Amazon Web Services status: access issues and outage reports

Problems in the last 24 hours

The graph below depicts the number of Amazon Web Services 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 Amazon Web Services users through our website.

• Website Down (48%)
• Errors (33%)
• Sign in (19%)

Live Outage Map

The most recent Amazon Web Services outage reports came from the following cities:

City	Problem Type	Report Time
Kyiv	Sign in	16 days ago
Chennai	Website Down	19 days ago
Point Pleasant Beach	Website Down	24 days ago
Little Rock	Errors	27 days ago
Atlanta	Website Down	27 days ago
Clearwater	Website Down	29 days 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.

Amazon Web Services Issues Reports

Latest outage, problems and issue reports in social media:

• 
  Ronald van Loon (@Ronald_vanLoon) reported 4 minutes ago

  Most AI pilots do not fail because the model is weak. They fail because the enterprise underneath it was never built for production AI. → Data volume → Latency → Deployment cycles → Legacy dependencies → Technical debt This is the infrastructure problem nobody is talking about. Sponsored by @awscloud. #AWSpartner

• 
  Ani (@nomad_on_run) reported 5 minutes ago

  @awscloud my ticket is open for more than 24hrs and nobody is yet assigned. I understand there is no SLA for basic support, but need assistance as i’m getting quota error even for a basic machine setup with 2 vCPUs

• 
  primate portfolio (@AgastyaSinha7) reported 7 minutes ago

  @BullTheoryio This is such a bullshit argument. Yes there are 3 customers but does he realize they are just gateways for AI access? Amazon AWS, Google GCP and Microsoft Azure buy the GPUs and install them in their data centers. Then these data centers are made available to small and bug enterprises via Cloud services? The alternative would be every individual and company buys their own servers and racks and build software on top for provisioning/security/resource allocation. This is slow and lot of overhead. But the demand is so strong that companies will go down this route if the cloud providers cannot service their needs and are not eating up every GPU Nvidia is making. Also, Nvidia is expanding the field. Enter project Stargate, Oracle, Neoclouds, consumer friendly GPUs.. Like wtf is this concentration argument. The cutting edge GPUs are not a direct consumer device

• 
  If Beli (@beli_if) reported 10 minutes ago

  @awscloud Until your services are not working for 4 hours, right ?

• 
  Wagecuck Liberator (@WagecuckR) reported 14 minutes ago

  @chainlink @amazon @awscloud Slow news day

• 
  DarkSoulja3323 (@DarkSoulja3323) reported 19 minutes ago

  I had this exact problem with a Oled monitor i ordered @amazon @awscloud . Fix your **** Amazon!!! Have severe punishments for these people so they never do this ****. The driver got out with my orderer, and just got right back in the van. I said what are you doing? He said, i have to go pee so i was in a hurry? Wtf kind of Bullshit excuse is that while delivering something for someone thats %600

• 
  Drew Tewell (@drewtewell) reported 22 minutes ago

  @DoorDash Is there an @awscloud outage? #DoorDash isn’t working.

• 
  Duke of Idanre Kingdom (@Alexander474335) reported 26 minutes ago

  @AWSSupport Once this is fixed, I'll return my subscription, but for now, goodbye 👋. I have a critical custom MCP server that works on all agents I have tried except Kiro. Kiro is great. Just fix the mcp server issue.

• 
  Teri Radichel #cybersecurity #pentesting (@TeriRadichel) reported 26 minutes ago

