Ten acquisitions and more than $13 billion in fresh funding point the same way: the value in AI is sliding below the model, toward the infrastructure that makes it cheap enough and safe enough to run.
In June 2026, SpaceX paid a reported $60 billion for a code editor. A month earlier, that sentence would have looked like a typo.
Anysphere, the company behind the AI coding tool Cursor, went from a $29 billion valuation to a $60 billion all-stock exit in roughly half a year. It was the loudest deal in a noisy month. It was not the most revealing one.
Look past the two headline numbers and a clearer pattern runs through the 48 deals and funding rounds we logged in our data, analytics, and AI market tracker last month. Buyers and investors mostly weren’t chasing the next foundation model. They were paying for the layers around the model: the software that makes inference cheaper, the controls that make AI agents safe to put into production, and the infrastructure that lets an organization run AI on its own terms.
That’s the throughline for June. The value in AI is drifting away from who trains the smartest model and toward who owns the layer that makes models cheap, governable, and sovereign to run.
Here’s how the month broke down, and where each cluster connects to questions our research keeps returning to.
June 2026 AI deals by the numbers: 48 deals and $13B in funding
A quick tally of what we logged last month, and what the raw numbers do and don’t tell us.
The overall count
48 events in total: 10 acquisitions, one SPAC listing (a stock-market debut through a ready-made shell company rather than a traditional IPO), 36 funding rounds, and one reported megaround still in the works.
Where the money went
Disclosed funding topped $13 billion, but a single $12 billion round for Jeff Bezos’s stealth industrial-AI company Prometheus accounts for most of it.
Strip Prometheus out and the rest adds up to roughly $1.5 billion, spread across a long tail of seed and Series A checks, many of them European.
The rest of the acquisitions went undisclosed, which keeps the full scale of the month’s consolidation out of view.
Everything below is grouped by the themes that shaped the month. A few deals sit in more than one place, and we flag those where it matters.
Compute and inference: making AI cheaper to run
No theme was more about cost than inference. In our conversations with data and analytics leaders, the hard part of AI in production is rarely the model. It’s the bill, so the companies that make inference cheaper are becoming strategic.
The approaches differ, though. At the serving layer, Qualcomm bought Modular for a hardware-agnostic runtime that frees models from any single chip, and Baseten’s reported megaround backs the same deploy-and-scale idea.
A step lower, Engram, Sail Research, and Ora Computing target the efficiency of inference itself, leaner runtimes and cheaper tokens per query. PointFive comes at the bill from the opposite end, on the FinOps side, hunting waste in the cloud spend organizations already carry.
AI governance, security, and compliance: keeping AI and agents in check
This was one of the two busiest clusters of the month, level with the last-mile group at 11 deals each, but it isn’t a single story. Part of it is squarely about agents: as they move from demos into production, someone has to keep them in check.
Patronus AI evaluates model and agent behavior, NeuralTrust and Archestra.AI secure agentic use cases, and Coalex.ai’s “decision firewall” and Bayshore’s compliance-as-code aim to constrain what AI does before it reaches a customer or a ledger.
The rest is broader security and compliance that would matter with or without agents: Nebulock’s threat hunting, Entryzero’s AI-driven threat-exposure management (mapping attack surfaces from the hacker’s perspective, with DORA and NIS2 compliance built in), Flagright and Kalipso on regulatory compliance, Databricks’ acquisition of Panther for security operations, and Wultra’s post-quantum digital identity, a hedge against future cryptographic attacks that has little to do with agents at all.
What ties the cluster together is less the technology than the direction of travel: regulation isn’t slowing this market down; it’s funding it, as buyers look for systems that can be audited, constrained, and explained.
AI observability: monitoring rebuilt for the AI era
Monitoring is having its own moment: classic observability is being redesigned for systems whose outputs are probabilistic rather than deterministic. Coralogix and Tsuga lead here, and Patronus (see Control) overlaps too, since evaluating model behavior is observability by another name.
Data sovereignty: from policy debate to a product you can buy
Data sovereignty became a product category. It moved from a policy conversation into a line item you can buy. 1001 and Spain’s Substrate AI both raised for sovereign AI infrastructure aligned with the EU AI Act and GDPR, and Modular’s vendor-neutral stack (see Compute) is a hedge against depending on any single chip supplier.
SAP ECC 2027 deadline: funding the S/4HANA migration race
Two rounds were really about the 2027 deadline for SAP ECC. Qorelo, a Berlin startup barely five months old, raised $3.5 million to automate ECC-to-S/4HANA migrations. Conduct, founded by former Palantir staff, raised $60 million to help enterprises operate and modernize legacy ERP, with SAP itself joining as a strategic investor.
For anyone following our SAP coverage, the message is familiar: the migration wall is close, and capital is flowing to whoever can make the move faster.
Agentic coding: build your own software or buy it?
Coding is turning into one of the strongest use cases for AI, which is exactly why it drew a $60 billion acquisition and why it sharpens the question our audience keeps asking: build your own software or buy standard packages?
SpaceX bought Cursor and OpenAI absorbed Ona (ex-Gitpod), while Codeplain and SuperPlane raised early money to push agentic software engineering further.
Strip out the headline themes and a large, quiet cluster remains, and for our readers it may be the most important one. These are the companies working the last mile: turning models and data into decisions and actions inside real enterprises.
Some sell the decisioning itself: Taktile for banks and insurers, ContextFab on the factory floor. Some are the agents that do the front-line work: Fin and Gradient Labs in customer service, Mobilewalla’s vertical agents in consumer intelligence. Some close the adoption and orchestration gap: Mendo on getting employees to actually use GenAI, INXM on making agents execute reliably. And some do the unglamorous plumbing that feeds them: Crux and NextGen Invent operationalizing data, Nagarro supplying the delivery muscle, Contentful handing Salesforce’s agents a content layer to act on.
