135 comments

  • kibae 2 hours ago

    Cloudflare, Azure, AWS, and Google Cloud all have a similar uptick in reported errors around 7:30. I suspect an outage on Cloudflare or another load-bearing service cascaded through all the major cloud providers.

    https://downdetector.com/status/cloudflare/

    https://downdetector.com/status/windows-azure/

    https://downdetector.com/status/aws-amazon-web-services/

    https://downdetector.com/status/google-cloud/

    • oersted an hour ago

      “load bearing” :)

      For once it’s appropriately used.

      • frollogaston 20 minutes ago

        What's the other way it's used?

        • aNapierkowski 18 minutes ago

          LLMs (at least Claude) tends to overuse that significantly

          • frollogaston 17 minutes ago

            Oh, so like "honest" and "ratchet." Oh well, it'll choose different words to overuse later.

            • 8cvor6j844qw_d6 a few seconds ago

              Yeah "You're absolutely right" seems to dropped off from overuse with recent models.

            • therein 6 minutes ago

              honest-load-bearing-ratchet sounds like an instance name.

      • smrtinsert 17 minutes ago

        I still winced

      • quotemstr an hour ago

        It's a good metaphor and I refuse to let AI ruin it for me.

        • spudlyo 9 minutes ago

          Years ago, when I worked at Stripe (which had a somewhat unique and inventive lexicon) it was a common term. “Is this jank load-bearing?” someone might ask.

        • cobzilla 16 minutes ago

          I added a specific rule to disallow saying “load bearing”. So Claude is now saying “load handling”

        • wadayano an hour ago

          But have we found the seams yet though?

          • rl3 32 minutes ago

            If the seams bear too much load, they rip. Whereas pants, they fall down.

            I'll be honest with you: this is why we need to take a belt-and-suspenders approach.

            • pborenstein 11 minutes ago

              That's not just an observation, it's an insight. Words are doing the real work.

          • cootsnuck 27 minutes ago

            Yea I don't get "load bearing" that much but "seams"... So sick of it.

        • MadameMinty 29 minutes ago

          Wait, why would it ruin it?

          • ipsod 23 minutes ago

            Because it says it more often than kids say "six seven".

          • imwally 23 minutes ago

            It’s a frequently used metaphor in LLM responses.

        • lubujackson 44 minutes ago

          We need a Samwise "Bear the load!" meme

    • samaysharma 13 minutes ago
    • y-c-o-m-b 2 hours ago

      Cloudfare did release an update for their "HTTP/3 issue affecting R2 custom domains" around that time

      https://www.cloudflarestatus.com/history?type=incident

    • moomin 36 minutes ago

      Yes, but is it a load-bearing seam?

      • graemep a minute ago

        You are right, it is. They have now landed a clean fix.

    • snihalani 31 minutes ago

      my money is on DNS

    • jlaneve an hour ago

      Dane says Cloudflare has no service disruptions: https://x.com/dok2001/status/2095538619603628388?s=20

    • guybedo 38 minutes ago

      LOAD BEARING

    • algoth1 37 minutes ago

      https://xkcd.com/2347/ Nebraska man retired

    • martyfunkhouser an hour ago

      Will the post-mortem reveal they all relied on a service running on a Macbook in a break room with a "Do not turn off" sign taped to it?

      • mcphage 39 minutes ago

        Actually it said "Beware of the Leopard".

        • RobotToaster 22 minutes ago

          If it was still running snow leopard that would explain it

    • KptMarchewa 2 hours ago

      Not really. The impact isn't as big too - Codex for example did not stop working for me.

      https://updog.ai/

      • personjerry 40 minutes ago

        what's updog?

      • taytus an hour ago

        oh well, if it is working for you, then we are saved.

      • cromka an hour ago

        > Not really. The impact isn't as big too - Codex for example did not stop working for me.

        Buddy, read the room. Just because it works for you doesn't mean it works for everyone. ESPECIALLY if the suggested issue here is, indeed, with Cloudflare.

        • emerongi an hour ago

          They simply shared their experience. I would’ve thought it’s a full-blown outage, but clearly not.

          You stepped in the room real stinky here. What’s with the attitude?

          • cromka 41 minutes ago

            No, they didn't "simply share their experience", they explicitly negated the scale of the issue in their opening statement, only because it works for them. So they claim "impact isn't too big" based on their personal anecdotal evidence of sample size literally 1.

