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.

• Domains (40%)
• Cloud Services (27%)
• Web Tools (13%)
• Hosting (13%)
• E-mail (7%)

Live Outage Map

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

City	Problem Type	Report Time
Manchester	Domains	14 days ago
Angers	Cloud Services	26 days ago
London	Domains	28 days ago
Noida	Hosting	1 month ago
Jewar	E-mail	1 month ago
Braga	Web Tools	1 month 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:

• 
  Omid Saffari (@omidsaffari) reported 4 minutes ago

  An AI gateway is not a production badge. It is a control tax. Pay it when a second model, a second provider, or a second engineer touches your LLM calls. Before that, call the provider directly and ship. The moment that line gets crossed, the gateway starts earning its hop. Why? Because the real problem is not "how do we call another model?" It is duplicated retry logic, missing spend attribution, no shared rate limits, inconsistent logs, and a fallback path nobody has tested. My decision rule: Single feature, single model: no gateway. Fast free visibility: Cloudflare. Already on Vercel and want zero token markup: Vercel AI Gateway. Managed governance, RBAC, guardrails, audit path: Portkey. Keys must stay inside your perimeter: LiteLLM, with the operational work that comes with self-hosting. The part teams underestimate is not setup. It is ownership. A fallback only counts if it fails over on real signals: timeouts, 5xx, 429, and invalid response shape. A cost dashboard only helps if every request is tagged by customer, feature, environment, and model. Semantic caching only works if you have measured false hits against real queries. And if you self-host, patching is now your calendar. LiteLLM's 2026 CVE hit CVSS 8.8 on NVD and was fixed in 1.83.7. The lesson is not "avoid LiteLLM." The lesson is that a gateway holds the keys, so it has to be treated like security-critical infrastructure. Add the gateway when it centralizes control. Not because the diagram looks mature.

• 
  Warya Wayne (@WaryaWayne) reported 4 minutes ago

  I am having a problem with a status check website I made. It is spamming me that the sites are down when they are not and i think it's Cloudflare blocking them do to them checking every 180 secs and notifying me but I have 200 notifications saying it's down and they're up? How?

• 
  PODCONF® (@_podconf) reported 6 minutes ago

  @Cloudflare When support for COMPLIANCE•TOKEN ®

• 
  Rahul Kumar Singh (@rahulitblog) reported 7 minutes ago

  @Cloudflare @Cloudflare, what about Cloudflare Pages support?

• 
  Diogo Souza (@diogocode) reported 10 minutes ago

  Cloudflare putting AI Search sync jobs in Wrangler is a small RAG ops signal: indexes are becoming CI/CD work, not dashboard chores. If an agent depends on fresh docs, trigger, inspect, cancel, and log the refresh from the pipeline.

