Microsoft’s Agentic AI Push Wins KPMG and Atos as Investors Balk

Microsoft signed KPMG and Atos onto its Agent 365 platform this month and rolled out new Azure chips and models at Build 2026. Its AI business is now growing 123% year over year. Its stock has still lost 16.6% over the past six months.

Enterprise demand for agentic AI looks genuine. Two more workforces, each tens of thousands of employees, committed to Microsoft’s agent platform in a single week in June. But the same expansion is pushing capital spending toward $190 billion, compressing margins, and exposing security holes Microsoft’s own engineers had to patch in the framework running the agents.

KPMG and Atos Bet Big on Agent 365

Microsoft and professional services giant KPMG expanded their global alliance on June 9, 2026. KPMG will use Microsoft Agent 365 to extend its own Trusted AI framework and roll out Microsoft 365 Copilot across a workforce of more than 276,000 professionals. Microsoft now classifies KPMG as a “Frontier Firm,” the label it uses for customers furthest along in deploying agents at enterprise scale.

Atos Group made a similar move the same day. The French IT services company is deploying Microsoft 365 Copilot E7 to all 56,000 of its employees across 54 countries, while managing a 19,000-agent fleet through Agent 365, Microsoft’s control plane for governing agents, which reached general availability on May 1.

Both deals landed the same day, a sign Microsoft is treating large, simultaneous workforce commitments as proof points for the platform rather than one-off pilots.

Build 2026 Adds New Chips, Models and Networking to the Stack

The KPMG and Atos news landed in the same week Microsoft used its Build 2026 developer conference to stack new infrastructure underneath all of it.

• Azure Cobalt 200: an early access preview of Arm-based virtual machines delivering up to 50% better generational performance than Cobalt 100, with the chip now deployed in more than 10 global regions.
• Multipath Reliable Connection: an open networking protocol co-developed with AMD, Broadcom, Intel, NVIDIA and OpenAI.
• Azure HorizonDB: a new database service that entered public preview.
• Microsoft Foundry catalog: expanded with public preview access to multiple Anthropic Claude models and the general availability of OpenAI’s GPT-5.5 on June 3, 2026.

Microsoft also gave agents a shared memory layer at Build, a move covered in Microsoft IQ’s shared-memory layer for enterprise agents, while continuing to diversify away from a single model supplier, part of what one analysis called a five-year bet on AI model independence.

The Growth Numbers Behind the Blitz

The deal-and-launch pace lines up with what Microsoft told investors about its fiscal third quarter.

• $82.9 billion in fiscal third-quarter revenue, up 18% year over year, with Microsoft Cloud reaching $54.5 billion, up 29%.
• $37 billion annualized run rate for the AI business, up 123% year over year, as Azure grew 40%.
• 20 million paid Microsoft 365 Copilot seats, up from 15 million the prior quarter, with seat additions accelerating 250% year over year.
• 90% of the Fortune 500 now running agents built through Copilot Studio’s low-code tools, with over 300 customers on track to process more than a trillion tokens on Foundry this fiscal year.

That trajectory tracks a broader forecast. Research firm IDC predicts 1.3 billion agents will exist by 2028, a scale Microsoft is explicitly building Agent 365 to keep manageable before it isn’t.

Gross Margin Slides as Capex Climbs Toward $190 Billion

Microsoft’s own numbers trace a trend across three straight quarters. Company gross margin ran 69% in the fiscal first quarter. It fell to 67.6% by the third quarter, ended March 31, the narrowest since 2022. Microsoft is now guiding to roughly 64% for the fiscal fourth quarter, citing continued AI investment.

Capital spending is the reason. Microsoft told investors on April 29 that capital expenditures for the year will reach $190 billion, up 61% from 2025 and roughly $35 billion above what Wall Street had modeled, with finance chief Amy Hood attributing part of the jump to soaring memory prices. Shares fell nearly 4% the next day to close at $407.78.

The reaction hasn’t faded. Wolfe Research cut its price target on Microsoft to $525 from $570 this week, after raising its fiscal 2027 capital expenditure estimate to $270 billion from $230 billion. Analyst Alex Zukin pointed to component cost inflation flagged in Micron’s latest earnings as the trigger.

• Wolfe Research (Alex Zukin) trimmed its price target, arguing rising memory and component costs will keep squeezing margins through fiscal 2027.
• Zacks Investment Research, the stock-rating firm, gives Microsoft a Value Score of D and a Rank #3 (Hold), citing a forward Price/Sales ratio of 7.91X against the industry’s 6.81X.
• Other Wall Street desks have mostly held Buy and Overweight ratings, betting that a commercial remaining performance obligation near $627 billion justifies the spending.

Why Did Microsoft’s Own Security Team Find Holes in Its Agent Framework?

Microsoft’s security researchers disclosed two critical vulnerabilities in Semantic Kernel, the company’s open-source framework for orchestrating AI agents, in May 2026. The flaws let attackers achieve remote code execution through prompt injection alone, no malware or browser exploit required, in a framework carrying more than 27,000 stars on GitHub.

