AI Infrastructure M&A: what is happening now?

What are the latest AI Infrastructure M&A deals?

When we look at all the M&A deals in AI Infrastructure over the last 24 months, the market is clearly not buying “AI startups” in a vague way.

The buyers are going after the layers that make AI cheaper, faster, safer or easier to deploy: compute, data centers, optical interconnect, AI runtimes, model deployment, MLOps, data governance and cloud security.

Date	Target	Acquirer	Value	Strategic rationale	Status and additional comment
Jun. 25, 2026	Synaptics	onsemi	~$7.0B EV	onsemi is buying edge AI compute, connectivity and human-machine-interface assets to move deeper into Physical AI infrastructure	Announced, expected to close mid-2027. Important because this brings AI infrastructure closer to factories, cars, robotics and devices
Jun. 25, 2026	Seldon AI	TrueFoundry	Undisclosed	TrueFoundry is buying model-serving and MLOps infrastructure for enterprise AI deployment	Announced. Smaller deal, but very relevant because model deployment is still consolidating
Jun. 24, 2026	Modular	Qualcomm	~$3.9B all-stock	Qualcomm is buying an AI software stack that helps workloads run across different hardware	Announced, expected to close H2 2026. This is a major bet against hardware lock-in
Mar. 11, 2026	Wiz	Google	$32.0B cash	Google is buying cloud and AI security infrastructure to strengthen Google Cloud	Closed. This is the largest deal in the dataset and Google’s largest acquisition ever
Feb. 17, 2026	Koyeb	Mistral AI	Undisclosed	Mistral is buying serverless AI cloud deployment infrastructure	Announced. Reported as Mistral’s first acquisition
Jan. 2026	XConn Technologies	Marvell	~$540M	Marvell is buying CXL and PCIe switching technology for AI data centers	Announced, expected to close in early 2026. Relevant because memory and accelerator connectivity are becoming AI cluster bottlenecks
Dec. 3, 2025	neptune.ai	OpenAI	Undisclosed	OpenAI is buying experiment tracking and model-training observability infrastructure	Definitive agreement announced. This is infrastructure for frontier model research, not an application deal
Dec. 2, 2025	Celestial AI	Marvell	$3.25B upfront, up to $5.5B with earnout	Marvell is buying optical interconnect technology for next-generation AI data centers	Announced. One of the clearest signs that AI cluster networking is now a strategic M&A category
Oct. 21, 2025	Securiti AI	Veeam	$1.725B	Veeam is buying data security, governance and AI trust infrastructure	Closed. Shows that backup and recovery platforms now want to own governed AI-ready data
Oct. 6, 2025	Monolith AI	CoreWeave	Undisclosed	CoreWeave is buying industrial AI and simulation tooling	Announced. Expands CoreWeave from generic GPU cloud into industrial AI workloads
Sep. 4, 2025	Hexagon Design & Engineering business	Cadence	~$3.16B	Cadence is buying simulation and system-design infrastructure for Physical AI workflows	Closed Feb. 2026. Relevant because AI infrastructure increasingly includes design and simulation tooling
Sep. 3, 2025	OpenPipe	CoreWeave	Undisclosed	CoreWeave is buying reinforcement learning and fine-tuning infrastructure	Announced. Fits the same stack-building move as Weights & Biases
Sep. 2025	Tecton	Databricks	Undisclosed	Databricks is buying real-time feature and data-serving infrastructure for AI agents	Announced. Important because agents need fresh operational data, not only static datasets
Jul. 7, 2025	Core Scientific	CoreWeave	~$9.0B all-stock	CoreWeave is buying data-center capacity and power infrastructure	Announced, expected to close Q4 2025. This is one of the strongest signs that physical capacity is now AI infrastructure
