content of this submission.
Competing interests: The author has no current financial interest in the outcome of this inquiry. The author is a co-author of the academic work cited in this submission (Shabad et al., 2026).