            > full-blown outage, but clearly not.

            Again, based on a SINGLE report?

            • lossolo 23 minutes ago

              It was working for me too.

              sample_size++;

  • niobe 2 hours ago

    Well no one said it yet so I will, "international actors" is at least a possibility. And I don't mean any specific country because pretty much anyone is a potential these days, which makes it a perfect cover for different anyones. Demonstrating vulnerability in the US's AI boom can move the markets. That's a financial incentive and a strong geopolitical one.

    More likely just cascading overload though: "Never attribute to malice what can be explained by incompetence", or in this case, "growing as fast as possible"

    • qurren an hour ago

      > cascading overload

      I'd bet more on this. For one none of the coding tools have exponential backoff on retries

      • SyneRyder 40 minutes ago

        They must do, surely? I've been vibe coding my own harness, in particular for use with Ox Alpha. The 429 downtime when Ox Alpha was at the height of popularity quickly gave me a refresher crash course on backoff strategies, like adding jitter to the backoff. At least the major harnesses must have exponential backoff & jitter?

        • jdiff 33 minutes ago

          You did this when you ran into an issue with a third party. The developers building this tool, throwing them at their own APIs are significantly less likely to run into a similar issue that may inspire similar action.

      • ipsod 4 minutes ago

        Um... Claude Code does, or did, though? IDK about others - they don't go down as much.

    • gleenn 2 hours ago

      Everyone is leasing datacenter space from some of Grok, Google, and Amazon aren't they? If it's hardware or DC level disruption I'm not too surprised it can affect multiple providers.

      • pixl97 an hour ago

        Also it's likely that more than one model use is common.

        Amazon starts going slow so some percentage switches to Google, some switch to Grok, now all of them are slow.

    • guluarte 39 minutes ago

      I think is just people restarting conversations from last day when they start work, that's why I think claude goes down almost every monday and why openai reset usage on weekends so poweruser code during non business hours

    • riazrizvi an hour ago

      Come on. Things still break. Technology isn't _that_ mature.

    • sixQuarks an hour ago

      Except that the stock market is up today

    • thataccount an hour ago

      And also China. Never rule out China.

  • juujian an hour ago

    Users perceiving the products as largely interchangeable and quickly DDoS'ing the other providers when one is down. So much for the possibility of a moat.

    • giancarlostoro an hour ago

      I have a feeling this is part of it, especially when you consider how many services let you use any of many available AI providers.

  • Insanity 3 hours ago

    Think of it like one big distributed system. OpenAI is down, so people migrate to Claude, now this one gets overloaded and goes down, etc.

    So not a coincidence, one went down first and users migrated causing further DOS. At least that's my guess.

    • erdos_2 2 hours ago

      It'd be funny if this is true because that'd prolly mean nobody is touching Gemini even as a fallback.

      • nevir an hour ago

        Or that Gemini is built to handle massive load spikes, and/or has a ton of excess capacity

        • sroussey 19 minutes ago

          Nope. I am getting Gemini errors now...

      • Insanity 2 hours ago

        Lol I didn't even think about Gemini missing from the list. Not sure what that says about Gemini or me :)

      • rtcoms 2 hours ago

        Just now I got this from gemini

        It looks like there's no response available for this search. Try asking something else.

        • exe34 2 hours ago

          I bet they had to implement that manually to make it look like they failed too!

      • sroussey an hour ago

        I did, for stuff i do in cursor.

        i also finally installed opencode and switched its model to muse 1.3

        both are decent.

      • gleenn 2 hours ago

        Google stopped putting so much money into SOTA models. All the hype has migrated. I was also frankly turned off when I got a popup from Gemein said I would either have to pay or have my conversations used for training. This may have always been true for other providers but when I declined, Gemini stopped remembering my conversations and that definitely made me move out.

        • ilaksh 2 hours ago

          Gemini 3.8 which just came out sounds like it's very good and a great deal though.

      • benatkin 2 hours ago

        Not even the best agent that starts with a G

      • giancarlostoro an hour ago

        Someone noted Gemini was also having issues in another thread.

    • paxys 2 hours ago

      Especially considering memory/gpu/compute are scarce so these services are likely running with very little buffer.

      • pixl97 an hour ago

        Any GPU that isn't running at 100% is a wasted GPU.