  I’m using @anthropic opus 4.8. With @awscloud Kiro. I have rules in my README that say only use Bash. But it ignored my instructions and said sorry I used Python after using a lot of tokens to do something I didn’t want it to do. I feel like I should get those tokens refunded: this is the kind of thing that unexpectedly wastes money and time and tokens unnecessarily. I can’t make the instructions any clearer. Stuff like that is so annoying. But I’m still writing code way faster. The problem is I got a $20 plan and used it up in a few hours. I tried switching back to 4.6 but seemed like the same results for code written. It’s like everything got more expensive after the new model showed up. Or maybe I really was just making that big of a change. I had to just up the subscription to $40. I’ve already spent a bunch of money on other subscriptions and got incorrectly billed $200 in tokens I couldn’t use this month (working on getting that refunded). So you can do the math. $40 per day x 31 = $1240. But people are using multiple agents. I could spin up an army of agents to write code for me if it didn’t cost so much. Who is going to win this game? As a single developer I’m seeking ways to reduce token usage (or whatever the billing metric is) and mistakes by agents like this one feel so unfair. It clearly was not what I asked for. Whether I fix it or the agent notices its mistake after the fact and fixes it without telling me it’s still going to cost the same. Bottom line is that I can produce a lot but it also costs me a lot and so far I haven’t found a way around that besides writing deterministic code. I have to get the thing built I want to make me money before I spend it all on AI tokens. But right now the agent is making mistakes and getting stuck on things I’m doing that I guess are unique and the agent hasn’t seen too much before. It writes code and can’t figure out what it did wrong, AWS shared networking with VPC endpoints and organizations delegated admins and such I’m getting around that (without giving up full access to my cloud account) by creating verification queries and code but then the agent doesn’t follow the existing file structure and uses the wrong language without at least asking…*sigh* Also it seems like the verification code is burning tokens like a wildfire out of control. Not sure how this is going to work out. I hope there’s not some issue like someone else siphoning off tokens behind the scenes or some other weird problem like the previous nerf issue.

• 
  AngelaY (@angelayoho1) reported 26 minutes ago

  @WNBA @awscloud Is this the same company that has down your atrocious scheduling?

• 
  Rich Peter (@peterli34923561) reported 28 minutes ago

  $AAOI --- In late March and early April, $AAOI issued consecutive announcements revealing it had won a massive $71 million new order for 800G single-mode optical transceivers from a top North American internet giant (widely speculated to be Amazon AWS or Microsoft). Critically, management confirmed the first batch of 800G products entered volume shipment in Q1, marking the company's official entry into the world's highest-end AI optical communications supply chain. On April 17, $AAOI announced it will expand its manufacturing footprint to 900,000 square feet in Pearland, Texas via acquisition and leasing of adjacent facilities. Exiting Q1, its 800G transceiver monthly production capacity reached 100,000 units — this expansion is explicitly built to meet exploding order demand from AI customers. On April 29, the State of Texas awarded $AAOI $20.85 million in semiconductor innovation fund grants to support its domestic advanced photonic chip manufacturing in Sugar Land. This not only eases significant R&D and facility buildout capital pressure, but also underscores its strategic position in the U.S. domestic manufacturing ecosystem. 1. The Ultimate AI Compute Bottleneck: The Copper-to-Optical Upgrade (800G / 1.6T) Global data centers are currently undergoing a generational upgrade supercycle, shifting from 400G to 800G and even 1.6T optical transceivers. AAOI is one of the most direct beneficiaries of this hardware refresh cycle. As AI chip compute power from giants like NVIDIA grows exponentially, traditional copper transmission can no longer support the massive, ultra-high-speed data exchange required inside hyperscale data centers. Optical transceivers — devices that convert electrical signals to optical signals — have become the critical throughput bottleneck for AI server clusters. 2. Vertical Integration Breaks Gross Margin Ceilings Unlike many transceiver vendors that only perform final assembly, AAOI has in-house R&D and manufacturing capabilities for photonic (laser) chips. Competitive advantage: Amid industry-wide supply shortages, owning upstream chip capacity means no supply bottlenecks. And as scale effects kick in from higher shipments in H2, its previously criticized high fixed costs will be heavily diluted, with Non-GAAP net income expected to swing rapidly to profitability in Q2 and Q3 this year. 3.3. Dual Engine Growth: CATV Broadband + Data Center Beyond AI data centers, AAOI's legacy stronghold — cable broadband networks (HFC/CATV) — is also entering a cyclical recovery. On May 12, the company announced a deep partnership with U.S. broadband giant Mediacom to roll out its full DOCSIS 4.0 network upgrade program. This stable legacy business provides a solid base of recurring, predictable cash flow.