Jun. 9, 2025	Alphawave Semi	Qualcomm	~$2.4B EV	Qualcomm is buying high-speed connectivity silicon for AI data centers	Completed later in 2025. Helps Qualcomm expand from mobile and edge into data-center AI
Jun. 6, 2025	Untether AI engineering team	AMD	Undisclosed	AMD is buying AI inference chip talent	Talent acquisition. Not a full-company deal, but still relevant because inference talent is scarce
May 27, 2025	Informatica	Salesforce	~$8.0B equity value	Salesforce is buying enterprise data governance and integration infrastructure for agentic AI	Closed Nov. 2025. One of the clearest “AI needs governed data” transactions
May 14, 2025	Neon	Databricks	Reported ~$1.0B	Databricks is buying serverless Postgres infrastructure used by AI-generated apps	Announced. Moves Databricks closer to operational databases
May 2025	Brium	AMD	Undisclosed	AMD is buying compiler and AI inference optimization talent	Closed. Strengthens AMD’s AI software layer around its chips
Apr. 28, 2025	Protect AI	Palo Alto Networks	Undisclosed, reported up to ~$700M	Palo Alto is buying AI and ML application security infrastructure	Closed Jul. 2025. Direct evidence that AI systems now need a dedicated security layer
Mar. 19, 2025	Ampere Computing	SoftBank	$6.5B cash	SoftBank is buying Arm-based data-center CPU infrastructure	Closed Nov. 2025. Fits SoftBank’s broader AI infrastructure strategy
Mar. 10, 2025	Edge Impulse	Qualcomm	Undisclosed	Qualcomm is buying edge AI developer infrastructure	Announced. Connects Qualcomm hardware with AI developer workflows
Mar. 4, 2025	Weights & Biases	CoreWeave	Reported ~$1.4B to ~$1.7B	CoreWeave is buying MLOps infrastructure for experiment tracking and model development	Closed May 2025. This was the start of CoreWeave’s visible move from GPU supplier to AI cloud platform
Feb. 25, 2025	DataStax	IBM	Undisclosed	IBM is buying NoSQL, vector database and Langflow infrastructure for enterprise generative AI	Closed May 2025. Strengthens IBM watsonx around AI data infrastructure
Feb. 27, 2025	HashiCorp	IBM	~$6.4B	IBM is buying cloud infrastructure automation and security software	Closed. Relevant because enterprise AI still needs automated hybrid-cloud infrastructure underneath
Dec. 30, 2024	Run:ai	Nvidia	Reported ~$700M	Nvidia is buying GPU orchestration and AI workload-management software	Closed. One of the most direct AI infrastructure deals in the dataset
Sep. 25, 2024	OctoAI	Nvidia	Reported ~$165M to $250M+	Nvidia is buying AI inference and model deployment infrastructure	Closed. Strengthens Nvidia beyond chips, into model-serving software
Sep. 5, 2024	Own Company	Salesforce	~$1.9B cash	Salesforce is buying data protection and backup infrastructure	Closed Nov. 2024. Relevant because enterprise AI requires protected and recoverable data
Aug. 2024	Robust Intelligence	Cisco	Undisclosed	Cisco is buying AI security infrastructure	Closed Oct. 2024. The asset later became core to Cisco AI Defense
Aug. 2024	ZT Systems	AMD	~$4.9B	AMD is buying hyperscale AI server and rack-level infrastructure expertise	Closed Mar. 2025. Key move for AMD’s AI data-center strategy
Jul. 17, 2024	Brev.dev	Nvidia	Undisclosed	Nvidia is buying GPU cloud developer infrastructure	Closed. Helps developers access and use GPU compute more easily
Jul. 10, 2024	Silo AI	AMD	~$665M cash	AMD is buying enterprise AI software and model-building expertise	Closed Aug. 2024. Helps AMD sell fuller AI solutions, not only chips

