The AI ecosystem is awash in capital and rife with transactions, but it is not yet experiencing an M&A wave. A BCG analysis of 41,149 majority acquisitions in the technology sector since 2020, of which 11,372 are linked to the AI ecosystem, finds that strategic dealmaking has broadened from buying technology and talent to assembling platforms and scale. However, the analysis also finds that the ecosystem has barely begun to consolidate.
That distinction matters. Funding rounds, infrastructure commitments, and commercial partnerships show where companies and investors are committing capital and resources. Strategic M&A transactions show whether AI is forming a competitive structure of its own or is being absorbed into existing structures.
Using 2020 as the base year, the deal data shows that this question remains open—even six years into the most heavily capitalized technology build-out in a generation.
CEOs, boards, and corporate development leaders should not wait for an ecosystem-wide consolidation wave to take shape before preparing to make deals. Different AI ecosystem segments are maturing at different speeds. To succeed in dealmaking as the industry evolves, leaders need to distinguish capability purchases from platform-building moves, understand which boundary in the stack matters to their strategy, and develop the due diligence and integration discipline required as deals become larger and more interdependent.
Most AI Transactions Are Not Acquisitions
Reading the M&A market correctly starts with separating strategic acquisitions from the much larger flow of capital into and around AI.
Transaction structures blur the line. Licensing-plus-talent arrangements can transfer valuable technology and senior engineering teams without transferring ownership. Minority investments can cement a strategic relationship without conferring control. And single outsized transactions can overwhelm aggregate deal values while saying little about the underlying pattern: Nvidia's reported $20 billion licensing arrangement with Groq is larger than its reported $12.9 billion agreement to acquire Hugging Face, yet only the latter is an acquisition. LSEG M&A data, which captures minority and majority transactions but not pure commercial partnerships or financing, reveals that only about one-third of M&A transaction volume qualifies as strategic dealmaking; the remainder is driven primarily by financial investors. (See Exhibit 1.)
For our analysis, we examined only strategic majority acquisitions and mergers involving technology companies, focusing on players in the AI ecosystem. We differentiated the AI ecosystem into five segments: compute, data center, and physical infrastructure; data and AI development infrastructure; models, inference, and access; applications, agents, and workflows; and AI deployment and transformation services. We mapped acquirers and targets to an ecosystem segment on the basis of their business descriptions, applying a simple AI-relevance threshold: we included a company only in instances where AI was central to the product, asset, business unit, or stated transaction rationale.
Our sample excludes transactions by financial investors, minority investments, and other noncontrol deals. It also excludes the physical build-out of data centers, power plants, and chip factories, which is financed largely through hyperscaler free cash flow, project debt from asset managers, selective equity issuance, and venture and private capital backing for neocloud providers. That build-out has more in common with the fiber and rail programs of earlier eras than with a wave of acquisitions, and including it would obscure rather than illuminate the strategic deal flow.
What remains is the part of the market where companies are changing their competitive positions rather than funding their growth—and that part behaves in a recognizable way.
Acquisition Motives Follow a Maturity Arc
As a dominant structure takes shape in any industry, competition begins to shift away from experimentation and toward scale and efficiency, and the number of companies tends to shrink. Acquisition motives move along the same arc, which we simplify into four phases:
- Phase 1: Acquire capabilities. Buyers purchase technology, talent, or know-how that they cannot build quickly enough themselves. Targets are small, often prerevenue, and valued for what they can do rather than what they earn.
- Phase 2: Expand products. With core capabilities secured, acquirers broaden the offering—filling gaps in the product line, adding features, and moving from point solutions to more complete propositions.
- Phase 3: Assemble platforms and scale. Deals get larger and more strategic as they focus on such objectives as entering adjacencies, tying products into an integrated platform, and buying volume, distribution, or geographic reach. Often a small group of serial acquirers accounts for a growing share of dealmaking activity.
- Phase 4: Consolidate. Horizontal transactions arise as buyers acquire direct competitors—firms selling substantially the same product to the same customers—a reflection of the fact that maturing categories can’t support a large field of rivals. The strategic logic shifts toward cost and market position, and the buyer set narrows significantly.
The phases are neither rigid nor mutually exclusive. Capability buying continues in mature industries; consolidation begins early in some categories and never dominates others. Read as a maturity arc rather than a sequence, the mix of motives shows where an industry's dealmaking currently sits—and whether in certain areas it has moved earlier or later than the pattern would predict. When applied to acquisitions made by AI-ecosystem companies themselves, it produces an unusually clear reading.
A Few Years In, the Ecosystem Is Still Buying Capabilities
From January 2020 through August 2026, capability-building and product expansion deals account for 4,394 and 2,653 transactions, respectively; together, they represent 95% of all acquisitions by ecosystem companies. That mix has been remarkably stable, with capability buying easing only from 63% to 57% of annual volume since 2020. The dominant motive in AI remains what it was at the outset: obtaining a team or a technology that would take too long to build.
Platform assembly is the only motive moving materially. Those deals have risen from 26 transactions and 3% of volume in 2020 to 67 transactions and 9% of volume in the first part of 2026— roughly a tripling in share. (See Exhibit 2.) This is a clear indication that strategic logic is maturing. A minority of acquirers has moved from only buying capabilities to buying a position in the ecosystem.
Consolidation, however, has not begun. Horizontal deals account for 23 transactions across the entire period, just 0.3% of the sample, with no trend at all. The strongest single year was 2020, with eight. Moreover, all 23 deals involve the compute, data centers, and physical infrastructure segment. In a field that contains thousands of funded companies in the applications segment alone, the absence of activity in other areas is as informative as any positive finding in the analysis.