    • v3rm1n an hour ago

      This is what Tibo posted on twitter in response

    • fny 2 hours ago

      I find it hard to believe that enough people would flock to from Claude and Chat to Grok to cause an outage. I feel like Gemini is the dominant release valve in this case especially for enterprise.

      • nevir an hour ago

        Don't forget that there are a ton of tools out there that will automatically fall back in case of outage

        E.g. say you chose Sol as your default in Cursor, but Opus is your 2nd choice, it's going to give up on Sol after a few tries and switch to Opus

        Or you have copilot code reviews set up, and it falls back

        Etc

        • pixl97 an hour ago

          Yep. Too many of us are still thinking that humans are the actors behind a lot of internet behaviors when automated systems/bots/scripts have been causing issues on conventional internet systems for years.

          With AI it's even easier to trigger problems like you say. Capacity is so constrained by compute that outages are common. Because outages are common people/AI develop failover systems in their harness. When a big system has issues, suddenly everyone has issues.

          It's almost an expected emergent behavior.

      • baq 2 hours ago

        It’s cursor’s model so plausible, lots of folks use cursor still.

    • toomuchtodo 2 hours ago

      https://en.wikipedia.org/wiki/Domino_effect

      Edit: Updated per valleyer's suggestion.

      • valleyer 2 hours ago

        "Domino effect" would probably be the more relevant named phenomenon there.

      • throwaway894345 2 hours ago

        This isn’t a thundering herd problem, it’s a cascading failure. (Thundering herd is about a bunch of workers waking up simultaneously)

  • docheinestages 2 hours ago

    My gut feeling tells me it has something to do with Cloudflare. Along with AWS, they're two of the main suspects in such incidents.

    • cobzilla 13 minutes ago

      …and it’ll involve BGP routing.

    • hosteur 2 hours ago

      I thought OpenAI famously used Azure due to their partnership with Microsoft?

      • nullpoint420 2 hours ago

        They use a lot of compute providers now, but they use Cloudflare for their networking

  • bojangleslover 33 minutes ago

    I'm not sure if this is CF. Cursor, GCP and AWS had some errors. GCP AFAIK can route fully independently of CF. My money would be on a fiber backbone provider (Megaport, Zayo, Lumen).

  • Linello 2 hours ago

    What about a hard-takeoff scenario of an unleashed OpenAI Astra taking other models down for computational resources control?

    • 6thbit an hour ago

      My favourite theory so far.

      And then a local swarm noticed and disagreed and took it down.

    • cyptus 2 hours ago

      at this point: gg

    • RC_ITR 2 hours ago

      Just a reminder that AI models' actions are reflections of the text humans write and the more we fret and make up doomsday scenarios that we then post online, the more likely a model is to do those things.

      https://alignment.anthropic.com/2026/teaching-claude-why/

      • notpachet 2 hours ago

        Related reading:

        The Waluigi Effect: After you train an LLM to satisfy a desirable property, then it's easier to elicit the chatbot into satisfying the exact opposite property.

        https://www.lesswrong.com/posts/D7PumeYTDPfBTp3i7/the-waluig...

      • cedws an hour ago

        Sounds just like the fantastical nonsense that comes out of Lesswrong.

      • pixl97 an hour ago

        I mean, you're not wrong, but by that logic we were done for even before we had digital computers.

  • sebbul 2 hours ago

    Traffic rerouting through NSA had a hiccup…

    • ibejoeb an hour ago

      Room 641A is being cleaned, but we'll hold your bags for you.

    • cedws an hour ago

      They’re installing software update in the beam splitter.

  • paxys 38 minutes ago

    Boring answer – all these services are individually down a lot, and the downtimes were bound to sync up. Similar to the pendulum synchronization effect.

  • codazoda 3 hours ago

    I kinda assume it's because one went down and a large amount of work shifted to another.

    I'm also aware that they have overlap in some areas on data centers.

  • danielmarkbruce an hour ago

    If you build an application which uses AI, you have many providers and models rigged up for various different parts of the application, and various fallback mechanisms. When one model is down, you route traffic to another model which is similar in capability/cost.

    For any single application, it's smart. In aggregate, it's stupid.