Is AI Infrastructure M&A still active now?

Yes, AI Infrastructure M&A is still active now, but the activity has become more selective and more strategic.

When we look at all the AI Infrastructure M&A deals over the last 24 months, we count 32 deals. The previous 12-month period had 18 deals, while the most recent 12-month period had 14 deals. That is a decline of 4 deals, or about 22%.

The first read is that volume slowed. The better read is that the market moved from many software and tooling acquisitions toward fewer assets that control bigger bottlenecks. The recent period includes Google-Wiz, CoreWeave-Core Scientific, Qualcomm-Modular, onsemi-Synaptics, Marvell-Celestial AI and Cadence-Hexagon Design & Engineering. Those deals are not small capability patches. They touch cloud security, data-center capacity, AI software portability, edge AI hardware, optical interconnect and engineering simulation.

So, it looks like the market is becoming choosier. Buyers are doing fewer deals, but the deals they do are closer to the real constraints of AI deployment.

Is AI Infrastructure M&A getting bigger recently?

Yes, AI Infrastructure M&A is getting bigger recently in terms of strategic size, even if deal count has fallen. The most recent 12 months include at least six very large disclosed transactions: Google-Wiz at $32B, CoreWeave-Core Scientific at about $9B, onsemi-Synaptics at about $7B, Qualcomm-Modular at about $3.9B, Marvell-Celestial AI at $3.25B upfront and Cadence-Hexagon Design & Engineering at about $3.16B.

That mix matters because these are different kinds of infrastructure control points. Wiz gives Google a major cloud and AI security platform. Core Scientific gives CoreWeave power and data-center footprint. Modular gives Qualcomm a software layer that can reduce friction across hardware. Celestial AI gives Marvell optical interconnect for AI clusters. Synaptics gives onsemi edge AI compute and connectivity.

The strong conclusion is that AI Infrastructure M&A is not getting bigger because everyone is buying more. It is getting bigger because the market has identified a few layers where scarcity is obvious and delay is expensive. When a buyer sees a control point in security, power, networking or hardware portability, the check size can jump quickly.

If you want more recent data on this point, please see our latest AI infrastructure market report.

Are AI Infrastructure buyers still mostly buying software these days?

No, AI Infrastructure buyers are no longer mostly buying pure software. Software still matters, but the center of gravity has moved closer to physical, semiconductor and systems infrastructure. In the earlier part of the 24-month window, we saw many software-heavy deals: Silo AI, Brev.dev, OctoAI, Robust Intelligence, Run:ai, DataStax, Weights & Biases and Protect AI.

The recent period looks different. CoreWeave bought Core Scientific for data-center and power capacity. Marvell bought Celestial AI and XConn for AI data-center connectivity. Qualcomm bought Alphawave and Modular to combine connectivity silicon with AI software portability. onsemi announced Synaptics to move into edge AI hardware. Cadence bought Hexagon’s design and engineering business to strengthen Physical AI and simulation workflows.

The important shift is that AI Infrastructure M&A is moving from software around the model toward the layers that determine whether AI can run economically at scale. That includes power, racks, interconnect, runtime portability, simulation and edge compute. Software is still part of the story, but it is increasingly being bought to complete a hardware or platform strategy.

Is GPU cloud becoming a real consolidation market now?

Yes, GPU cloud is becoming a real consolidation market now, and CoreWeave is the clearest case. Over the period, CoreWeave bought Weights & Biases, OpenPipe, Monolith AI and Core Scientific. That gives us four different signals from the same buyer: MLOps, reinforcement learning, industrial AI and physical data-center capacity.

This is a stronger pattern than just “CoreWeave is acquisitive.” Weights & Biases helps developers track experiments and train models. OpenPipe supports fine-tuning and reinforcement learning workflows. Monolith brings AI into physical engineering and simulation-heavy industries. Core Scientific brings power and data-center capacity, which are some of the hardest resources to secure in AI infrastructure.

The conclusion is pretty direct: GPU cloud providers do not want to remain commodity GPU landlords. If the only product is rented compute, differentiation becomes fragile. CoreWeave is trying to own more of the workflow above the GPU and more of the capacity below it. That is what makes this consolidation important.

If you want more recent data on this point, please see our latest AI infrastructure market report.

Are chip companies buying AI Infrastructure again?

Yes, chip companies are buying AI Infrastructure again, but they are buying around the chip as much as the chip itself. AMD bought Silo AI, ZT Systems, Brium and the Untether AI engineering team. Nvidia bought Brev.dev, OctoAI and Run:ai. Qualcomm bought Edge Impulse, Alphawave and Modular. Marvell bought Celestial AI and XConn. onsemi announced Synaptics.

