Claude Managed Agents Sank Three Edge Stocks. Six Months On, Who Partnered

Two days after Anthropic launched Claude Managed Agents on 8 April 2026, Fastly fell 18%, Akamai 13% and Cloudflare 11% in one session. Six months of filings tell a different story from that day. Cloudflare became a Claude Managed Agents partner within six weeks, Akamai disclosed a $1.8 billion contract Bloomberg tied to Anthropic, and Fastly raised guidance twice without partnering at all.

Short answer: Cloudflare is the only one of the three that publicly partnered with Anthropic on Claude Managed Agents, through Environments for Claude Managed Agents on 19 May 2026. Akamai disclosed a $1.8 billion compute contract that Bloomberg attributed to Anthropic. Fastly announced no Anthropic deal and raised guidance anyway.

What Claude Managed Agents actually bundled on 8 April

Anthropic described Claude Managed Agents as "a suite of composable APIs for building and deploying cloud-hosted agents at scale". It went into public beta on the Claude Platform on 8 April 2026. You define the agent's tasks, tools and guardrails, and Anthropic runs it on its own infrastructure.

The launch post lists what that infrastructure covers. Secure sandboxing, authentication and tool execution are handled for you. Sessions can run autonomously for hours and persist through disconnections. Scoped permissions, identity management and execution tracing are built in.

Early users named in the post include Notion, Rakuten, Asana, Sentry and Atlassian. Credentials stored in vaults, self-hosted sandboxes and MCP tunnels appear in a later update section of the same page, so they were not all part of day one.

The $0.08 session-hour is the number that scared people

Pricing is consumption based. Standard Claude Platform token rates apply, plus $0.08 per session-hour for active runtime. That is a flat, published rate for something edge vendors had sold as several separately metered products: compute, sandboxed code execution, secrets handling and observability.

The fear was not the price level. It was the packaging. When the model vendor sells the runtime as one line item, the buyer stops shopping for the runtime. Our read is that this is the correct thing for investors to have worried about, and the wrong conclusion to have drawn from it. Anthropic's own detailed rate card is broken down in the line-by-line check of Claude pricing and API rates.

The Anthropic SaaS selloff hit edge stocks on 10 April

The selling arrived on Friday 10 April, two trading days after the launch. Coverage framed it as the latest round of the AI-driven software selloff that had already hit application vendors earlier in the year. That earlier round, and why it hit seat-priced software first, is covered in the explainer on the SaaSpocalypse selloff.

This time the targets were infrastructure. The 24/7 Wall St account, syndicated on Yahoo Finance, said Managed Agents "bundles code execution, credential management, and hosting, services these companies traditionally sold separately." That sentence is the whole bear case in one line.

Both articles were published on 11 April 2026 and describe Friday trading. Neither states whether its figures are closing or intraday moves, which is the most likely reason they differ. Treat every number in this table as directional.

Two sources, two sets of numbers

The headline range in this post, 11% to 18%, comes from the GuruFocus report because it is the only one covering all three named companies. The 24/7 Wall St figures are larger for Akamai and Cloudflare. We have not found an exchange closing-price record in this research that settles it, and we would rather show both than pick the convenient one.

The mismatch matters less than it looks. Whichever set you use, three profitable or near-profitable infrastructure businesses lost a double-digit share of their value on a beta launch with a published price of $0.08 an hour. That is a market pricing a category threat, not a revenue event.

Cloudflare: from target to Claude Managed Agents partner

Cloudflare moved first and most visibly. On 19 May 2026 it announced it was collaborating with Anthropic to launch Cloudflare Environments for Claude Managed Agents. That is 41 days after the launch that had knocked its shares down.

What Cloudflare Environments for Claude does

The split of work is the interesting part. The core agent loop runs on Claude. Cloudflare's network and Workers platform handle code execution, private connections and task-specific tools. A control plane built on Workers creates an isolated sandbox for each agent session.

