Public AI adoption money buys the first months, not the person who keeps the system running

The pattern is clearest in the UK. Its AI Upskilling Fund, a £7.4 million pilot open only to professional and business services firms, paid up to half the cost of external AI skills training, up to £2,500 for a micro firm and £10,000 for a medium one; it would not pay for AI technology, consulting, recruitment or internship salaries, and all funded activity had to end by 31 March 2025.

The government's own evaluation, published in January 2026, shows how thin that kind of help can be. The pilot was meant to train 4,300 employees in 2,000 businesses; in the end 327 firms were approved and £381,096 was awarded. Only 55% of approved firms went on to claim their grant, and the planned randomised trial was judged unfeasible because applications were lower than expected.

The evaluation also shows where firms expected the value to go. Of 73 approved firms surveyed, 70% hoped trained staff would pass their skills on to colleagues, and 14% named the risk that trained staff would leave without doing so.

Elsewhere the money takes the form of advice. Australia's AI Adopt Centres give eligible small firms free training, one-to-one consultations, roadmaps and demonstrations. The EU pays up to half the cost of its European Digital Innovation Hubs for 36 months, and each hub decides whether firms pay anything for its services. In the US, one federal route is the Manufacturing Extension Partnership, whose centres must raise at least 50% of their funding from non-federal sources and list AI among the technologies they help firms adopt.

FDE AI reads these designs as sensible ways to lower the cost of starting. What they do not fund is the employee who keeps the work going once the expert has left.

The newest rounds follow the same shape. Innovate UK's BridgeAI Deployment Catalyst opened on 21 October and closes at 11:00 UK time on 8 December 2026; it offers grants of £100,000 to £200,000 per project, covering up to 70% of eligible costs depending on size, for 3 to 5 months, and every project must end by 31 August 2027. The lead applicant must be a UK small or medium-sized firm that already has an AI solution to deploy, but the deployer must be a medium-sized or large firm or a public body, so small firms cannot take part as the adopter. This call is the first of three possible phases. Innovate UK says a third phase of about 22 months would cover live deployment, user adoption and benefits measurement, but it calls the later phases indicative, not a commitment.

Canada added C$500 million over five years on 13 October, bringing its Regional Artificial Intelligence Initiative to C$700 million; regional agencies deliver it, and intake dates vary by region. In southern Ontario applications close on 19 January 2027, or on 24 November 2026 for projects whose costs start before 31 March 2027. FedDev Ontario's adoption stream lists building AI expertise, training and change management among eligible activities, but the money is a repayable contribution of C$125,000 to C$5 million covering up to 50% of costs, for firms with at least five full-time equivalent employees and three years of trading. Nationally the initiative counts 218 projects since 2024, creating or maintaining more than 1,100 jobs. Those are counts of projects and jobs, not of systems still in use.

Where programmes look at later results, they measure the firm, not the role. The EU hubs repeat each firm's digital maturity assessment 18 to 24 months after support, the latest confirmed per-firm follow-up among the schemes reviewed here. Even that measures a maturity score, not whether anyone inside the firm is responsible for keeping the systems working.

Singapore's Enterprise Compute Initiative draws the line in a different place. It pays 70% of consulting costs, up to S$105,000 per company, for up to six months, but only for firms with CEO-level sponsorship, at least 10 staff in Singapore and a tech team of 2 or more. The use case is meant to lead to an AI team inside the company, and the firm must produce a 12-month change plan and a product tested with at least five end-users who give written feedback.

Enterprise Singapore's newer Champions of AI programme goes a step further and lists building an AI team to sustain the change among its support areas. Neither scheme's published terms say it will pay for that team after support ends; the Champions of AI page publishes no funding terms at all, and the Enterprise Compute Initiative is aimed at larger firms than a typical small business. But they show that a programme can ask, before it pays, who will still be running the system.

The adoption data explains why the question matters now. In September 2026, 29% of UK firms with fewer than 10 employees used AI, against 51% of those with 250 or more. In the US the national rate was 19.8% in May, against 37% for firms with 250 or more employees and under 20% for those with four or fewer. Australia's statistics office puts small and micro businesses at around 11% in 2024 to 2025, against around 35% for large ones. Eurostat's narrower measure puts EU firms with 10 or more employees at 20.0% in 2025, up from 13.5% a year earlier. In a broader survey, 42% of EU small and mid-sized firms say they use AI or algorithmic management, mostly in limited, trial-stage ways, and 48% cite a lack of expertise. These surveys define use differently and should not be compared directly, but each one that reports by size shows small firms behind.