Those deals cover very different layers: enterprise AI software, rack-scale AI servers, compiler optimization, inference talent, GPU orchestration, model serving, edge AI tools, high-speed connectivity, optical interconnect, CXL switching and edge AI hardware. The common thread is that silicon companies are trying to solve the full deployment problem, not only produce faster chips.

This is one of the strongest non-obvious reads from the table. AI chip competition is no longer only about performance benchmarks. It is about whether the buyer can deliver the surrounding system: software, servers, networking, developer tools and workload management. That is why AMD buying ZT Systems and Qualcomm buying Modular may matter as much strategically as a pure chip acquisition.

Is AI Infrastructure M&A mostly about data centers now?

No, AI Infrastructure M&A is broader than data centers, but data-center capacity has become one of the most valuable new M&A layers. CoreWeave-Core Scientific is the obvious signal, but it is not alone. AMD-ZT Systems points to rack-scale server design. Marvell-Celestial AI and Marvell-XConn point to the connectivity bottleneck inside AI data centers. Qualcomm-Alphawave also sits in the same high-speed data-center connectivity theme.

That means data centers are not just buildings in this market. They are a bundle of constraints: power, server architecture, networking, memory access, optical interconnect and upgrade speed. Buying Core Scientific helps CoreWeave with power and footprint. Buying ZT helps AMD with rack-level AI systems. Buying Celestial AI helps Marvell attack bandwidth and latency inside AI clusters.

So yes, data centers matter a lot. But the smarter point is that the data center itself is becoming an M&A stack: power below, silicon inside, interconnect between machines, orchestration above, and software that makes the whole thing usable.

If you want more recent data on this point, please see our latest AI infrastructure market report.

Are AI security deals becoming part of AI Infrastructure M&A?

Yes, AI security has become part of AI Infrastructure M&A. The proof is not just Google buying Wiz. We also see Cisco-Robust Intelligence, Palo Alto-Protect AI, Veeam-Securiti AI, Salesforce-Own and Salesforce-Informatica. These deals sit across cloud security, AI application security, data security, backup, governance and trust.

The reason is simple: enterprise AI creates new failure points. Models can leak data, use untrusted inputs, access sensitive systems, generate risky actions and depend on messy internal data. A normal cloud stack was not designed for all of that. This is why the security layer is moving closer to the AI deployment layer.

Google-Wiz is the largest proof point, but the more interesting detail is the variety of buyers. Cisco, Palo Alto, Veeam, Salesforce and Google are not all solving the same problem. Yet they are all buying around AI trust, governed data or secure deployment. That makes AI security infrastructure a real M&A category, not a marketing label.

Are enterprise data platforms buying AI Infrastructure now?

Yes, enterprise data platforms are buying AI Infrastructure now because AI systems need usable data more than they need another demo. Salesforce bought Own and Informatica. IBM bought DataStax. Databricks bought Neon and Tecton. Veeam bought Securiti AI. These deals cover backup, data integration, governance, vector and NoSQL databases, operational databases, real-time feature serving and data security.

The important thing is how these pieces fit together. DataStax helps with AI data and vector-style workloads. Neon gives Databricks a serverless operational database layer. Tecton adds real-time data serving for AI agents. Informatica gives Salesforce governance and integration. Securiti AI gives Veeam a data trust and security layer.

AI Infrastructure buyers are trying to close the gap between enterprise data and AI execution. A model is only useful if it can safely reach the right internal data, update its context and operate inside governance rules. That is why data infrastructure has become one of the main M&A lanes.

If you want more recent data on this point, please see our latest AI infrastructure market report.

Are AI Infrastructure acquirers buying complete companies or just talent?

AI Infrastructure acquirers are mostly buying complete companies, but talent deals appear when the missing capability is extremely specialized. Most of the large transactions are full acquisitions: Wiz, Core Scientific, Modular, Synaptics, Celestial AI, Informatica, ZT Systems, HashiCorp, Ampere, DataStax and Weights & Biases. These are not acqui-hires but actual platform, product or asset purchases.

