Yes — but not by buying what the leaders bought. The tools that gave early adopters their head start are now cheap and nearly interchangeable; the advantage that remains sits in the slow work around the tools: redesigned processes, usable data, and people who know what to do with the output. A late adopter that invests there can close the gap. One that only buys software will stay behind with better software.
The conventional answer is bleaker. In 2018, Vikram Mahidhar and Thomas Davenport argued in Harvard Business Review that the fast-follower strategy would not work for AI and that late adopters might never catch up. Half of that argument has aged well. The other half has not, and the difference tells a lagging company exactly where to spend.
Is the gap between AI leaders and laggards still widening?
Yes, by the best available measure. BCG’s 2025 survey of 1,250 companies classified only 5% as “future-built,” achieving AI value at scale, while 60% reported little or no material value despite substantial investment. The leaders report 1.7 times the revenue growth and 3.6 times the three-year shareholder return of the laggards, and they plan to spend more than twice as much on AI.
Two cautions apply. The categories are the consultancy’s own, and much of the data is self-reported. Neither caution reverses the direction: the distance is real and growing. Why the leaders got there is a separate question, covered in what separates the companies pulling ahead from those falling behind. Here the question is narrower: what the distance is made of, and which parts of it a laggard can close.
What can a late adopter now buy that leaders had to build?
Almost all of the technology. Stanford’s 2025 AI Index estimates that the cost of querying a model at GPT-3.5-level performance fell from $20 per million tokens in November 2022 to $0.07 by October 2024, a reduction of more than 280-fold in about eighteen months. Over a single year, the performance gap between the top-ranked model and the tenth narrowed from 11.9% to 5.4%.
A company starting today gets, at a fraction of the price, capability that early movers paid to pilot. On the tool layer, waiting was not a mistake. It was a discount.
What can’t be bought, and why does it take so long?
The complements. Brynjolfsson, Rock and Syverson’s work on the productivity J-curve (2021) shows that general-purpose technologies require large intangible investments (new processes, new skills, new business models) that depress measured productivity before they pay off. This is the half of the 2018 argument that holds: AI usually supports tasks, not whole processes, so value appears only after the work around it is redesigned.
That redesign cannot be downloaded. It is also where BCG’s future-built firms differ most: nearly all report a deeply engaged C-suite, against 8% of laggards, and they rebuild operating models around human–AI workflows rather than layering tools onto old ones.
This is the mechanism behind asymmetric adaptation, the pattern in which AI narrows some gaps and widens others: it compresses differences between firms where the tool is the bottleneck and amplifies them where the complements are. Leaders are not ahead because they have more AI. They are ahead because they finished the expensive part of the J.
Do late movers have any real advantage?
Yes, but a bounded one. Lieberman and Montgomery’s classic analysis of first-mover advantage (1988) identifies what late movers gain: they free-ride on pioneers’ investments, enter after technological uncertainty resolves, and avoid commitments to designs that turn out to be wrong.
All three apply to AI now. Laggards can learn from documented failure modes, skip architectures that early movers are locked into, and buy at today’s prices. The advantage is weakest where leaders have built proprietary data loops, systems that improve with every use. Those compound, and no purchase replicates them.
Where should a company that fell behind start?
With a workflow, not a platform.
- Choose two or three workflows with measurable output, and redesign each end to end before selecting tools.
- Budget the complements explicitly. If the plan is mostly licenses, it is a plan to stay behind with better software.
- Put the program under the CEO, not under IT. In BCG’s data the sponsorship gap is wider than the spending gap.
- Keep the tool layer cheap and swappable. The late mover’s advantage disappears the moment it locks into a single vendor.
The emphasis shifts by region. US firms are often tempted to close the gap by acquisition, which buys capability but not the redesign. In the Gulf, where executive sponsorship can move quickly, the binding constraint is more often process depth than commitment. In Central and Eastern Europe, cost discipline becomes an asset: the late-mover discount rewards precisely the firms that declined to overpay for early pilots.
Frequently asked questions
How long does it take to catch up?
Longer than buying tools, shorter than the leaders took. Complements take years rather than quarters to build, but a late adopter skips much of the experimentation the pioneers paid for. No credible study offers a single figure.
Should we wait for AI to mature before investing?
On tools, waiting has paid. On complements, it has not: process redesign and skill-building take roughly the same time whenever they start, so every month of waiting there is a month added to the gap.
Can we catch up by acquiring an AI startup or hiring a chief AI officer?
Both buy capability. Neither redesigns your workflows. Acquired capability still has to be adapted to how your business actually works, and that adaptation is the slow part.
Does the late-mover advantage apply in every industry?
No. It is strongest where AI serves general-purpose tasks and weakest where leaders have built proprietary data that improves with every use.
Sources and further reading
- BCG (2025). The Widening AI Value Gap: Build for the Future 2025. https://media-publications.bcg.com/The-Widening-AI-Value-Gap-October-2025.pdf
- Stanford HAI (2025). AI Index Report 2025.
- Brynjolfsson, E., Rock, D., & Syverson, C. (2021). The Productivity J-Curve: How Intangibles Complement General Purpose Technologies. American Economic Journal: Macroeconomics, 13(1), 333–372. https://www.aeaweb.org/doi/10.1257/mac.20180386
- Lieberman, M. B., & Montgomery, D. B. (1988). First-mover advantages. Strategic Management Journal, 9(S1), 41–58.
- Mahidhar, V., & Davenport, T. H. (2018). Why companies that wait to adopt AI may never catch up. Harvard Business Review, December 2018.