No single deal defined the month here, but together they mark where the value is landing, not in the model but in making it do the work.
A smaller but telling group is rebuilding an established software category from the ground up for the AI era, often led by veterans of the very incumbent they mean to replace. PhoenixAI (formerly CelerData) rebuilt the analytical database so that autonomous agents, not human analysts, are the primary users.
Golden Analytics, founded by Tableau’s former product chief, is doing the same to BI: analytics designed for the AI era rather than retrofitted to it. And Zaro, built by engineers who helped ship Salesforce’s Agentforce, is deliberately its “anti-vendor” opposite: one AI workspace the customer owns rather than the software vendor.
It is a small sample, but a signal worth watching: the shift is deep enough to reopen categories that looked settled.
The rest of the field: deals at the edge of the data and analytics stack
The month’s two biggest checks matter least to us: Prometheus’s $12 billion physical-AI bet and SandboxAQ’s $500 million AI-plus-quantum play are striking in scale but far from the D&A stack.
Isometric’s carbon-certification round is the one we are quietly rooting for, a hint that ESG plus AI agents could regain momentum, while the vertical plays are useful mainly as evidence of how far agentic AI is spreading into niches.
Who is buying AI companies, and why: Salesforce, Databricks, Qualcomm, SpaceX
The acquirer list matters less as a roll-call than for the intent behind a few of the moves. Salesforce bought twice, Contentful and Fin, which reads as Marc Benioff doubling down on his agent-driven CRM vision: own the content layer and the AI customer-service layer, not just the database underneath.
Databricks’ purchase of Panther, its third security deal in a row, signals that data platforms now treat security and governance as part of the core stack, not an add-on.
And the most telling buyers came from outside software altogether: Qualcomm and SpaceX are betting that owning AI compute and tooling is strategic even if your business is chips or rockets, a sign of how far this value has spread beyond the traditional software industry.
What to watch next: agentic engineering, sovereignty, and maturity over hype.
June’s shape was a barbell. Two enormous deals at one end, a crowd of small European seed and Series A rounds at the other, and comparatively little in the middle. That tells us capital is concentrating at the top while early-stage experimentation stays healthy, especially across Germany, France, Spain, Austria, Slovenia, and the Nordics.
We wouldn’t read the smaller European checks as a weakness. The US market is louder and hype travels faster there, so Europe’s more modest, less frothy rounds may prove the more durable bet, backing companies built on a sober business case rather than a narrative and carrying less of the correction risk that comes with froth.
There is a geopolitical read here too. Unease about American dominance of the IT and data stack is no longer a European peculiarity; it is spreading, and the Gulf is the clearest new example, with 1001 raising regional sovereign money to build sovereign AI operating systems for critical infrastructure across the GCC.
The irony is that Washington has done as much as Brussels to fuel this. In February the Trump administration ordered US diplomats to lobby foreign governments against data-sovereignty rules, framing them as barriers to US technology exports. Then in June came the Fable 5 shutdown: a US export-control order forced Anthropic to suspend access to Claude Fable 5, at the time its most capable model, for all foreign nationals, taking the model dark for customers outside the US within hours. What had been an abstract dependency suddenly felt very concrete. Both moves have handed the sovereignty cause fresh momentum rather than slowing it.
Two themes will shape how we read the rest of the year, and beyond them a set of open questions we will keep tracking.
Agentic engineering is coming for software itself. The Cursor, Ona, Codeplain and SuperPlane deals are early markers of a deeper shift: agents that write, test, and ship code are becoming the future of software development, not just a productivity add-on. That is a direct threat to established software vendors, whose per-seat licenses assume humans do the work and whose customers are starting to ask why they should buy a packaged product an agent could now build for them. And it is not only the coding agents: the same logic drives the AI-native rebuilds that reconstruct the database, BI, and workspace from scratch. When even settled categories reopen, the build-versus-buy question turns existential for incumbents.
Maturity is quietly beating hype. Agentic AI owns the headlines, but the money flowing into governance, monitoring, and cost control is the more telling signal. The same holds on the last mile: capital is flowing into decision automation, adoption, and agent-ready data, toward where AI does the work, not where it’s demoed. It suggests the technology is maturing fast. The shift is from experimentation to disciplined, production-grade use.
Open questions remain
Will the bet on the compute and inference layer pay off, or will hyperscalers absorb that value themselves? And will this month’s efficiency gains (cheaper inference, leaner models, and smarter serving) keep energy and compute from becoming the scarce resource many now expect?
As the EU AI Act’s obligations for general-purpose and high-risk systems phase in through 2026 and 2027, will the governance, compliance, and sovereignty spend harden into a permanent budget line, or fade once companies have done just enough to tick the box?
How many of these small European rounds become the next acquisition targets for the same handful of consolidators?
Who ends up owning the application and decision layer: the consolidating platforms like Salesforce and Databricks, or the AI-native challengers now rebuilding their categories?
We will keep logging the answers.
Data, BI and Analytics Trend Monitor
The new Data, BI and Analytics Trend Monitor shows you which trends are shaping the BI, analytics and data management market. Find out now which trends are really worth investing in!
Alexander Seeliger is an Analyst for Data & Analytics and Data Scientist at the Business Application Research Center (BARC). He advises companies on use case identification for data analysis and tool selection for advanced analytics. He conducts proof of concepts in the field of advanced analytics and provides data science and data literacy coaching. Alexander Seeliger is the author of BARC market studies and research articles. He speaks at conferences and conducts BARC and in-house seminars. He is mainly responsible for data management, data preparation and data enrichment of the BARC product and service overviews.
DATA festival Online. Experience Data & AI with the community shaping what’s next.