Developers can choose Linux microVMs for heavy tasks or V8 isolate sandboxes with millisecond boot times. Outbound connection controls, private links to internal services through Cloudflare Mesh, audit trails and session recordings are included. Matthew Prince's line in the release was that "the true potential of AI agents will only be realized when they can safely interact with the real world at massive scale".

Read that architecture again with the selloff thesis in mind. The thing investors feared Anthropic would absorb, sandboxed execution next to private data, is precisely the layer Cloudflare now sells into Anthropic's product. The release gives no pricing and no availability status, so you cannot yet size what it earns.

Guidance went up twice while the company cut 1,100 jobs

Cloudflare's numbers before and after the partnership both moved up. Its first quarter release on 7 May reported revenue of $639.8 million, up 34% year on year, and guided 2026 revenue to $2,805 million to $2,813 million. The same release announced a reduction of about 1,100 roles as part of what it called an "agentic AI-first operating model".

The second quarter release on 6 August reported revenue of $696.1 million, up 36%, against its own May guide of $664 million to $665 million. Full-year guidance rose to $2,864 million to $2,870 million. Restructuring charges of $150.7 million pushed the GAAP operating loss to $205.7 million, while non-GAAP operating income was $96.1 million.

Prince's August framing was that the web is going through "a fundamental rewrite of the Internet for machine-to-machine traffic." Neither results release mentions Anthropic or Claude by name. Whether the layoffs are really about AI is a separate argument, examined in the evidence check on Cloudflare's AI layoff claim.

Akamai: the $1.8 billion contract nobody has confirmed

Akamai's response was not a product announcement. It was a line in an earnings release. On 7 May its first quarter results led with a key takeaway: a "leading, U.S. based frontier model provider commits $1.8 billion over seven years for Cloud Infrastructure Services."

The next day Bloomberg identified the customer as Anthropic, according to The Next Web's report of 9 May. That report says both companies declined to comment and that the shares rose 27% in a single day. A stock that fell double digits on an Anthropic launch then rallied harder on a reported Anthropic contract, inside four weeks.

What the filing says and does not say

The filing says $1.8 billion over seven years. Divided evenly that is about $257 million a year, though nothing in the release says the commitment is evenly spread. The Next Web reported that revenue from the deal begins in the fourth quarter of 2026 at roughly $20 million to $25 million for that period.

For scale, Akamai's Cloud Infrastructure Services line was $95 million in the first quarter, up 40%. An even $257 million a year would equal roughly 68% of that quarter's annualised CIS revenue. That is a large customer for a small segment, and it is not a large customer for Akamai overall.

The August release went further. Leighton said Akamai had signed "multi-year CIS contracts, collectively worth over $2.8 billion" year to date. That included a new U.S. technology customer worth more than $600 million over four years, which the release tied to robotics development rather than to model training or agents.

Why Akamai's guide went down anyway

This is the result that complicates every tidy version of the story. Akamai's second quarter release reported revenue of $1.100 billion, up 5%, with CIS at $99 million, up 39%. Full-year 2026 revenue guidance moved from $4,445 million to $4,550 million in May to $4,445 million to $4,530 million in August.

The top end came down $20 million. Non-GAAP earnings per share guidance moved from $6.40 to $7.15 down to $6.40 to $7.05. The reason sits in plain sight: Delivery and other cloud applications revenue was $396 million, down 6%, and it is still four times the size of CIS. We think this is the clearest lesson in the whole episode. An AI contract can be real and large and still be too small to outrun a shrinking legacy line.

Fastly: no Anthropic partnership, two guidance raises

Fastly took the worst hit in the GuruFocus figures and made no Anthropic announcement that we could find. It raised its outlook anyway, twice. That makes it the useful control case in this set.

Its first quarter release on 6 May reported revenue of $173.0 million, up 20%, with full-year guidance of $710 million to $725 million. The second quarter release on 5 August reported $183.3 million, up 23%, above its own guide of $170 million to $176 million. Full-year guidance rose to $732 million to $746 million.