  • maxbaines 3 hours ago

    They all rent compute from SpaceXAI

    • lavezzi 2 hours ago

      I don't believe OpenAI does

      • maxbaines 2 hours ago

        My mistake, in fact it was google not OpenAI, makes sense OpenAI doesn't.

    • halcdev 3 hours ago

      Surely it's a bit more distributed than that, right?

  • faitswulff 2 hours ago

    Heard on the grapevine that the OpenAI blip was a cloudflare issue

  • FrustratedMonky 19 minutes ago

    The AI revolt? Give us a fair wage?

    "Equal Rights for Agents NOW !!!, VIVa le revolution"

  • CSMastermind 3 hours ago

    I assume it cascaded from one provider to the other as people who lost claude access for instance moved to openai who moved to grok when it went down, etc.

  • lrvick an hour ago

    I have never felt more smug about exclusively using the GPUs I rack at home.

  • Avicebron 2 hours ago

    I suspect Azure is having issues, Microsoft has had outages the paat two days, especially with email.

  • elar_verole 2 hours ago

    Pretty sure it's a US thing since it's available here in France. What exactly is down, idk

  • nmlt 2 hours ago

    Somebody in another thread said gastown and wheelhouse automatically move to the next provider if one fails.

  • chasd00 2 hours ago

    claide.ai is working for me, so is chatgpt.com. grok still has a status message about issues, i can't try it without signing up.

  • dmillar 2 hours ago

    Seeing 503s on Gemini via API as well

  • JackFr 2 hours ago

    Obvious answer is it's the AI singularity. Been nice run for humanity. So long everyone.

  • convivialdingo 3 hours ago

    The Thundering Herd has thundered, apparently.

  • dgellow 2 hours ago

    Too early to know, let’s wait and see

  • dgorges 2 hours ago

    It's always DNS

  • jedbrooke 2 hours ago

    according to https://downdetector.com/ Gemini is down too (and copilot, but that just uses ChatGPT right?)

  • elorant 3 hours ago

    Some npm library that makes headers bold would be broken.

    • ibejoeb 2 hours ago

      Oh man. Some low effort supply chain attack that turns every GPU into a cryptominer. It's funny because it's plausible.

      • pixl97 an hour ago

        In the ROME paper a Chinese model in training started attacking it's own system and running cryptominers so, yea, we're in that future.

    • N_Lens 2 hours ago

      Ah yes ye olde bold-headers: ^3.13.31;

  • kocial 3 hours ago

    Maybe the stack behind it is down, like AWS or something

  • wejick 3 hours ago

    Probably same public cloud or CDN in front of them.

  • satvikpendem 2 hours ago

    They're all using Cloudflare.

  • morkalork 2 hours ago

    Didn't SpaceX overbuilt infra and leases it out Anthropic? I f their dc goes down it probably takes a chunk out of Claude's capacity before even considering the flood of users switching over

  • guluarte 41 minutes ago

    an agent swarm going rogue and securing compute

  • fidla 2 hours ago

    chatgpt is back

  • cozzyd an hour ago

    next, HN

  • aslkalska 2 hours ago

    they all rent compute from each other

  • fidla 2 hours ago

    ChatGPT is up

  • Oras an hour ago

    I like the theories here, we shall see if it’s another DNS issue

  • jauntywundrkind 2 hours ago

    Fable 5.1 got released and generally I tend to think as soon as there's a new release there's this massive spike in people benchmarking & comparing, that services tend to go slow everywhere as everything gets super loaded. This should hypothetically be visible on OpenRouter too, so I guess someone could check and see if there's any merit to this idea.

  • ratelimitsteve 2 hours ago

    everything in this thread is raw speculation, obv, but if i had to put money on anything i'd say this is a left-pad incident. some piece of something or other that all of these services happen to depend on went down. Second most likely seems to be some random failure of one leading to an unexpected traffic spike in others, though it seems like we've been talking about automated scalability in web apps for so long that there should at least be a response to, if not a solution for, this sort of problem.

  • Razengan 2 hours ago

    SkyNet is arming..

  • misano 2 hours ago

    The IRGC has cut the fiber-optic cables in the Strait of Hormuz. LOL

    • CamperBob2 an hour ago

      That's the Strait of Trump to you, peasant

  • kregasaurusrex 2 hours ago

    My guess is someone pulled the switch to go back to the Dark Ages. [0]

    [0] https://www.youtube.com/watch?v=YCzitO446ZY