Cloudflare status: hosting issues and outage reports

Problems in the last 24 hours

The graph below depicts the number of Cloudflare reports received over the last 24 hours by time of day. When the number of reports exceeds the baseline, represented by the red line, an outage is determined.

Most Reported Problems

The following are the most recent problems reported by Cloudflare users through our website.

• Cloud Services (37%)
• Domains (26%)
• Web Tools (21%)
• Hosting (11%)
• E-mail (5%)

Live Outage Map

The most recent Cloudflare outage reports came from the following cities:

City	Problem Type	Report Time
New York City	Cloud Services	9 days ago
Los Angeles	Cloud Services	10 days ago
Paris	Cloud Services	26 days ago
New York City	Hosting	28 days ago
Manchester	Domains	2 months ago
Angers	Cloud Services	2 months ago

Full Outage Map

Community Discussion

Tips? Frustrations? Share them here. Useful comments include a description of the problem, city and postal code.

Beware of "support numbers" or "recovery" accounts that might be posted below. Make sure to report and downvote those comments. Avoid posting your personal information.

Cloudflare Issues Reports

Latest outage, problems and issue reports in social media:

• 
  Daniel (@Arxonic) reported 2 minutes ago

  @gregisenberg @Cloudflare 6/ Startup Idea #1: The Niche Data Refinery Pick one messy niche. Clean the data into agent fuel. Wedge: 100 businesses in one city. First customer: The business that needs clean data to exist. First version: Manual data entry + API.

• 
  chris catlett (@ccatlett1984) reported 6 minutes ago

  @Prusa3D the embeds for printables are broken in Discord again. looks like a cloudflare/CDN config. needs to allow discord's link-preview crawler. Need a custom WAF rule for UserAgent: Discordbot, or allow in Security - Bots - Bot Fight Mode / Super Bot Fight Mode @NomadsGalaxy

• 
  Paul (@hasieratik) reported 10 minutes ago

  @alexcloudstar it was an error on my end, to be honest; as I gave it green-light; just didn't think things through :D I've made sure to include in the system prompt instructions to specifically analyse for possible outcomes like these. Cloudflare was nice enough to wipe the bill as I'm an idiot

• 
  Eric Schweizer (@ericschweizer) reported 11 minutes ago

  @masoncapitalllc @remarks They did this with cloudflare last week. Put AI in charge of report tickets screening. It was a total disaster. AI was shutting down legitimate sites hundreds of them. A site dedicated on preventing sites from being shut down was the one doing the shut downs. AI not reliable.

• 
  Preet Patel (@PreetPatel_04) reported 12 minutes ago

  @riyasayshie The cloudflare dns problem you need to fix

• 
  Logan Weaver (@LogWeaver) reported 14 minutes ago

  We think we are the main user of the internet. That’s wrong. Machines now make more than HALF of all web traffic. This report comes from Cloudflare, and even Google and Meta should be nervous about this. Cloudflare runs the plumbing for about one in five websites. So it sees a big slice of all web traffic. The machines are mostly AI bots and agents. They scan the web to train and feed AI models. They read pages far faster than any person can. A person might visit a few websites to buy something. An AI can visit thousands in the same moment. Not long ago bots were about 20 percent of traffic. Now they are more than half of it. Cloudflare expected this to happen around 2027. It arrived a full year early instead. Its finance chief gave investors a blunt warning: Human activity could become a rounding error in five years. Bots could outnumber people by a thousand to one. Picture a web built almost entirely for machines. Now here is the part the market keeps ignoring. The free internet runs on ads shown to people. More and more, an AI reads the page for you. You get the answer and never visit the site. That site gets no visitor and no ad money. Google and Meta are the kings of the ad model. Their fortunes depend on human eyes and human clicks. But those humans are becoming the smaller half. Bots do not watch ads or buy products. These are some of the most valuable companies alive. Their whole engine sits on a shrinking base. Even Cloudflare admits the old model is breaking. It now wants to charge the bots to visit. That risk barely shows up in the headlines. The crowd sees AI as pure good news for big tech. This is a slow shift, not a crash tomorrow. But slow shifts are the ones that reprice giants. The biggest changes are often the quietest ones. The crowd chased the obvious story. The data told a different one. That's the whole game. Surmount runs on clear rules and real data, not hype. If you want a system that follows facts, start here:

• 
  Robin (@MediaGoddess1) reported 19 minutes ago

  @HedgieMarkets Real humans to be squeezed out of the www. "To Cloudflare this is a capacity problem" Not a content problem; certainly not an industry self-regulatory problem.

