My observation is that almost every technological revolution begins with abundance, too many players, too many promises. The industrial era had railroads. The internet had dot-coms. Artificial intelligence is now entering its own equivalent phase. If I put this into a form of a metaphor, in evolutionary biology, periods of rapid diversification are always followed by mass extinctions. Not because species fail to adapt, but because only a few environmental niches can sustain long-term survival. The AI sector is seeming to be entering its own evolutionary bottleneck. 

What looks today like explosive growth and endless opportunity is, in reality, the prelude to a market consolidation cycle. The AI industry is approaching a structural shakeout where only a small number of platforms, ecosystems, and operating models will capture the majority of long-term value.

The global AI market is projected to grow from roughly $375 billion in 2026 to around $2.48 trillion by 2034, reflecting extraordinary investment momentum. Yet history seems to suggest that growth at this scale does not produce many winners, it produces a few dominant architectures and many displaced ambitions. For business leaders, the critical question is no longer whether AI will reshape industries, but how market share in AI will consolidate and what structural forces will determine who survives that consolidation.

I am sharing here 3 dynamics that has the potential to define this shakeout.

1. The Shift from Model Competition to Ecosystem Control

Public discourse frames AI competition as a race between models: which foundation model is smarter, larger, or more capable. This narrative is already becoming obsolete.

The real battle is not over models, it is over ecosystems.

My observation is that as AI becomes embedded in enterprise workflows, value increasingly shifts toward platforms that control:

• cloud infrastructure
• data integration layers
• developer toolchains
• compliance and governance systems
• vertical application marketplaces

In other words, the winners will be those who design the environment in which AI operates, not merely the intelligence itself.

Cloud infrastructure data already reflects this dynamic. Hyperscalers continue to dominate spending and increasingly bundle AI services directly into core enterprise stacks. This means that many innovative AI companies will not fail because their technology is inferior, they will fail because they are positioned outside dominant ecosystems.

For business leaders, this has a subtle but powerful implication; the strategic risk is not choosing the wrong AI model , it could be aligning with the wrong AI platform.

2. The Geopolitical Fragmentation of AI Markets

The AI market is not global in the traditional sense. It is becoming geopolitically segmented.

A significant but under-appreciated trend is the redistribution of AI research leadership. China now accounts for a rapidly growing share of global AI publications, reshaping the geography of innovation.

This shift is already producing real commercial consequences. Export controls on advanced chips are accelerating the development of domestic AI stacks in China, reducing reliance on Western providers. 

At the same time, regulatory regimes are diverging:

• The EU is prioritizing safety and governance.
• The US is prioritizing innovation speed.
• China is prioritizing state-aligned infrastructure.

The result is not a single global AI market but multiple regional AI economies, each with distinct standards, architectures, and champions. For multinational firms, this introduces a new strategic complexity. Hence, my observation is that AI strategy must now resemble geopolitical strategy balancing localization, compliance, data sovereignty, and ecosystem partnerships across jurisdictions. Those who design multi-architecture strategies, not monolithic global deployments, should have an edge.

3. The Frontier-Laggard Divide Inside Enterprises

Perhaps the most overlooked dimension of the AI shakeout is not between companies but within them. Early enterprise adoption data shows a widening gap between “frontier organizations” that deeply integrate AI into decision processes and the majority that use AI as an experimental layer.

Usage metrics already reveal that frontier adopters engage more frequently, across more functions, and with higher strategic integration.  

This divide is not technological. It is organizational.

Frontier firms are redesigning:

• how decisions are made
• how insights flow into leadership forums
• how experimentation replaces static planning
• how governance enables speed instead of constraining it

Laggards, by contrast, often treat AI as a tool rather than a system deployed at the edges but excluded from core strategic logic. The coming shakeout will therefore not simply eliminate firms that fail to adopt AI. Seeming that it will reward firms that restructure themselves around AI-driven cognition, where learning cycles replace planning cycles, and strategy becomes a real-time process.

The Structural Reality of the Shakeout

The AI shakeout will not resemble a typical technology cycle. It will be quieter, more structural, and more permanent.

It will not be announced by bankruptcies. It will be revealed through:

• shrinking ecosystem relevance
• declining developer engagement
• reduced access to strategic partnerships
• loss of platform influence
• inability to attract frontier talent

In this sense, the AI shakeout is less about failure and more about gravitational drift. Organizations are seeming to be slowly moving out of the centers where innovation, capital, and strategic power accumulate.

What This Means for C-Suite Leaders?

The market share battle in AI does not look like about buying tools. It is about ‘designing position’.

The leaders who emerge from the shakeout will be those who:

• treat ecosystem alignment as a core strategic decision
• build geopolitical adaptability into AI roadmaps
• redesign governance to enable continuous learning
• integrate AI into leadership cognition, not just operations

My take is that AI will not replace strategy. But it will replace static strategy. 

