Read the passage and mark the letter A, B, C or D on your answer sheet to indicate the best answer to each of the following questions from 31 to 40.
Calls to treat compute as a public utility signal the strategic return of the state, not least through industrial policies like the U.S. Inflation Reduction Act and the EU’s Net-Zero Industry Act. In AI, compute – the capacity to train and serve large models – has become the decisive lever. The UK’s £900m AI Research Resource (AIRR) is framed as world-class infrastructure to widen access and seed public-interest innovation. [I] Yet if access remains scarce and extractive, such investment could merely gild existing hierarchies instead of redressing them.
AI relies on an interlaced stack: chips and data-center infrastructure, orchestration software, model development and hosting, then downstream applications. Where this stack is vertically integrated or steered by venture-capital imperatives, frontier research narrows, and “open” projects become dependent on hyperscalers’ platforms and purse strings. Partnerships that canalize public research into private clouds risk entrenching monopoly control over experimentation, evaluation, and deployment. [II] The upshot is a compute divide that attenuates academic participation and sidelines smaller firms with socially valuable but less lucrative aims.
On present trajectories, the gulf between researchers’ needs and domestically available capacity will widen, especially where public provision is thin. Governments can procure cloud credits as an interim fix while seeding on-shore capacity and “friend-shoring” resilient supply chains. [III] Relying on hyperscalers to fulfil short-term demand, while deferring public capacity-building, risks deepening structural dependency. Without credible plans for governance, efficiency, and sustainability, procurement becomes a subsidy to incumbents rather than a pathway to plural, equitable research ecosystems.
Public compute should be governed to create public value: user-centric boards; safety, auditing, and documentation duties; contributions to digital commons; and efficiency commitments that reward doing more with less compute. Treating major compute providers as utilities – ensuring fair dealing and interoperability – would curb lock-in and enhance contestability. [IV] To move beyond an arms race narrative, policymakers must define public benefit precisely, then align access rules, funding, and oversight with that vision so that AI serves society rather than merely scale.
(Adapted from Ada Lovelace Institute, “The role of public compute”.)
Question 31. The word gulf in paragraph 3 mostly means ______.
A. roughly equivalent B. marginally divergent
C. vastly asymmetrical D. neatly congruent
Question 32. Where in the passage does the following sentence best fit?
Pooling capacity under AIRR is intended to democratise experimentation by lowering the prohibitive costs individual labs currently face.
A. [I] B. [II] C. [III] D. [IV]
Question 33. Which of the following best summarises paragraph 2?
A. Venture-driven integration concentrates control of the AI stack, shrinking genuine openness and constraining non-commercial research.
B. Public labs already dominate compute markets, ensuring academic freedom and broad experimental diversity across disciplines.
C. Hardware scarcity matters less than software tooling, which guarantees fair access regardless of cloud ownership structures.
D. Decentralised hobbyist clusters outcompete hyperscalers, reversing dependence and restoring pluralism in foundation-model research.
Question 34. What is framed as a short-term stopgap?
A. Banning vertical integration B. Building national fabs
C. Buying cloud credits in bulk D. Mandating open weights
Question 35. According to paragraph 4, treating compute providers as utilities would ______.
A. require fair dealing and interoperability to limit customer lock-in across the stack
B. guarantee free access for all researchers regardless of safety obligations or quotas
C. eliminate venture capital funding from AI infrastructure within five fiscal years
D. nationalise every major data centre and centrally set all training priorities
Question 36. What does effective AIRR governance need to include, as suggested in the passage?
A. Exclusive partnerships with hyperscalers to accelerate procurement and suppress duplicative projects
B. User-centred boards, safety auditing, commons contributions, and efficiency-linked access incentives
C. A permanent moratorium on frontier research to conserve energy and reduce ecological footprints
D. Unlimited compute allowances for accredited universities without documentation or throughput reporting
Question 37. The phrase arms race in paragraph 4 refers to ______.
A. chip yields B. peace talks
C. zero-sum competition D. budget windfall
Question 38. Which of the following can be inferred from the passage?
A. Public boards should avoid engaging users, since expert-only governance accelerates safe and plural innovation.
B. The UK already commands ample supercomputing; additional public compute would be redundant for academic laboratories.
C. Open-source releases alone are sufficient to neutralise market concentration throughout the compute supply chain.
D. Without clear public-benefit criteria, investments risk fortifying incumbents rather than broadening socially valuable AI research opportunities.
Question 39. Which of the following best paraphrases the underlined sentence in paragraph 3?
Relying on hyperscalers to fulfil short-term demand, while deferring public capacity-building, risks deepening structural dependency.
A. Contracting with hyperscalers briefly then cancelling guarantees sovereignty regardless of domestic infrastructure development or long-term planning efforts.
B. Buying bulk credits today and expanding public capacity tomorrow will permanently eliminate reliance on commercial platforms for research infrastructure.
C. Meeting near-term needs solely through big clouds while postponing state capacity would lock public research into lasting dependence on incumbents.
D. If governments delay procurement and rush capacity-building, vendors will forego leverage and encourage decentralization promoting competitive research ecosystems.
Question 40. Which of the following best summarises the passage?
A. Private leadership in AI is inevitable; public investment should avoid governance and concentrate on speed alone.
B. Public compute, governed for public value, can counter concentration and align AI with democratically articulated social aims.
C. The only barrier to plural AI is hardware scarcity; policy choices are largely irrelevant to outcomes.
D. Academic projects should exit public funding and rely solely on venture capital for experimentation.