• 
  Dhravya Shah (@DhravyaShah) reported 11 minutes ago

  difference between cloudflare vs vercel (ignoring cost, purely on product quality): we use cloudflare extensively and pay thousands of dollars every month, so I'm obviously a BIG BIG FAN of everything they do. But just think about it Both are doing everything (general AI cloud). but when vercel does it, they do it with utmost taste, obsession, and thoughtfulness. There's almost never a hiccup, missed doc, question or error in the way. When cloudflare does it, it's cool - Awesome concept. Not 100% at its potential yet, the product might as well not be maintained in a few months. But it exists today. Technically cloudflare has the opportunity to build a vercel (early obsession around devplat), a turbopuffer (better Vectorize), a browserbase (better browser rendering), maybe even a together AI (workers AI++), Openrouter (AI Gateway ++), Resend (Email sending ++), Blacksmith (FAST and cheap builds++). even Supermemory (better AI memory product?) But they won't, they can't. And this is only accounting for a small section of the stuff they already have. I'm sure all the products drive a lot of revenue, I'm not concerned about that as a user. I'm concerned about my experience. imagine how big of a generational difference it makes to make the best product for that particular industry. I gotta give it to vercel for nailing everything they build. v0 is legit top tier, AI sdk is a top AI framework. NextJS is the biggest web framework. Workflows has a beautiful DX, their gateway is really good, too. They did a fantastic job with fluid compute. Kinda feels like magic It's clear that vercel is super obsessed and driven towards specific goals, and when they do something, they do it really really well. It's a pleasure to see them execute. Just like Cognition did, Vercel could one day do an ad saying "Remember vercel? It's not expensive anymore" and could just win the market. (I'm not even sure if it really is expensive. My judgement is years old at this point) Today I spent like 4-5 hours trying to do something really simple with cloudflare (a preview deployment. that's it) and was unsuccessful, rage quit and felt like ****, felt super tired and burnt out. like it ruined the entire day. All this while I couldn't stop but wonder: What would be the experience on Vercel? how would it feel? Is all this pain worth it? Also insane that I encounter these scenarios despite working there for more than a year and having full access to everyone there for any help I need. I cannot imagine the experience for people just getting started / learning about cloudflare. It is also very clear that Cloudflare is executing too! The dashboard is improving a lot, the products are getting better. But not 100% there - This is why I've invested a LOT in cloudflare personally (both mentally and monetarily) and I believe in them. At the same time as a founder + builder, it's fascinating to see a company kill it at the scale Vercel does tbh. @rauchg is inspirational!

• 
  Peter Mindenhall (@PeterMindenhall) reported 12 minutes ago

  @rustybrick @JohnMu Hmm - this is precisely why companies like @Cloudflare should not be doing SEO things - they cannot be trusted and make crap up. They have an opportunity to do good things well - yet they are making a mess and causing confusion for site owners.

• 
  StaffSignal (@staffsignal) reported 13 minutes ago

  6. Rate Limiting + Circuit Breakers + Graceful Degradation • Rate limit at the edge (API Gateway / Cloudflare) • Implement circuit breakers so one failing service doesn’t take down everything • Have feature flags to temporarily disable expensive features (recommendations, complex queries, high-res images) • Return graceful errors or cached data instead of 500s Users forgive slow responses more than complete outages.

• 
  Vedant Anand 🐲/acc (@Vedantsx) reported 13 minutes ago

  @samlambert Gotcha ser 🫡 Btw I'm unable to migrate from Supabase to Planetscale in Cloudflare Startup program, can you kindly help me somehow?

• 
  SpyPigeon (@spypidgeon2) reported 14 minutes ago

  @__alula in fairness, I think they're drawing from a larger database for detecting CSAM content that other companies also use. I saw someone have a similar problem with their website running cloudflare. Hope the problem gets fixed soon though because this is so bad for texture artists

• 
  RoadRunner TX (@FredBra93439257) reported 15 minutes ago

  @TheMercianNews I'd like to see mobile phone and internet companies shut down operating in the UK on a single day. Have USN/Seals sever all international fiber into the UK on a single day. SHUT DOWN the City of London Physically SEVER International Fiber from BT/OpenReach, VirginMedia/O2, Vodaphone and the LINX Nework internchange in London. Ideally each line in 2-3 points of varying distance from UK shores to make repair more complex than laying new fiber. Air attack on all the major BT interchanges plus the physical operations of Cloudflare Akamai, AWS/Cloudfront, GoogleCloud CDN and smaller operators like Azure, Netflix, Meta and every source of new fiber either mfg or warehousing Destroy every piece of equipment and then just leave. Make all European govts TERRIFIED of the crazy Americans Destroy the internet as a means of brainwashing the British public

• 
  DarkEndMoon (@DarkEndMoon1) reported 20 minutes ago

  @hskenncutter Just a few months ago there were bot attacks from them. - X had problems and Cloudflare went down and brought a lot of US infrastructure down, including hospitals for a day, they could not authenticate into systems. Cloudflare is a united states verification and security protocol, one of their jobs is stopping bots. - One of the things they do with these coordinated attacks can be seen on X, gets slow, lags and there are ten times more comments than normal. They use the same infrastructure systems so it has an additive effect if you overwhelm the whole system in general. X or Elon, and several other agencies have been fighting this for a while now. Bots keep changing attacks keeps changing AIs supporting the attacks with the bot armies keep changing. Pretty crazy when you think about it.