Still, the talent signal matters. AMD’s Untether AI transaction was essentially an engineering-team move around inference chip expertise. Brium also looks like a capability acquisition around compilers and inference optimization. Meta’s hiring of Virtue AI founders, while not a clean M&A deal in this table, supports the broader point that AI security and infrastructure talent can be strategic even without a full-company acquisition.

The practical interpretation is that buyers acquire companies when the asset has product maturity, customer footprint or infrastructure ownership. They go after teams when the value is concentrated in scarce engineering knowledge, especially around inference, compilers, AI security or chip-adjacent software.

Are first-time acquisitions happening in AI Infrastructure now?

Yes, first-time acquisitions are happening in AI Infrastructure now, but they are not the core driver of the market. Mistral buying Koyeb is the clearest example because it was reported as Mistral’s first acquisition.

The logic is straightforward: Mistral cannot only be a model company if it wants to compete with bigger AI platforms. It needs deployment and cloud infrastructure around its models.

CoreWeave’s Weights & Biases acquisition also matters because it was an early major step in its public platform strategy. After that, CoreWeave kept buying: OpenPipe, Monolith and Core Scientific. That sequence tells us the first deal was not a one-off experiment. It was the start of a broader stack-building plan.

Still, most AI Infrastructure M&A is being driven by repeat acquirers: AMD, Nvidia, IBM, Salesforce, Qualcomm, Google, Cisco, Palo Alto, Marvell and Databricks. The market is therefore led by companies that already know what they are missing and are using M&A to close those gaps faster.

Is AI Infrastructure M&A consolidating around a few repeat buyers?

Yes, AI Infrastructure M&A is consolidating around repeat buyers, and that is one of the strongest patterns in the table. AMD appears several times, CoreWeave appears several times, Qualcomm appears several times, Marvell appears several times, Databricks appears several times and Salesforce appears several times.

That matters because repeat acquisition behavior usually reveals a roadmap. AMD is building around full-stack data-center AI execution: software, servers, compilers and inference talent. CoreWeave is building around AI cloud vertical integration: developer workflow, reinforcement learning, industrial AI and data-center capacity. Qualcomm is moving from edge and mobile into data-center AI through Edge Impulse, Alphawave and Modular. Marvell is doubling down on AI connectivity through Celestial AI and XConn.

The market therefore looks less like a scattered land grab and more like a stack-control race. The buyers that come back again and again are not just collecting assets. They are assembling missing layers.

If you want more recent data on this point, please see our latest AI infrastructure market report.

Are valuations in AI Infrastructure M&A still aggressive?

Yes, valuations in AI Infrastructure M&A are still aggressive when the target controls a scarce layer. The largest disclosed deal is Google-Wiz at $32B. The table also includes CoreWeave-Core Scientific at about $9B, Salesforce-Informatica at about $8B, onsemi-Synaptics at about $7B, SoftBank-Ampere at $6.5B, IBM-HashiCorp at $6.4B, AMD-ZT Systems at $4.9B and Qualcomm-Modular at about $3.9B.

Those are not all the same type of asset, which is exactly the point. Buyers are paying large amounts for cloud security, power and data-center capacity, enterprise data governance, edge AI hardware, data-center CPUs, hybrid-cloud automation, AI server systems and AI software portability. The common denominator is not “AI branding.” It is scarcity plus strategic fit.

The market is therefore expensive in a very specific way. A generic AI wrapper does not deserve a major multiple. But a target that gives the buyer control over security, power, connectivity, data governance or hardware portability can command a large price because it reduces execution risk across a much bigger platform.

Is AI Infrastructure M&A moving toward the edge again?

Yes, AI Infrastructure M&A is moving toward the edge again, but not in the old IoT sense. The recent edge-related deals are more serious because they connect edge devices, AI compute, connectivity and physical-world use cases. Qualcomm bought Edge Impulse for edge AI developer tooling. Qualcomm also bought Modular to improve AI software portability across data-center and edge environments. onsemi announced Synaptics to move deeper into edge AI hardware, connectivity and human-machine interfaces.