In terms of maturity, the segments do not land where the public narrative would put them. (See Exhibit 3.) Compute, data-center, and physical infrastructure—the segment representing more than half the sample—has the ecosystem’s lowest platform-assembly rate and all of its consolidation deals. That is what a mature structure looks like: buying capabilities and products within settled boundaries, and buying rivals in categories that have run out of room. This places it in almost complete opposition to the platform-building stage that its headline transactions imply. In contrast, the applications, agents, and workflows segment produces nearly as many platform-assembly deals on roughly one-third of the volume—but two-thirds of its strategic transactions are still capability buying, a field that is being assembled rather than reduced.
Two segments deserve closer attention. Data and AI development infrastructure shows the highest platform-assembly rate outside the models segment and no consolidation whatsoever. In this segment, acquirers are building integrated toolchains from a fragmented field of specialists, a step that typically precedes consolidation. It is the segment where the ecosystem's first substantive horizontal activity is most likely to occur, and the behavior of an identifiable handful of serial acquirers over the next 12 to 18 months will determine how quickly that moment arrives.
Models, inference, and access—the segment that dominates public attention—accounts for less than 1% of strategic acquisitions. Its activity forms a barbell: model companies buy either capabilities or ecosystem position, with a single product-expansion deal in six years providing the bar. The buyer set is small and concentrated, with more than one-third of segment deals coming from a single company, and capital reaches the segment through funding rather than through acquisitions. A segment that is so low in strategic acquisitions is unlikely to be in equilibrium.
Absorption Is Running in Both Directions at Once
Roughly seven in ten intra-ecosystem transactions involve parties that operate within the same segment. Deals within the compute, data-center, and physical infrastructure segment represent the single largest flow. Currently, aside from hyperscalers, the segments of the AI stack seem to be building separately rather than assembling into one another.
Where deals cross boundaries, value concentrates. Cross-segment transactions represent only 28% of intra-ecosystem activity, but the largest deals in our sample cross a boundary rather than staying inside one. The direction of deal flow differs for deal volume and deal value. For deal volume, up-stack and down-stack flows are nearly balanced. For deal value, the up-stack transactions dominate—including Nvidia's reported agreement for Hugging Face and SpaceX's completed $60 billion purchase of Cursor’s parent—while down-stack activity consists largely of smaller semiconductor and data center deals. The narrative of infrastructure owners extending into software is visible in the value data but not in the volume data.
The most consequential pattern is directional in a different sense. Among transactions for AI targets, buyers outside the ecosystem outnumber those within the ecosystem roughly two to one—the signature of an industry being absorbed into existing structures. But ecosystem companies are themselves highly acquisitive, and their acquisitions of non-AI targets outnumber their acquisitions of each other by nearly three to one. (See Exhibit 4.) Absorption is running both ways. That is not the same as AI being folded into enterprise software, and it is not the same as AI forming a competitive structure of its own. It is an industry whose structure remains to be determined, which is precisely why the positions taken in the next two years will be difficult to reverse.
What This Means for Your Next Deal
Our research findings have several implications for dealmakers:
- Build integration muscle before making platform deals. The window in which capability buying is competitively sufficient has not yet closed, but it is narrowing. The acquirers that are likely to hold durable positions are the minority now assembling platforms. The operational consequence is immediate because platform and consolidation transactions demand what capability deals do not—synergy planning and synergy capture, rigorous integration management, and post-merger discipline. Many ecosystem acquirers have never run a program of that kind, and building the required capabilities takes longer than signing a deal. The wise course is to build them before you need them.
- Expect price discipline and speed to remain in conflict, with the market rewarding speed. Assets standing at the chokepoints between models and the workflows that use them typically reprice in months, not years. For example, OpenRouter was valued at $1.3 billion in a May 2026 funding round and agreed to be acquired for more than $7 billion just a few months later. These assets are exceptions, not the main pattern, but they set the clearing price for anything in a similar position. Exercising conventional valuation discipline might price you out of precisely the assets that platform assembly requires. Overall, the environment rewards boldness but punishes lack of discipline. Rigorous due diligence and long-term thinking, supported by scenario analysis, are more important than ever.
- Treat alternative deal structures as a live structuring option rather than a reporting anomaly. A structured deal will move faster than an acquisition if it entails a substantial payment for a nonexclusive technology license, offers to the target’s engineering team, and a minority stake in what remains. A deal of this kind also delivers no control, no consolidated revenue, and no ownership of the asset's direction. Boards should evaluate the proposed transaction on its merits and recognize when a competitor is using such structures to take a capability off the market without announcing a deal. As such arrangements proliferate, a growing share of consequential transactions will lie outside both deal databases and merger review, creating a measurement problem for analysts and a policy question for regulators.
- Watch the boundary rather than the sector, and watch it from both sides. A defensive posture that focuses on same-segment rivals will miss an acquirer arriving from an adjacent segment—or from outside the ecosystem altogether, as when Workday paid $1.1 billion in cash for Sana in 2025 to add AI-native enterprise knowledge tools to its array of capabilities. Companies that are more likely to be targets than buyers should emphasize preparation, not defense: identify which segment you are most valuable to, and know what you are worth to a buyer seeking to control distribution or a chokepoint rather than to capture earnings.
The structure of the AI industry is still forming through deals, rather than through consolidation, but this stage will not last indefinitely. The acquirers that decide now which part of the ecosystem they intend to own—and build the M&A and integration capabilities necessary to make those acquisitions work—will shape their position before the market structure hardens around them.
The authors are grateful to Greg Emerson, Val Elbert, Francesca Pietrogrande, and Duc Loc Nguyen for their valuable insights and support in the preparation of this article.