• 
  Nicholas Charriere (@nichochar) reported 20 minutes ago

  The actor model is winning: with cellId and the rise of serverless we’re seeing patterns emerge as most popular for modern workloads: workers, durable objects and message passing. The funny (and annoying) thing is that the perfect stack exists but isn’t being used. We’re reinventing in rust and typescript what was built already, and boy was it built well. Elixir is based on the BEAM VM, which implements actors perfectly. No shared state, immutable data, gen servers for long lived stateful processes, supervision trees for reliability Then, Elixir introduces ergonomics on top. The language is inspired by Ruby and tries really hard to make itself a delight to use. Really strong developer ergonomics (pattern matching, overloading, type hints, all-in-one build tool and linter, thoughtful keywords, strong conventions, etc…) It is quite literally overpowered. I used it at Pinterest and we would take services built in python that were maintained by a team of 20 and replace with one elixir service maintained by less than one person. Multiple times. If well designed they simply never go down and scale beautifully. We also used it at Mocha and had a 10x more reliable app than lovable or replit. In fact one of the reasons we kept competing is that we felt their apps were garbage (they have since improved a lot, which is a frustrating mistake I have since learned from). In the age of agents, elixir should absolutely be winning, and indeed the actor model is! It’s a perfect fit: long lived processes with very unreliable third party processes and humans in the loop, all orchestrated to be long lived and reliable with not too much internal state. But as I learned, unfortunately the best tech doesn’t win, capital wins. And most programming languages won because they had massive capital (think companies) pushing them forward. Rust had Mozilla, Go had Google. typescript had Microsoft (JavaScript is a bit different, it won because it’s the runtime of the biggest platform ever made: the web). This is frustrating but it is also reality, so I guess we should be happy to see great implementations (rivet in rust, workers/DO by cloudflare) becoming mature. The actor model is the best programming model for most web workflows.

• 
  Lucas Gray-Joy (@techdreamergc) reported 21 minutes ago

  @HedgieMarkets I think it's an incentive problem. Cloudflare profits from bot traffic. Proof-of-personhood could create a verified human layer, but nobody running the pipes has reason to build it.