That combination tells us the edge AI thesis is changing. The old version was mostly “put small models on devices.” The new version is broader: make AI systems run across cloud, data center, car, factory, robot and device environments without rebuilding everything from scratch.

This is why the Qualcomm-Modular and onsemi-Synaptics deals matter together. One is about software portability. The other is about edge AI hardware and connectivity. Together, they suggest that AI Infrastructure M&A is preparing for AI workloads to spread outside centralized GPU clusters.

So, what is the latest update on AI Infrastructure M&A now?

AI Infrastructure M&A is changing shape more than it is slowing down. The deal count fell from 18 deals in the previous 12 months to 14 deals in the most recent 12 months, which is a 22% decline. On volume alone, that looks like a slowdown.

But the composition says something different. In the recent period, we see fewer small software deals and more large strategic control-point deals: Wiz for cloud and AI security, Core Scientific for power and data centers, Celestial AI for optical interconnect, Modular for AI software portability, Synaptics for edge AI hardware and XConn for CXL and PCIe switching.

AI Infrastructure M&A is maturing. The easy first wave was buying obvious software gaps. The current wave is about infrastructure layers that are harder to build internally and harder to replace once competitors control them.

Check	Current status
Deal activity now	AI Infrastructure M&A remains active, with 32 deals over the last 24 months. The latest 12 months show fewer deals than the previous 12 months, but the recent deals are more strategic and more infrastructure-heavy
12-month momentum	Deal count declined from 18 to 14, a drop of about 22%. That is real volume deceleration, but not a collapse, because recent deals include several multi-billion-dollar platform moves
Deal size	The market still supports very large checks. Wiz, Core Scientific, Synaptics, Informatica, Ampere, HashiCorp, ZT Systems, Modular and Celestial AI all show that strategic AI infrastructure assets can still clear multi-billion-dollar prices
Main buyer behavior	Repeat buyers dominate the market. AMD, CoreWeave, Qualcomm, Marvell, Databricks and Salesforce are not buying randomly; each one is filling visible gaps in its AI infrastructure stack
Compute and hardware	Hardware has moved back to the center of the market. Ampere, ZT Systems, Alphawave, Celestial AI, XConn, Modular and Synaptics show that AI infrastructure is now deeply tied to chips, servers, connectivity and edge devices
Data-center capacity	Physical capacity is now a direct acquisition target. CoreWeave-Core Scientific shows that power, footprint and lease control can be as strategic as software in AI infrastructure
MLOps and deployment	MLOps is still consolidating, but it is now being absorbed into larger AI cloud and model platforms. Weights & Biases, OpenPipe, neptune.ai, Seldon AI, Run:ai and OctoAI all support that pattern
Data foundation	Data infrastructure is one of the main enterprise AI M&A lanes. Salesforce-Informatica, IBM-DataStax, Databricks-Neon and Databricks-Tecton show that buyers need governed, live and usable data for AI systems
Security and governance	AI security has become a core infrastructure layer. Wiz, Protect AI, Robust Intelligence and Securiti AI show that enterprise AI needs secure deployment, governed data and model protection
Edge AI	Edge AI is reappearing in a more credible form. Qualcomm-Edge Impulse, Qualcomm-Modular and onsemi-Synaptics suggest that AI infrastructure is preparing to move across data centers, devices, factories, cars and robots
Valuation signal	The market is not paying huge prices for every AI asset. The largest valuations are attached to scarce control points: cloud security, power capacity, data governance, hardware/software portability, AI connectivity and edge AI platforms

Who is the author of this content?

NEW MARKET PITCH TEAM

We track new markets so founders and investors can move faster

We build living "market pitch" documents for emerging markets: AI, synthetic biology, new proteins, and more. Instead of outdated PDFs or hallucinated LLM answers, our clients get a clean, visual, always-updated view of what's really happening: key players, deals, regulations, and signals that matter. Learn more about us.