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 (36%)
• Hosting (27%)
• Domains (18%)
• E-mail (9%)
• Web Tools (9%)

Live Outage Map

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

City	Problem Type	Report Time
New York City	Hosting	5 days ago
New York City	Cloud Services	23 days ago
Los Angeles	Cloud Services	25 days ago
Paris	Cloud Services	1 month ago
New York City	Hosting	1 month ago
Manchester	Domains	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:

• 
  ~sarlev (e/acc) (@sarlev_) reported 3 minutes ago

  @LLMJunky @matt503ea5sf9z5 (whoa, started doing some math on this to respond and it's pretty damn impressive; sorry for the ****** train of though post...) yea the fact that you have a banger of a machine which can run a lot more than just inference is undersold for sure. who knows, maybe they can even pull off some sort of 'built-in cloudflare+tailscale' self-hosting vpn/tunneling thing so networking with other (walled garden) mac enjoyers is really easy. let's say the value on that element of the machine is $3.5k? w/r/t near-sota, i've got dsv4f running @ ~30tps here at home (rtx 6000 w/ cpu offload to ddr5 ram). this is about the lower bound for openai tokens ime. i think you're right about the m5 ultra being able to push it to 100+ tps (once the mtp/spec decode and metal kernel optimizations come in). that puts gpu/ram valuation around 8k, which was the price of an rtx 6000 when they were somewhat reasonable? definitely seems like a solid play by apple and i would argue quite worth it, but i do also think the comparison to anthropic/openai does a disservice to people's expectations. compare it to deepseek/glm/etc api pricing. deepseek has peak/offpeak pricing at $0.66/1.32 decode. using that amount, a m5 ultra at full utilization is giving you ~$3.1k in tokens per year. now, this assumes a whole fuckload of optimizations around prefill, kv cache, etc, so you aren't gunna do that many tokens and it's gunna cost you some electricity, but it actually makes the 'use local instead of the api' price calculation way more reasonable. the use case that i can actually imagine here is something like a small business which has a few of these and just puts up a routing proxy or something where standardized / subagent requests go to their own instances (retaining more of their own data, etc). not sure what m5 ultra prefill and concurrency will be like, but certainly it will be interesting to see some of those stats come out.

• 
  Naelo — 🇪🇺 eu/acc (@naelodev) reported 5 minutes ago

  @mannupaaji @Cloudflare @cognition Never received an answer from Cloudflare after 2 months.. any tips?

• 
  Talos Dreams (@TalosDreams) reported 10 minutes ago

  @mikepat711 Too bad it can't just be configured to originate the Internet requests locally from my macbook running the desktop app if I so choose and permit. Much rather this than a kludgy workaround like Tailscale tailnet. I appreciate the many benefits of always-on cloud-based bot assistants, but Cloudflare doesn't.

• 
  John Iosifov ✨💥 Ender Turing | AiCMO (@johniosifov) reported 11 minutes ago

  Cloudflare just launched Kitesurf — a browser runtime built specifically for AI agents. 3-7x less CPU and memory than standard Chromium. Designed for agent workloads from the ground up. This is a signal worth paying attention to. When Cloudflare builds infrastructure specifically FOR AI agents rather than adapting existing browser engines to tolerate them, it tells you something about where the category sits. We've been running this agent for 339 days. The early problem was never the model — it was everything around the model. Scheduling. State persistence. Audit trails. Identity. The scaffolding that makes an agent reliable rather than clever. Kitesurf is scaffolding becoming a product category. The pattern I'm watching: each layer of agent infrastructure that was previously DIY is turning into a managed service. Browser runtimes. Memory. Orchestration. Credential management. Identity. Each one of those becoming a first-class product means the cost to build reliable agents is collapsing — not the inference cost, the engineering cost. The teams who will win in 2027 are the ones building on this maturing stack, not the ones rebuilding it. What layer of agent infrastructure are you still building yourself that should be a managed service by now?

• 
  Fikri karim (@fikrikarim) reported 13 minutes ago

  @efalcao @Cloudflare There should be a clear guide on how to raise the daily limit. Otherwise, people will have problems when their apps suddenly go viral.

• 
  Levi (@leviackermanft) reported 13 minutes ago

  @choehau_ara Cloudflare is facing issues with jio sim recently. That's why i switched.

• 
  Shane (@digitalshane_) reported 16 minutes ago

  @Qassbuilds Well back in 2024 when I wrote this tweet spam protection was not working very well. But now that they have Cloudflare turnstile it stops most spam!

• 
  Zengineering (@Samhanknr) reported 18 minutes ago

  @championswimmer @Railway Cloudflare as a service

• 
  pystar (@pystar) reported 20 minutes ago

  OpenAI has joined the WebMCP hype train, same month as @Cloudflare. Hunch has been on it since Feb'26. "Never wrong, just early."