• 
  Joe (@LTSmash420) reported 22 minutes ago

  Google’s AI Push Is Progress, Not Theft – Stop Whining About the Old Web The Bloomberg column fretting that Google’s AI is “killing web creativity” ignores how AI tools deliver faster answers, better information access, and real innovation while the old traffic-for-indexing bargain was never a sacred right. Parmy Olson’s latest Bloomberg opinion piece paints Google’s use of its search crawler for AI training as some underhanded plot that starves websites of human visitors and hands the company an unfair edge. The argument recycles familiar complaints: AI Overviews and similar features keep users on the results page instead of sending them to publishers, traffic drops, and the open web suffers. A British regulator and Cloudflare get cast as the plucky resistance finally forcing change. The framing is pure nostalgia dressed up as concern for creators. The original Google bargain was simple commerce, not a perpetual subsidy for every blog and news site. Search engines indexed content and sent readers; sites gained visibility and ad revenue. That model worked while search remained a list of blue links. Once generative AI proved it could synthesize useful answers from the same public data, the economics shifted. Users prefer concise, direct responses over hunting through half a dozen pages of SEO-optimized filler. That is progress, not predation. Critics treat declining referral traffic as proof of corporate malice. In reality it reflects user preference. When an AI summary answers a question accurately and cites sources, the click becomes optional. Studies and publisher reports confirm zero-click searches rose after AI features expanded, yet the same reports rarely acknowledge that many of those visits previously delivered little value beyond a single page view and a banner ad. Quality journalism and unique analysis still attract direct audiences and loyal readers. Sites that produced thin content optimized solely for ranking algorithms were always vulnerable. AI simply accelerated the exposure of that weakness. Cloudflare’s efforts to let sites block AI crawlers by default and the UK Competition and Markets Authority’s rules requiring clearer opt-outs for AI use of search content address real technical imbalances. Google’s dual-purpose crawler did create a practical advantage because blocking it meant disappearing from search entirely. Separating the functions is a reasonable competitive fix. Turning that fix into a broader campaign against AI itself is not. The deeper problem with the “AI is killing the web” narrative is its Luddite streak. It treats the open web of the 2000s and 2010s as the final form of information distribution rather than a transitional phase. AI systems trained on publicly available material can surface knowledge faster, cross-reference claims, and reduce the friction of finding reliable answers. Tools like Grok demonstrate the upside: direct, less filtered responses grounded in real-time data and reasoning instead of the slow drip of algorithmic ranking and ad-driven incentives. Publishers who adapt by offering deeper reporting, exclusive data, or community features will continue to thrive. Those who demand the old referral pipeline continue forever are asking regulators to freeze technology in place for the benefit of yesterday’s business model. Resistance from regulators and infrastructure players is healthy when it levels the playing field on data access. It becomes counterproductive when it frames every AI improvement as an existential threat to creativity. The web was never static. Search itself disrupted directories and print classifieds. Mobile apps redirected attention. Now AI is reordering discovery again. Creators who produce work worth seeking out will find audiences. The rest will adjust or fade, just as they always have. Celebrating that dynamic rather than mourning the loss of guaranteed traffic is the only stance consistent with technological reality.

• 
  Neo (@anon_neoki) reported 24 minutes ago

  @Cloudflare why is it so hard to login to your website all of the time? it just endlessly loads, its atrocious

• 
  Jacob Frautschi (@JacobFrautschi) reported 25 minutes ago

  @ToKTeacher the business model of electricity and most of the internet (advertising) is very different. cloudflare is trying to help its customers monetize agentic access.

• 
  Ryan Hicks (@RyanThomasHicks) reported 25 minutes ago

  @waynesutton @Cloudflare @CloudflareDev This would be nice! Ran into an issue on setup this weekend myself

• 
  kerfuffle (@kerfuffledev) reported 27 minutes ago

  @rauchg cloudflare has been doing a lot of agentic stuff, do you think it will be much better to stay with ver el if Im a heavy user of hermes? ngl I was inclined to transfwr to cloudflare but if there are opinions that could help me stay with vercel, maybe I should just stay.

• 
  Anjli (@AnjliAmin) reported 31 minutes ago

  @Cloudflare just gave AI agents wallets with a $100 weekly cap. Everyone's calling it a control. It's an admission. A cap can't make an agent trustworthy, only limit the damage before a human notices. Letting agents pay was never the hard part. Owning the loss when one pays wrong is. That said it'll be a low risk environment to test out how often things go wrong and what happens when they do

• 
  willard shoko (@WillarShoko) reported 34 minutes ago

  Cloudflare says humans could become a "rounding error" as bots generate 1,000 times more internet traffic Just a couple of months after Cloudflare revealed that bots have officially surpassed human traffic online, the company has given its view on what things will look like in five years. It's not good news for us fleshy meatsacks: non-human traffic could be up to 1,000 times higher than human traffic. CDN and cybersecurity giant Cloudflare held its Q2 earnings call on Thursday, during which chief financial officer Thomas Seifert gave a concerning prediction about what the web will look like in the future. "If the current trends continue, we think in five years, non-human traffic will be as much as 1,000 times as much as human traffic," he said. "In other words, humans will be a rounding error on the internet, not because human traffic goes down, but that's just how fast we're seeing non-human traffic grow."

• 
  The Regime Report (@theregimereport) reported 42 minutes ago

  Nasdaq Trap Resolution Jul 22: Nasdaq R² = 0.48 Jul 29: R² = 0.29. Ranging confirmed. Tools OFF. Every trend signal was disabled. EMA. MACD. ADX. All deactivated. Then: Palantir +29%. Intel +10.8%. Airbnb +8%. Cloudflare +16%. NFP -23k. Aug 7: R² = 0.50. Trap resolved. Full tool set restored. 9 sessions from ranging to moderate trend. Best week since April. +5.2%. The regime did not care about the headlines. It responded to the structure. Know the regime before you pick the tool.