My Final Reflection

Every major technological shift produces noise first and structure later.

We are leaving the noisy phase of AI and entering the structural phase.

The shakeout will crown the smartest architectures rather than models. And in the end, the true competitive advantage will not be artificial intelligence but organizational intelligence shaped by artificial intelligence.

A summary for Finance Directors

The recent emergence of activists on the shareholder registers of many Pan-European blue chips has highlighted a new type of risk for Finance Directors. This risk, related indirectly to MiFID and to liquidity fragmentation, is that of being caught by surprise by the eruption of an activist. The risk is not so much about control of your company, but more about interference, rise of a shareholder activist plus the time and resources that have to be expended to deal with activist demands.

How can a corporate accurately monitor its trading flows when the primary markets now account for less than half of total equity volumes? Many corporates admit to being “blind” and devoid of regular, reliable, information on this topic. Unfortunately, it’s no longer sufficient to have a friendly contact on a trading floor who may be able to explain unusual flows or an erratic share performance.

1. What are the implications of MiFID2 for liquidity fragmentation?

The first change of real significance dates from 1 November 2007 with the implementation of the European Directive on Financial Markets (MIF). The underlying objective was to create competition between the various platforms where orders were executed, i.e. Executive Recruitment with Investment between the primary markets (for example, Euronext in France) and Multilateral Trading Facilities (MTFs).

This has lowered transaction fees (the original objective) and accelerated the development of algorithmic trading and other electronic solutions (so-called “low-touch” trading, as opposed to “high touch” — more commonly known as voice trading). On average, 75% of European blue chip volumes (66% of midcap volumes) are now traded on low touch systems.

However, the growing diversity of platforms and solutions (dark pools, systematic internalisation (SI), intraday fixing, etc.) has also reduced visibility for finance departments on their volumes traded. It is not unusual for the primary market today to rise of a shareholder activist account for just one-third of a blue chip’s total traded volumes. Liquidity and volumes have become so fragmented that monitoring the plethora of trading platforms is becoming a near-impossible task for most finance departments.

2. What are dark pools and how do they function?

A dark pool is an automatic transaction system that, as the name suggests, is opaque (put simply, orders are not visible to other investors). The transaction only becomes visible once it has been completed (post-trade transparency).

Dark pools allow investors to trade significant positions anonymously, without any market impact. It is for this reason that they are often used by activists to build up a position quickly and unobserved.

On average, dark pools account for 4–6% of a European blue chip’s total traded flows. When these flows regularly exceed 20–30% and/or occasionally peak above 50%, there is reason to suspect that an activist is building a position.

3. How can the rise of a shareholder activist be detected? What mechanisms are available to build a position?

The list of trading opportunities available to an investor wishing to build a position (long or short) is endless.

To start with so-called “vanilla” solutions, the greater the liquidity, the easier it is to build a large position in just a short time. An investor can go to the listing market, seek blocks (although that has the drawback of attracting attention) or turn to dark pools. This latter solution has the advantage of allowing an investor to buy or sell a big stake without making any impact on the market and without being identified.

Then there are “structured” solutions. Changes are visible (i) on the lending/borrowing market and (ii) on the various markets on which the stock is listed. This market practice highlights a major development that should be integrated in any market monitoring: the growing porosity of the various asset classes. For example, the Exane BNP Paribas Daily360 includes the analysis of the lending/borrowing market (shorts), CDSs (credit risk) and options.

In our experience, it is generally possible to spot signs of potential activity by an activist. However, a structured and detailed analysis is necessary to identify true anomalies. Possible warning signs often turn out to be the result of normal activity, such as index re-balancing or a dividend payment. Experience enables an alert broker to distinguish between such cases and potential indications of “real” activity by an activist.

Stakes are usually built up silently, which allows an activist to take advantage of the “announcement impact”, which is generally an increase in the share price of the target. An activist relies on the surprise effect in order to (i) gain most benefit from the announcement and (ii) to catch management out and at the same time render a counter-attack difficult.

A management team that has received an early warning will have had the time to mount a proactive defense strategy, such as carrying out an internal audit, preparing counter-arguments, and, if deemed appropriate, seeking an ally or white knight (BNP Paribas has a unique experience of such mandates, built up over many years). This grace period which may last several weeks but could be just a few hours in the case of aggressive stake-building, is critical.

This document is neither a recommendation nor a piece of technical. Recipients of this document are invited to make their own independent decision to subscribe to the services described herein (the “Services”) based on their own judgment, considering its specific needs and, as the case may be, upon the advice of any relevant and independent advisors they have consulted (including legal, tax, accounting or other specialist advisors). The information contained in this document does not take into account the specific needs of the Services users. Exane cannot be held liable for any damages or losses resulting from the use of this information or part thereof.

Benedicte Thibord, Alexis Basser – May 2021