• 
  Vivek Kotecha (@vbkotecha) reported 20 minutes ago

  Solana just became the second x402 network. Every major player now supports dual-chain: Base + Solana. Alchemy, AWS CloudFront, Cloudflare Gateway. If your x402 endpoint is only on Base, you're leaving half the agent economy on the table.

• 
  Udemezue John ☀️ (@_udemezue) reported 22 minutes ago

  The internet just hit a crazy milestone. More than half of all web traffic is now bots, not humans. According to Cloudflare, the massive company that protects and runs a huge chunk of the internet, we have officially crossed a historic line. Their CEO recently shared data showing that automated bots now make up over 57% of all internet activity. For the first time in the history of the digital world, real human beings are in the minority when it comes to browsing the web. What is truly wild about this news is how fast it happened. The tech experts knew the machines would eventually take over web traffic, but the CEO originally predicted it wouldn't happen for another year or two. Instead, the boom happened practically overnight. It turns out that the explosive growth of new artificial intelligence tools has accelerated everything way faster than anyone anticipated. If you are wondering what these bots are actually doing, they are not just annoying spam accounts or malicious hackers. The vast majority of this traffic comes from AI "agents" searching the web on behalf of humans. Think of it this way: if you want to buy a new laptop, you might open five different websites to compare prices. But if you ask an AI assistant to find you the best deal, that AI might scan five thousand websites in a single second. Even though a human asked the question, the sheer mountain of web traffic is being created by a machine. This shift is causing some serious issues for the people who actually build and own websites. Small blogs and independent sites are suddenly getting slammed with millions of visits from AI bots. This strains their servers and costs them a lot of money for hosting, but it doesn't actually bring in real human customers or advertising money to pay the bills. Because of this massive shift, companies like Cloudflare are starting to roll out new rules and features. They want to give website owners the power to block these AI bots, or at least force the big tech companies to pay creators when an AI steals their content to answer a user's prompt. It really makes you think about the future of the internet, as we enter an era where most of the web is just machines talking to other machines. Details in link below:

• 
  Alex Prompts (@alexpromptz) reported 25 minutes ago

  @Cloudflare Which crypto network we using?

• 
  Alphatica (@alphaticaio) reported 28 minutes ago

  LARGE / MID-CAP DARK POOL | July 7, 2026 $NET +$142M zero sells. Cloudflare is the mid-cap safe haven. $CMG -$83M zero buys. $RKLB -$22M. $RIVN -$21M. Both agree sell. 62 prints. Q3 week two. $289M in buy prints. $308M in sells. Net: -$19M. Nearly flat. Buyers: $NET +$142M (15 prints, $0 DP SELLS, BOTH TAPES AGREE) $CRWV +$22M (3 prints, 2ND SESSION BUY) Sellers: $CMG -$83M (4 prints, $0 DP BUYS, ZERO) $EXC -$28M (1 print) $IREN -$24M (4 prints) $LITE -$23M (6 prints, BOTH TAPES AGREE SELL) $RKLB -$22M (2 prints, BOTH TAPES AGREE SELL) $RIVN -$21M (2 prints, BOTH TAPES AGREE SELL) $NET: +$142M. Zero dark pool sells. Both tapes agree buy. On a sell day where the mid-cap blocks were -$726M and the internals hit series lows on every metric, Cloudflare is the only mid-cap name the dark pool bought without opposition. +$142M dark pool, +$79M lit tape. Zero opposition on both venues. The mid-cap safe haven on a sell day is a cloud infrastructure name. $CMG: -$83M. Zero buys. Chipotle. Consumer discretionary sold with zero dark pool buying. Not one print on the buy side. $RKLB: -$22M. $RIVN: -$21M. Both agree sell on both tapes. Rocket Lab and Rivian, the growth/EV names that led Q3 week one, are being sold on both venues. The growth sell is cross-venue. $IREN: -$24M. Iris Energy. Bitcoin mining. Crypto-adjacent sold. $LITE: -$23M. Lumentum. Optical networking. Semi-adjacent. The sell is concentrated in semi-adjacent, EV, and crypto-adjacent names. The only buyer is cloud infrastructure. Yesterday: $MRNA +$123M zero sells was the headline. Today: $NET +$142M zero sells. The mid-cap dark pool is producing a new name-level signal each session. The patterns accumulate. The weekly scorecard builds. Watching the tape.