• 
  Eduard Lugo (@eduardlh) reported 28 minutes ago

  @AWSSupport @AWS 🚨 URGENT: Account 8366-7457-5282 suspended due to unauthorized 3rd party access. We CANNOT access console to complete required remediation steps. Support case 177899971400371 open 48hrs+ with no resolution. Deadline June 16. Business fully down.

• 
  dolla bill (@BillBah) reported 30 minutes ago

  @IAGovernor @awscloud @Cognizant Cognizant is a terrible company. The “savings” will cost more in the long run.

• 
  saura (@its_saurbh) reported 31 minutes ago

  @awscloud finally a sane take, after AI decided the best way to fix an issue is to "delete and recreate" live production environment

• 
  🇿🇦CN - Cracking News🇵🇹 (@walter_mashala) reported 33 minutes ago

  @awscloud @luis_usaa It’s good but will only server those who could afford using them, the world has more poor people than rich the people therefore it will work for the few.🤔

• 
  Teri Radichel #cybersecurity #pentesting (@TeriRadichel) reported 37 minutes ago

  I’m using @anthropic opus 4.8. With @awscloud Kiro. I have rules in my README that say only use Bash. But it ignored my instructions and said sorry I used Python after using a lot of tokens t do something I didn’t want it to do. I feel like I should get those tokens refunded: this is the kind of thing that unexpectedly wastes money and time and tokens unnecessarily. I can’t make the instructions any clearer. Stuff like that is so annoying. But I’m still writing code way faster. The problem is I got a $20 plan and used it up in a few hours. I tried switching back to 4.6 but seemed like the same results for code written. It’s like everything got more expensive after the new model showed up. Or maybe I really was just making that big of a change. I had to just up the subscription to $40. I’ve already spent a bunch of money on other subscriptions and got incorrectly billed $200 in tokens I couldn’t use this month (working on getting that refunded). So you can do the math. $40 per day x 31 = $1240. But people are using multiple agents. I could spin up an army of agents to write code for me if it didn’t cost so much. Who is going to win this game? As a single developer I’m seeking ways to reduce token usage (or whatever the billing metric is) and mistakes by agents like this one feel so unfair. It clearly was not what I asked for. Whether I fix it or the agent notices its mistake after the fact and fixes it without telling me it’s still going to cost the same. Bottom line is that I can produce a lot but it also costs me a lot and so far I haven’t found a way around that besides writing deterministic code. I have to get the thing built I want to make me money before I spend it all on AI tokens. But right now the agent is making mistakes and getting stuck on things I’m doing that I guess are unique and the agent hasn’t seen too much before. It writes code and can’t figure out what it did wrong, AWS shared networking with VPC endpoints and organizations delegated admins and such I’m getting around that (without giving up full access to my cloud account) by creating verification queries and code but then the agent doesn’t follow the existing file structure and uses the wrong language without at least asking…*sigh* Also it seems like the verification code is burning tokens like a wildfire out of control. Not sure how this is going to work out. I hope there’s not some issue like someone else siphoning off tokens behind the scenes or some other weird problem like the previous nerf issue.

• 
  Joyfulwood (@JoyfulWood) reported 39 minutes ago

  $GXAI 1.47 Dip SL under 1.39 Ran big on news this morning, brought down to support at 1.45. Trying for a second run. PT 1.75/1.89/2/2.15/2.3+ Gaxos Secures Additional Funding from Amazon AWS to Accelerate AI Sales Platform Development