The bugs, tracked as CVE-2026-25592 and CVE-2026-26030, were disclosed and patched by Microsoft’s own security team. In one proof of concept, the researchers wrote, “a single prompt was enough to launch calc.exe on the device running our AI agent, with no browser exploit, malicious attachment, or memory corruption bug needed.”

The stakes are structural. Agents no longer just generate text. They read files, query databases and run scripts with real credentials, so a flaw in how a framework maps model output to system tools becomes an execution risk, not a content problem.

Microsoft has been warning customers about a related issue for months: unmanaged agents installed directly by employees, outside any IT review, a phenomenon the company calls shadow AI.

We need to treat agents like humans and apply Zero Trust principles.

Vasu Jakkal, corporate vice president at Microsoft Security, made that case in a company report on agent risk. Agent 365 now extends discovery and blocking controls through Microsoft Defender and Intune to catch unmanaged agents on end-user devices, and an analysis from Futurum Group described the approach as a wager that “the harder and more durable problem is controlling them once they exist.” That governance question already reaches beyond agents into Copilot itself, where Copilot’s auto-install default for administrators has drawn its own scrutiny.

Amazon and Google Are Running the Same Playbook

Microsoft is not alone in racing to turn cloud infrastructure into agent infrastructure. Amazon has folded OpenAI’s GPT-5.5, GPT-5.4 and Codex models into Amazon Bedrock and, on May 7, previewed Bedrock AgentCore Payments, built with Coinbase and Stripe to let agents transact on their own. Amazon says its AgentCore SDK has been downloaded more than two million times since preview.

Alphabet moved just as fast. At Google Cloud Next ’26 in April, the company rebranded Vertex AI as the Gemini Enterprise Agent Platform, completing the console-level cutover on May 21. Its Agent Development Kit reached version 1.0 across four programming languages, and its agent-to-agent protocol is now running in production at more than 150 organizations.

The financial contrast is sharper than the product contrast. Google Cloud revenue grew 63% year over year to $20 billion in the same quarter Azure grew 40%, and Alphabet raised its 2026 capital spending guidance to as much as $190 billion, matching Microsoft’s own number. Alphabet’s finance leadership said the results reinforce its conviction to keep investing the capital needed to capture the AI opportunity.

Microsoft Guides to Double-Digit Growth into Fiscal 2027

Microsoft has told investors to expect $86.7 billion to $87.8 billion in fiscal fourth-quarter revenue, with capital expenditures exceeding $40 billion in the quarter alone. The company expects double-digit revenue growth to continue into fiscal 2027, a target that assumes Azure capacity constraints ease as new regions and chips like Cobalt 200 come fully online.

The Zacks Consensus Estimate puts Microsoft’s fiscal 2026 earnings at $17.35 per share, up 27.2% from a year earlier. Microsoft still carries a Zacks Rank #3 (Hold), the same week its agentic AI push produced new enterprise deals, new chips, and a patched hole in the framework running it all.

Frequently Asked Questions

What Is Microsoft Agent 365 and How Much Does It Cost?

Agent 365 is Microsoft’s control plane for observing, governing and securing AI agents across Microsoft and partner platforms, including agents built on AWS or Google Cloud. It reached general availability on May 1, 2026, priced at $15 per user per month, with registry sync to AWS Bedrock and Google Cloud available in public preview.

What Is Agent Sprawl?

Agent sprawl describes the buildup of AI agents operating outside IT oversight, often installed directly by employees using tools such as OpenClaw or Claude Code. Microsoft calls the unmanaged version of this problem shadow AI and built Agent 365’s discovery features, tied to Microsoft Defender and Intune, specifically to surface and block it.

What Were the Vulnerabilities Found in Microsoft’s Agent Framework?

Microsoft’s own security team disclosed two flaws, CVE-2026-25592 and CVE-2026-26030, in Semantic Kernel, its open-source framework for building AI agents. Both allowed remote code execution through prompt injection, and both were fixed before Microsoft published details in May 2026.

How Does Microsoft’s Agentic AI Strategy Compare to Amazon and Google?

Amazon is focused on letting agents transact autonomously through Bedrock AgentCore Payments, built with Coinbase and Stripe, while Alphabet rebranded Vertex AI into the Gemini Enterprise Agent Platform and pushed its agent-to-agent protocol into production at more than 150 organizations. Microsoft leans harder on governance, positioning Agent 365 as the control layer for agents built anywhere, including on rival clouds.

Why Did Microsoft Stock Fall Despite Record AI Growth?

Microsoft shares dropped after the company guided capital expenditures toward $190 billion for the year and gross margin toward roughly 64% for the fiscal fourth quarter, both signs AI infrastructure is costing more than analysts expected. The stock had lost 16.6% over six months even as Azure and Copilot revenue kept accelerating.

Disclaimer: This article is for informational purposes only and does not constitute investment advice. Stock prices, analyst ratings and price targets change quickly; consult a licensed financial advisor before making investment decisions. Figures are accurate as of publication on July 10, 2026.