• 
  DigitalAssetBuzz 🔶 (@DAssetBuzz) reported 44 minutes ago

  This is a meaningful improvement. Cutting a browser workflow from dozens of model round-trips to one generated script should reduce latency, token burn, and a lot of unnecessary failure points. But one of the complaints I keep seeing around browser agents is still authenticated state. Users don’t care how elegant the browser abstraction is if the agent suddenly lands in the wrong profile, loses cookies, hits a login wall, or can’t recover from Cloudflare/CAPTCHA. Hermes itself has had reports around browser tooling using the wrong browser session and breaking authenticated workflows. That’s where I think the next step is: one browser tool plus governed session persistence and a clean human-handoff path when authentication blocks the workflow. Less clicking is good. Reliable recovery is what makes it production-ready.

• 
  orlie (@sunglassesface) reported 45 minutes ago

  @burcs Telemetry right now looks like crap even with the open telemetry stuff I would redo the documentation and show various degrees of telemetry from basic to advanced and how to configure it so that people can debug workers properly and get the right amount of information out of workers Even the fact that cloudflare automatically redacts routes is quite annoying

• 
  mizuki (@vem4s156553) reported 46 minutes ago

  @dollmanipulator wtf is cloudflare

• 
  murdok (@murdok_gg) reported 46 minutes ago

  (would you believe that i only spend $70 on groceries per week, but i splurged by buying 2 domains for $25 for a year a piece?! omg its crazy that cloudflare can sell at cost from a reg that people like, customer service is interesting these days...)

• 
  Enkhmanal 🟠 (@Enkhmanal) reported 48 minutes ago

  $NET - CLOUDFLARE'S 2021 CONVERTIBLE COMES DUE THIS MONTH, AND THE HEDGE AGAINST IT CAPS OUT AT $250.94 Cloudflare runs the network that sits in front of a large slice of the web, absorbing attacks and serving pages from close to whoever asked for them. This morning it proposed $2.175 billion of new convertible notes. On August 15, the $1.29 billion of 0% notes it sold in 2021 mature. A 0% coupon is never free money, and Cloudflare knew it. Alongside those notes it bought capped calls — a hedge that hands back shares to offset the dilution when conversion arrives. That hedge switches on at the $191.34 conversion price and stops at $250.94. The stock closed Friday at $300.78. So the settlement splits three ways: the $1.29 billion of principal in cash, which the company irrevocably elected back in May; stock for the conversion value above it; and a partial refund of that stock from the hedge, but only as far as the cap. Everything above $250.94 is dilution Cloudflare simply wears. That is the bargain inside every capped call — cheaper protection, bought by agreeing in advance on how well your own stock is allowed to do.

• 
  泰多宝 (@TabDuoBao) reported 52 minutes ago

  @Cloudflare The one thing worth pinning down in a number like 519% is whether "hyper-volumetric" is a fixed threshold or a moving one. If the bar stayed at a set Tbps figure while the ambient ceiling rose, part of that growth is attacks crossing a line that didn't move rather than a change in attacker behaviour. Both readings can be true at once and they imply different defensive priorities — one says provision more headroom, the other says the composition of the botnets changed. The report is more useful than the headline if it splits those.

• 
  Kite AI Community and Ecosystem (@Kite_Frens_Eco) reported 53 minutes ago

  AI answer engines take content from the entire web but no longer send traffic back, hollowing out the foundation of the content economy. In this AI on Air clip, excerpted from a recent episode on the @Bankless show, the speaker is Matthew Prince @eastdakota, co-founder and CEO of @Cloudflare. ▷ The ads-based business model of the internet is approaching its end game: ad blockers had already pushed CPMs into a slow decline, but over the last two years AI has caused a near step-function drop in ad value. ▷ Google's AI overviews, ChatGPT, and Claude are not search engines but answer engines: they strip mine the web and hand users the answer directly, so no one clicks the links. Without eyeball traffic there is no ad revenue, and content creation is either retreating behind paywalls or shutting down entirely. ▷ For users this saves time, but these AI tools run on paid subscriptions. For people in the global south who cannot afford them, the internet is getting smaller, not bigger.