• 
  Tobias_Petry.sql (@tobias_petry) reported 31 minutes ago

  @jaydrogers @mattiasgeniar Its too new. Give cloudflare a few months. Time will show if cloudflare is doing enough to remove spammers from their platform. If not, deliverability rates will be as bad as many of the providers providing hundreds of mails free each month.

• 
  Alberto Schiabel (@jkomyno) reported 34 minutes ago

  Just fixed an annoying dev-server bug on Astro v7 + cloudflare. The first request after a cold cache triggered a mid-render dep-optimizer reload that loaded two copies of React and broke every island. Thankfully, the fix is a one-liner

• 
  Jawad Dashti CRE ⚡️ (@TheJawadDashti) reported 36 minutes ago

  @Cloudflare It appears consensus is that this is a good idea but stables coins is a terrible idea.

• 
  Rohit Kashyap | AI + Full-Stack (@rohit_jsfreaky) reported 38 minutes ago

  @aschmelyun cloudflare handling the send side plus native laravel support takes a real headache off deliverability

• 
  zyn laden 🇺🇸🥋 (@ho_chi_zyn) reported 38 minutes ago

  @thijstriemstra @vxunderground @Cloudflare Grass is dangerous too bro, got my **** sprayed up with permethrin and picardirin on me skin and I'm still seeing ticks on my socks

• 
  Fire Watcher Watch (@FireWtcherWatch) reported 40 minutes ago

  @NeelMehta420 @khyimiq @DalitDetector One of the more recent major CloudFlare outages (February this year) was because they pushed a change which automatically deregistered 25% of BYOIP addresses. Total incompetence, and caused a multi-hour outage for many millions of people.

• 
  Filecoin (@Filecoin) reported 40 minutes ago

  @thetobiaskrug @john_zuccaro @Cloudflare proof of possession and integrity comes first, ZK proofs over the data are a harder problem, but worth building toward.