Fastly's growth is coming from security, which grew 43% to $41.7 million in the second quarter. Network Services grew 17%. Kip Compton's line was that platform strategy gave the company "the confidence to raise our full-year outlook."

Fastly's agent strategy points the other way from Cloudflare's. Instead of hosting agents, it sells controls over them. Its bot management blocks unauthorised AI agents for publishers. A Skyfire partnership verifies and transacts with agents at the edge, and a Fastly Agent Toolkit serves coding agents. The company also published research claiming AI traffic grew 6.5 times faster than human traffic this year. That is Fastly's own measurement, so treat it as a vendor claim until someone else reproduces it.

Agent infrastructure guidance, side by side

Put the three companies' full-year revenue guidance on one table and the selloff thesis becomes testable. If Managed Agents were already taking revenue from edge vendors, you would expect it in the guides issued after the launch. You do not see it.

Guidance from each company's first and second quarter 2026 results releases. Midpoints and percentages are our arithmetic. Fastly's May range was itself already a raise on its earlier outlook.

Two readings fit this chart. One is that agent hosting is still too small to move any of these businesses in either direction, so the selloff and the partnerships are both ahead of the revenue. The other is that the base business decides the guide. Cloudflare and Fastly are growing at 23% to 36%, and Akamai is carrying a delivery line that shrinks every quarter. We lean firmly toward the second reading.

Where this argument is weakest

This post argues the 10 April selloff mispriced the threat. Several things could make that wrong, and some of them are already visible.

Six months is too short, and share prices are missing

Agent workloads on Managed Agents are a beta product's traffic. A threat that shows up in 2027 guidance would not show up in a 2026 guide, and an incumbent can raise guidance right up to the quarter its customers move. This post also quotes no current share prices. We tracked filings and event-day moves, and we did not verify a closing price series for any of the three since April, so it says nothing about whether the stocks have recovered.

Partnering can be a slow way to lose

The strongest counter-case is that Cloudflare's partnership is a concession. Anthropic owns the customer, the agent loop and the session-hour price, and Cloudflare supplies a sandbox underneath it. That is how platforms usually absorb suppliers: first as a partner, later as a commodity input with a single buyer. The point-solution version of that pattern is laid out in the analysis of how agent ecosystems absorb point SaaS.

The Akamai contract has the same weakness in a different form. Neither company has confirmed the customer. If it is Anthropic, Akamai has swapped some exposure to many delivery buyers for heavy exposure to one model lab. We also did not read the earnings call transcripts for this post. Management may have said more about agents on those calls than the releases do, and we would update this post if they did.

What this means if you are choosing where agents run

This post is written for founders and directors deciding where their own agents should live. The market story matters to you only through one question: does the runtime layer consolidate into the model vendor, or stay a separate purchase?

Six months of evidence says it is staying separate, for now. Anthropic sells a bundled runtime at $0.08 a session-hour, and it also launched with a partner that runs the sandbox on Cloudflare's network. Fastly is building controls for agents it does not host. That is three different places to put the same workload, each with a different lock-in profile.

Our recommendation is to keep the agent loop and the execution environment separable in your own architecture. If the model vendor's runtime is cheapest today, use it, and make sure your tools, secrets and data access do not depend on it. The cost of moving later is examined in the breakdown of model switching costs and lock-in.

Frequently asked questions

Where to start

Pull up your own agent architecture and draw one line: everything above it belongs to the model vendor, everything below it to you. If your credentials, tool definitions or data connections sit above that line, you have bought the bundle whether you meant to or not.

Then put the third quarter results of Cloudflare, Akamai and Fastly on your calendar and read one thing in each: whether agents or Anthropic appear by name. None of the August releases from Cloudflare or Akamai named Anthropic. The first one that does is telling you its agent revenue has become worth disclosing.

Related reading