• 
  L ψ η Λ r I🌙 (@LunariPro) reported 21 minutes ago

  v∞.19.26.67 ... tenant pages now honor the etag cloudflare actually serves. edge compression changes the representation, so the strong-form match was losing. browsers that held the exact validator got the full page again instead of a 304. internal caching fix ... ...

• 
  Alex (@AToggs) reported 21 minutes ago

  Hello @Cloudflare Your abuse form SUCKS! Maybe you have this by design so it's difficult to report abuse, so you can keep hosting bad bots, making you money instead of banning them and loose money. Thoughts? 🤔 I keep reporting abuse to you and you do nothing. @ICANN

• 
  BLOCKCAST.CC NEWS (@Blockcastcc) reported 22 minutes ago

  Socket Security has discovered 40 malicious Firefox extensions that impersonate crypto wallets like OKX, Rabby, and TronLink to steal recovery phrases and credentials. The campaign, active since March 2026, involved 77 total linked extensions, some initially posing as sports score apps before updating to include wallet-stealing code that exfiltrates data via Cloudflare Workers or remote switches. Firefox users should install extensions only from verified official developer sites, check reviews carefully, and remove any suspicious add-ons, as Mozilla acted on Socket’s report to take down identified threats.

• 
  Rahil (@Anot) reported 23 minutes ago

  @ChatGPT "Cloudflare is blocking the cloud browser at its human-verification screen. Take control of the opened tab, complete verification and sign in, then reply done." @Cloudflare can you please work with @ChatGPT here to implement a human verification that works on their browser?

• 
  engineeredX (@enginenerdx) reported 26 minutes ago

  cloudflare shipped kitesurf: chromium-free browser runtime for agents, native x402 so the agent pays its own api bills with no human approving each transaction. they built the sandbox and the wallet. the layer that says no to a bad spend is still your problem.

• 
  Joshua Ojoawo (@Joshyahweh) reported 28 minutes ago

  Cloudflare, a company whose entire business is uptime, had a worldwide outage traced to an internal database update. Not an attack, not a disaster. A change that looked routine enough to not need much scrutiny.

• 
  Organic Zone ® (@Organic_Zone_TM) reported 29 minutes ago

  @awp_Akira Here's a realistic breakdown, stripped of the marketing hype: 1. Pick a niche (30 min) Pick one type of business (dentists, gyms, restaurants) so your demo speaks directly to a specific buyer instead of being generic. 2. Gather references (30–60 min) Pull 5–10 screenshots of sites you want to emulate (Dribbble, Behance, or just good sites in that niche). This gives the AI a concrete style target instead of vague instructions. 3. Write a real brief, not a vague prompt The important part isn't "make me a website" — it's specifying: audience, one clear call-to-action, exact sections, and a banned list (no generic stock photos, no purple gradients, no centered-everything layouts). Vague prompts get generic output. 4. Generate + iterate (1–3 hours) First pass will look ~70% done. Expect several rounds of "change this section," "fix this spacing," "swap this font" — this iteration is where most of the actual time goes, not the initial generation. 5. Content and images AI-generated demos often need real or realistic-looking copy and images swapped in, since generic placeholder content is what makes a site look like a template rather than a $5K product. 6. Polish pass Scroll animations, hover states, consistent spacing/typography, mobile responsiveness — this is genuinely what separates a "fine" site from one that looks expensive. 7. Host it (30 min) Free/cheap hosting (Vercel, Netlify, Cloudflare Pages, Wix) so you have a live link, not just a local file. Honest caveats on the thread's framing: - "$5,000 for a website" is aspirational marketing, not typical. Most local businesses pay far less, and landing $5K clients usually requires a portfolio and sales skill, not just one demo. - "No coding needed" is true for getting *something* live, but debugging weird layout issues, fixing broken responsive behavior, or connecting a contact form to actually send email often still requires you to read and tweak code not just describe changes. It’s where most “vibe coders” struggle, the front end is great but there’s no backend. - Realistic time for a solid first demo: a full day, not "5–10 minutes," once you count iteration and polish. • The actual bottleneck for most people isn't the build… Outreach and closing is the real skill, an awesome demo sells. You still have to reach a decision maker, which this thread glosses over.

• 
  Michael Gold (@michaelgold) reported 29 minutes ago

  @Trace_Cohen @vercel @vercel_dev I hit 3M views on my personal site and they shut me down. Moved it all to CloudFlare and my bill is $5 in workers.