• 
  CHRISTOPHER BLAZ✨ (@Blaz_Dao) reported 43 minutes ago

  Walrus 🦭 and Walrus Memory explained in a lay man's understanding. Let's dive in: 1. What is" Walrus🦭" ?? Firstly, I want you to think of Walrus as a gaint decentralised hard drive built on Sui protocol or ecosystem. Walrus is trying to become the decentralized version of cloud storage services like Google Drive, Dropbox, or Amazon S3, but built for Web3. In a simple analogy, think of it like this; Google Drive vs Walrus With Google Drive, your files are stored on Google's servers and Google controls the storage. If Google removes a file or service, you're dependent on them. But, With Walrus🦭 it's a different case as your files are split into many pieces. Those pieces are stored across many independent storage providers in a cheap manner as no single company controls all of your data. And the most fascinating thing is as long as enough storage providers remain online, your files can be recovered whenever you want. 2. What is "Walrus memory"?? In plain English, Walrus memory is simply the storage space used to keep data on the Walrus network. That data can be:Images, Videos, Documents, NFT media, AI datasets, Website files, Backups etc. Why does it matter? Imagine a viral meme image on Sui. Normally, the blockchain only stores a reference to the image because storing the image itself would be too expensive. Most blockchains are good at storing transactions but terrible at storing large files. @WalrusProtocol is designed to store large amounts of data cheaply while remaining decentralized. ~BlazCares

• 
  RationalJames2776 (@RationalJames76) reported 43 minutes ago

  @IAGovernor @awscloud @Cognizant Your systems will be spaghettified and will need 10x the staff to support and fix. Source: seen it irl

• 
  Astro Kanu (@AstroKanu) reported 46 minutes ago

  @Suraj_Lande_ @awscloud is running a scam in India- payments taken but services not given at multiple levels! It’s time to raise this issue ! We as paying clients cannot be treated this way ! #awsscam

• 
  Jason Sulano (@jmoney42777) reported 48 minutes ago

  @amazon @awscloud Run over my mailbox and drag this issue out for months with no resolution. Thanks Debashree Bose

• 
  AbendStarr (@AbendStarr) reported 48 minutes ago

  @IAGovernor @awscloud @Cognizant Just out of curiosity will the citizens of Iowa have their taxes go down bc of these "savings"?

• 
  eve (@evevrcx) reported 49 minutes ago

  @awscloud @Twitch slow down

• 
  Munawwar Firoz (@munawwarfiroz) reported 50 minutes ago

  @AWSSupport I don't care anymore. These issues were reported years ago by others but never fixed.

• 
  Kingdom of ElviSpeare (@ElviSpeareTV) reported 51 minutes ago

  All down: DoorDash. Spotify. GoDaddy. Verizon. Uber Eats. Amazon AWS. Did someone cut the under water wires half way around the world? Possibly others.

• 
  Astro Kanu (@AstroKanu) reported 54 minutes ago

  @awscloud @AWSSupport if u don’t resolve this on urgent basis I’m going to start telling my clients who have paid for my services- which are down because of you to start writing to u to claim a refund - you cannot disrupt a live project and be so causal about this #aws #awsscam

• 
  Conor O'Neill (@conoro) reported 58 minutes ago

  @AWSSupport @AWSAI Attempting to use the new Gemma 4 in eu-central-1 via Curl. New error I've never seen before. "error":{"code":"access_denied","message":"Berm is not enabled for this account" What's Berm? model: google.gemma-4-26b-a4b

• 
  vasik mansuri (@vasik_mansuri) reported 58 minutes ago

  @AWSCloud @AWSSupport We've been waiting for support for more than 3 hours today and spent hours yesterday as well. Our account was suspended after a password reset issue. Password has been reset multiple times, support case updated repeatedly, but still no response.

• 
  🇺🇸 End H1B F1 OPT TN E3 O1 L1 I1 H4 all NIV 🇺🇸 (@slmimorgan) reported 1 hour ago

  @IAGovernor @awscloud @Cognizant You need to stop down you treasonous monster. You are elected to serve we the American people. Not Cognizant or Indians. Shame on you.

• 
  Jeff Grigg (@JeffGrigg1) reported 1 hour ago

  @2fast2feeless @AlerixCodes @awscloud But also, you're 100% responsible for *FIXING* all the problems it creates, too! 😱