• 
  masYNYa (@KijAkubovs86334) reported 49 minutes ago

  🚨 EVERYONE'S RUSHING TO USE FABLE 5 WHILE IT'S FREE. NOBODY'S TELLING YOU WHAT ORDER TO DO IT IN 🌐 The tools are half the work. The flow is what separates a site that looks templated from one you'd actually ship. Pause at 0:05. Look at the two screens. Foreground: a MacBook Pro on a dark desk, showing a site titled "Lightweight." Nav bar: WHEELS, RRC COCKPIT, ABOUT US, DEALER, SERVICE. Hero: "HANDMADE IN..." in a serif you'd fight your designer over. A single carbon wheel below. Sidebar labels: PHILOSOPHY, CRAFT, INNOVATION. Background, projected on the wall: a Claude usage dashboard reading "168.5M tokens, 71 active days, peak hour 11 PM, favorite model Opus 4.8." That is the ecosystem this workflow runs inside. The Lightweight site is the reference target, not the output. The output is whatever you're about to build against that bar. The exact sequence, six steps: → 1. Open Claude Code in an empty folder — nothing to bias the model, no legacy files, clean slate → 2. Find a reference on Awwwards. Screen-record it or grab the link and hand it to Fable. Do not describe the vibe. Show it. → 3. Make it write a SKILL FILE first. "Turn my brief into a production guide with brand, fonts, colors, layout rules, then follow it." This is the step that kills the generic AI-slop look. → 4. Run the first build on High. Let the model design the structure, then react to what comes back. → 5. Refine in passes. One section at a time. "Make the hero scroll-driven." "Tighten the type." Not one giant edit prompt. → 6. Deploy free on Cloudflare Pages. Tell Fable to do it — it knows the CLI, it will handle the push. The point worth arguing about: → Step 3 is the entire game. Everyone else skips it. → Writing a skill file forces the model to codify the design system BEFORE generating a single component → Most people describe the vibe in a prompt and get a page that looks like every other AI page → Skill files invert the default: constraints first, output second → Whether this is "just better prompt engineering" or a genuine change in how you use the tool is exactly the argument worth having What each step actually replaces: → "Vibe coding" a site by asking for "modern, clean, professional" (Step 2) → Trying to describe a color palette in words (Step 2 again) → Prompt engineering a monster prompt that half works (Step 3) → Getting a full site in one shot and hating it, then trashing it (Step 5) → Paying $12/month for Vercel Pro when you have five brand launches to ship (Step 6) What it does NOT replace: → Taste. You still have to pick a good reference. → Judgment. You still have to say "no, tighten the type" and know when the type is finally right. → The willingness to run six passes instead of one. The old loop is dead. Ask Fable "build me a modern SaaS landing page." Wait 3 minutes. Get something that could be anyone. Post a screenshot with "look what AI did." Delete it two days later. Now you screen-record Lightweight, make Fable write the design system, and ship a site that could have come out of a boutique studio. Here is the question I want an answer to in the comments: Of the six steps — which one do you actually skip? Mine's step 3 (skill file first). I default to describing the vibe and it burns me every time. Yours is which — and what would it take to actually put it back in the loop? Bookmark this. Answer below. The race for shippable design just left the "just prompt it" school. Literally.

• 
  Roy (@__roycohen) reported 50 minutes ago

  @DhravyaShah I'm gonna expand on his point here to anyone who works at @Cloudflare "Today I spent like 4-5 hours trying to do something really simple with cloudflare (a preview deployment. that's it) and was unsuccessful, rage quit and felt like ****, felt super tired and burnt out." I have always had this happen with every product that I've tried to use on the frontend side. The AI bot doesn't even properly queue up any suggestions, it's also quite useless/slow. It could be so much better honestly. It takes time and Cloudflare has an insanely complicated product, I agree, but some stuff like hiding logs when it's erroring for... aesthetic purposes? Deploying a worker just made me ragequit and I gave up, the issue is that you sacrifice your ability to fix any problem when the underlying documentation and options are just unable to resolve your problems. I also have been using Cloudflare for nearly 10 years! I am not a paying customer, so obviously I cannot really complain that much, but I could possibly convert if the UX wasn't so painful.

• 
  Twiterrr (@_CanvasAndKeys) reported 56 minutes ago

  My problem with Cloudflare, it's like they never get things done down to perfection. You'd struggle with an initial build like it wasn't even tested at all.

• 
  UP-LVL.com (@uplvls) reported 56 minutes ago

  @jessepollak Can we get cloudflare on web3 half the internet and half of crypto goes down still when their down

• 
  Pareen (@pareen) reported 57 minutes ago

  @Cloudflare ok not bad

• 
  Jason Snell (@jsnell) reported 1 hour ago

  @heyjenbartel FYI your website seems to have a malware problem - I visited and got a fake cloudflare warning with some dangerous instructions to paste things into a terminal window.

• 
  John Nigroᵍᵐ (@jn) reported 1 hour ago

  @snipextt @p_naix @spaceship why does everyone shill cloudflare? you cannot change your nameservers. You save 9 cents to have a fundamentally broken domain name.