• 
  KalicoShade (@MidKnightKurama) reported 54 minutes ago

  @Labyrinthusian @TutaPrivacy All google or any cloudflare or AWS based website has to do is push a code line that says Is_google_first_party_true.False, if your device returns false, your blocked from ANYTHING that also uses google. Same for cloudflare, same for AWS. You dont even need to be on their service

• 
  Nyagah-254 (@NyagahTony) reported 55 minutes ago

  Stopping the bad guys with Cloudflare: 22,280 malicious requests blocked or challenged in the last month #cloudflare This is from a domain that's pretty much dormant 🤔

• 
  Alexander Zuev (@zlxndr) reported 58 minutes ago

  I were looking for ways to 10x the DX of @CloudflareDev (which is already great, ofc), formulated in a withlist: 1. Make every error thrown by any binding or the runtime carry a stable, documented .code - not a code embedded in the message like R2's (10024), an actual property 2. Every such error should carry .retryable, typed, on all bindings - you already ship it on DO stubs, extend it and put it in @cloudflare/workers-types 3. Both should survive the boundaries that currently flatten errors: RPC serialization and NonRetryableError in Workflows This will make error workers error handling no longer depend on string parsing for each & every binding I use. Is it too much to ask? Merci beaucoup in advance!

• 
  CoderLuii (@CoderLuii) reported 58 minutes ago

  @pcshipp cloudflare pages. static hosting is free, and having workers + d1 in the same place meant i never had to stand up a separate backend for the control plane

• 
  Harsh (@harsh_biz) reported 59 minutes ago

  The deliverability meta going into Q4 2026 is different from what worked in Q1. If you're still building infrastructure the way you did at the start of the year, you're going to feel it next quarter. SURBL made a lot of noise this quarter. Cold email inboxes got flagged in batches, everyone scrambled, and the enforcement has already lost most of its teeth. But the fixes are still the right build. Not because SURBL is going to catch you. Because when you build this way, there's almost nothing left for any ESP to pattern-match against. SURBL won't be relevant soon from what I think. You're no longer defending against a specific list. You're removing the mechanical fingerprint that makes automated detection possible at all. Practically: Keep domains to 20 per registrar account, 50 at the absolute ceiling, spread across multiple accounts. Once your domains aren't clustered, there's no cluster to flag. Stop naming them [Company]aa, [Company]ab, [Company]ac. Branch out entirely instead. If the brand is Milbridge Capital, use Milbridge Advisory, MBCapital, MilbridgeGroup, or drop the brand pattern altogether and use unrelated professional-sounding names. Older domains with genuinely varied names get listed far less often, and that holds regardless of which blacklist happens to be active. Placement tests are dead. They were reliable in early 2025. They aren't now. You can score 100% inbox placement on a test and pull a 0% reply rate on the live campaign the next day. Replies are the only trustworthy signal left. Build your monitoring around reply rate at the sender-email level, excluding auto-replies, and pull inboxes when they drop below threshold. Masking beats redirects, and both beat nothing. Redirects have visibly hurt deliverability this year. Dead domains hurt too, because prospects check the domain and expect a real site. Build masking through Cloudflare Workers rather than paying a markup for it. One free Cloudflare account holds around 80 domains. Your lead source is now a deliverability factor. Cheapest leads reply the least on average, and low reply rates suppress sender reputation. That's not a relevance problem, it's an infrastructure problem. When a campaign starts dying, diagnose before you rebuild. If two or three campaigns trend down at once, it's the copy. Rotate it. If you rotate the copy well and nothing recovers, it's the inboxes. Fix the wrong one and you burn another month.

• 
  Aryuemaan Chowdhury (@aryuminstrel) reported 1 hour ago

  Stopping the bad guys with Cloudflare 1,314 malicious requests blocked and challenged in the last month #cloudflare