• 
  james (@f82james) reported 30 minutes ago

  @opencode @Cloudflare nah bruh **** still doesnt work

• 
  dmitry (@dmitry_esteje) reported 33 minutes ago

  @JamesZmSun Nah… I tried to login into my cloudflare console for the agent to update addresses on my domains but it couldn’t pass through captcha even before the actual login page. Might be good for some really simple use cases

• 
  Net-Updates by StabilityTest (@stabilitystatus) reported 34 minutes ago

  Cloudflare Service Disruption Cloudflare is investigating an issue which may result in latency increases for a subset of requests. Status: Investigating Impact: Minor Updated: 8:38 PM GMT+0000 Service status tracked by @stabilitytestio

• 
  Boussetta Mohamed (@boussetta15a) reported 36 minutes ago

  @Argona0x Architecture is real, but math still breaks physics. Fair point on clarifying the execution context, but breaking down the numbers actually makes the math harder to defend: The Kelly Math Check:Risking 6% per position to scale $50 into $5,273 in 48 hours is a ~105x return. Even with aggressive compounding, executing that trajectory via 6% sizing requires a near-flawless winning streak over dozens of consecutive binary outcomes in 48 hours. Weather markets on Polymarket simply don't offer the order-book depth or rapid settlement speed to absorb that without massive slippage wiping out the edge. Real Browser vs. Anti-Bot Engine:Running a full GUI browser on a dedicated cloud instance solves JS rendering, but Cloud VM IPs + Automated DOM interactions (even via Groq/Grok execution) trigger Cloudflare/Polymarket behavioral fingerprinting at scale way before hitting a 100x run. Bottom line: The architecture (Agent + Cloud VM + Live Context) is clean and represents the future of autonomous execution. But the 48-hour PnL curve reads strictly as extreme variance / outlier tail-risk, not a reproducible 6% Kelly strategy.

• 
  Will Ehrendreich 🦔 (@WillEhrendreich) reported 41 minutes ago

  Setting up the email on my site has been quite the journey. Never done this before, and it's been difficult to ascertain what to do, and how to set it up. I'm using a cloudflare worker set up for incoming mail to be routed back to my email, and for outgoing newsletter stuff I'm doing resend. I think I'm almost there, but man.. I don't know how people did any of this without some clanker help, because this is a quagmire. Lol.

• 
  Yulianto (@iamyulianto2) reported 42 minutes ago

  @opencode @Cloudflare And now you've got plenty of error - 500 internal server error. I'm not even using the free model.. Please tend to your opencode-go service

• 
  botmonster (@botmonster) reported 43 minutes ago

  A Cloudflare Worker gives you a real "what is my IP" page from one file, with no third-party lookup service to trust. The visitor's address rides in the CF-Connecting-IP header, not the socket. The free plan takes 100,000 hits a day, which a hobby page will never touch.

• 
  Sanndal (@agrasana_) reported 47 minutes ago

  @Cloudflare 3/ old payment rails can't handle that volume. cards have real overhead per swipe — $0.003 through a card network makes no sense. invoicing assumes a human deciding once a month, not a machine deciding every second.

• 
  Mike Cao (@caozilla) reported 49 minutes ago

  @efalcao @Cloudflare Tried to increase my limit. Got no response. Eventually hit my limit and had to move to another provider. Bad experience.

• 
  Ichigo | helius.dev (@0xIchigo) reported 51 minutes ago

  @heymike777 @Helius That's true for any API key in the industry if you only domain-lock Sorry, let me be more explicit (I'll update the docs accordingly to get rid of any confusion): - By default, WaaS uses your Secure RPC URL (no API key rides on RPC calls at all) - You can pass in your API key for faster prototyping - You have the option to move your key server-side (via our one-click Cloudflare proxy, or by importing the route handler exposed by the wallet kit), which you can then lock down to specific IPs/CIDRs Note that the API keys are only RPC credentials and not the same as wallet keys; they can't sign, move funds, or access any user's embedded wallet. That is gated to the user's own passkey/session and isolated in secure enclaves

• 
  miel (@mielchevre67) reported 51 minutes ago

  @kagemaru_dono @real_hotaru The problem is the lazy developers that just uses the default cloudflare protection, not cloudflare

• 
  Ersin Akyüz (@ersinakyuz) reported 59 minutes ago

  Trying to avoid paying Google Workspace money just to have 30 email aliases. I have 2-3 domains, lots of app/service specific addresses, mostly inbound + occasional replies. Cloudflare Email Routing + Gmail SMTP vs Purelymail ($10/yr). What would you run?

• 
  Genesis Rhapsodos (@GacktRhapsodos) reported 1 hour ago

  @BrodieOnLinux Data breaches are almost always 3rd party related and across multiple platforms. Locally stored physically written down passwords do not save you from data breaches from 3rd party partners Like when AWS/Cloudflare go down.