THINKING ALOUD

2025: The year AI came of age – and the questions ahead for 2026

The technology has moved from boardroom curiosity to balance sheet necessity

Summarise
阅读简体中文版 (beta)
    • 2025 proved AI's commercial viability beyond doubt. The challenge for 2026 is ensuring that velocity doesn’t outpace wisdom.
    • 2025 proved AI's commercial viability beyond doubt. The challenge for 2026 is ensuring that velocity doesn’t outpace wisdom. IMAGE: REUTERS
    Published Wed, Dec 24, 2025 · 03:00 PM

    AS 2025 draws to a close, artificial intelligence (AI) has unmistakably transitioned from experimental technology to essential infrastructure.

    This was the year AI moved from boardroom curiosity to balance sheet necessity, reshaping how we work, create and compete on the global stage. The numbers tell a compelling story – 44 per cent of US businesses now pay for AI tools, up dramatically from just 5 per cent in 2023, while average AI contracts reached US$530,000, according to Air Street Capital and AI investor Nathan Benaich’s State of AI Report 2025. More striking still: AI-first startups grew 1.5 times faster than their peers. These aren’t pilot projects anymore – they’re core operations.

    The technical achievements were equally remarkable. AI system performance surged across major benchmarks, with scores rising by 18.8, 48.9 and 67.3 percentage points on MMMU (designed to evaluate multimodal models), Graduate-Level Google-Proof Q&A (GPQA) and SWE-bench (for evaluating large language models on real world software issues collected from GitHub), respectively, in just one year.

    Perhaps most significantly, language model agents outperformed humans in programming tasks in some settings, signalling AI’s evolution from assistant to autonomous actor.

    Yet 2025 also exposed the growing pains of rapid adoption. The geopolitical race intensified, with the US producing 40 notable AI models compared to China’s 15 and Europe’s three, though Chinese models rapidly closed the quality gap, with performance differences shrinking from double digits to near parity. Meanwhile, AI-related incidents rose sharply, while standardised safety evaluations remained rare among major developers.

    The regulatory landscape fragmented dramatically. In the US, the Trump administration’s January executive order reversed the Biden-era approach, prioritising innovation over oversight.

    Thirty-eight states enacted about 100 AI-related measures in 2025, creating a patchwork that has businesses scrambling for compliance. Texas, California, Colorado and Illinois all have comprehensive AI laws taking effect throughout 2026, covering everything from hiring discrimination to frontier-model safety.

    Looking towards 2026, several critical questions loom large. Can we bridge the gap between recognising AI risks and taking meaningful action? Will the ongoing federal-state regulatory tension resolve into coherent governance, or will fragmentation stifle the innovation both sides claim to protect?

    And as enterprises push beyond content generation into AI reasoning for their own data, how will we ensure transparency and accountability in systems making increasingly autonomous decisions?

    The infrastructure challenge cannot be ignored. Multi-gigawatt data centres backed by sovereign funds signal a new wave of compute infrastructure, with power supply emerging as the new constraint. The question isn’t just whether we can build AI systems smart enough to transform industries – it’s whether we can power them sustainably and govern them responsibly.

    2025 proved AI’s commercial viability beyond doubt. The challenge for 2026 is ensuring that velocity doesn’t outpace wisdom. As AI systems grow more capable and autonomous, our frameworks for oversight, accountability and international cooperation must mature just as rapidly.

    The technology has come of age. Now our governance must catch up.