Amazon status: access issues and outage reports

Problems in the last 24 hours

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

• Website Down (45%)
• Errors (32%)
• Sign in (24%)

Live Outage Map

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

City	Problem Type	Report Time
Hayange	Errors	19 hours ago
San Nicolás de los Garza	Sign in	19 hours ago
Miguel Hidalgo	Website Down	19 hours ago
Paris	Website Down	3 days ago
Guadalajara	Errors	4 days ago
New York City	Website Down	4 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 Issues Reports

Latest outage, problems and issue reports in social media:

• 
  Sulabh Awasthi (@sulabh_612) reported 6 minutes ago

  @GodrejAppliance @amazonIN SR no.-82340075,82526340 Amazon:404-4456525-0815564 Godrej denied any support for Amazon-suplied appliances and demanded additional service charges. It was installed in July-26 & the unit was down from 12-Aug, &they closed the tickets.Beware while purchasing godrej

• 
  Peter (@pw7fam) reported 12 minutes ago

  @DavidDTawil AI FINANCING ~~~~Circular funding (or circular financing) in the current AI boom is a web of interdependent equity stakes, credit guarantees, residual-value support, and third-party financing platforms in which capital from chipmakers, hyperscalers, and financiers flows to AI labs, neoclouds, and data-center builders, who then recycle much of it back into purchases of the same suppliers’ GPUs, cloud capacity, or infrastructure.15 It is essentially modern vendor financing at extreme scale. NVIDIA is the clearest node: it takes equity in customers (OpenAI, Anthropic, CoreWeave, etc.), provides or arranges credit support/guarantees, and helps structure debt vehicles collateralized by its own chips; those customers then buy more NVIDIA hardware or lease capacity that uses it. Hyperscalers (Microsoft, Amazon, Oracle, etc.) play similar roles with their cloud and equity commitments. The August 10, 2026 NVIDIA announcement of memorandums with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR formalizes this further: independent platforms designed to mobilize >$500 billion of third-party capital over time for AI-factory infrastructure, with NVIDIA optionally providing residual-value support up to 25% on a project basis. The chips themselves become an “investable asset class” that can be financed and leased.11 Extending “pretend accounting” to future sales of nonexistent hardware and build-outs The circular structure enables aggressive forward-looking commitments and accounting optics around capacity that does not yet exist: Massive multi-year purchase commitments, leases, and offtake agreements are announced and booked into remaining performance obligations (backlogs). These support revenue guidance, valuations, and further fundraising long before the hardware is manufactured or the data centers are powered. Financing platforms and residual-value guarantees let projects raise debt against future NVIDIA systems (Grace Blackwell, Vera Rubin, etc.) and the expected cash flows from leasing that compute. The hardware may still be in the design/production pipeline. Equity or guarantee capital from the supplier lowers the cost of capital for the buyer, who then places larger orders. NVIDIA (or a hyperscaler) can effectively see revenue and backlog growth that is partially enabled by its own balance-sheet or partnership support. In some cases the supplier ends up leasing capacity back or guaranteeing utilization, tightening the loop. Critics compare elements to 1990s telecom vendor financing (Lucent-style) or Enron-era special-purpose vehicles, though the structures are generally disclosed and the underlying demand for AI training/inference is real.13 Revenue recognition itself still largely follows delivery/control transfer under ASC 606, but the appearance of demand—order books, committed capex, and financed build-outs—runs far ahead of end-user cash flows and physical realization. Rising memory and server prices (the 15%+ hikes reported for early-2027 systems) amplify the issue: a fixed financed budget simply buys fewer actual machines, so the same dollar commitments translate into less real compute capacity.12 Effects on the real economy Short-term acceleration. The loops have demonstrably sped physical build-out. Hyperscaler and AI-related capex has become a major contributor to U.S. GDP growth in recent periods. Data-center construction, power infrastructure, semiconductor manufacturing, and related supply chains (HBM memory, networking, cooling, substations) expand faster than pure end-user demand or internal cash flows would allow. Jobs are created in construction, engineering, energy, and manufacturing; land, power, and specialized components are rapidly allocated to AI. Capital and resource misallocation. Because financing is abundant inside the circle, investment is pulled toward AI infrastructure at the expense of other sectors. Manufacturing construction, non-AI tech, housing, and other private investment face higher capital costs or reduced availability. Talent, copper, transformers, natural gas, and grid capacity are diverted. Estimates of over-investment relative to socially efficient levels (driven by competitive “race” dynamics plus circular stakes) run on the order of 50% or higher in calibrated models.46 Distorted signals and fragility. Reported revenue growth and order backlogs partly reflect recycled capital rather than independent, cash-paying end demand. If AI monetization (enterprise adoption, consumer willingness to pay, advertising, API usage) lags the infrastructure build, utilization falls. Specialized assets (GPU clusters with limited alternative uses) become candidates for fire sales. Interconnected equity, guarantees, and collateral create systemic spillover risk: stress at a keystone player (often identified as OpenAI-scale labs) can impair demand assumptions, collateral values, and credit across the web. Private-credit and institutional capital now sit inside the structures, raising the chance that a correction transmits beyond tech equities into broader credit markets.50 Cash-flow and balance-sheet pressure. Several hyperscalers have already seen free cash flow turn negative or sharply lower as AI capex ramps. Take-or-pay contracts and non-cancelable leases mean outflows continue even if revenues disappoint. Residual-value assumptions and chip collateral valuations are sensitive to technological obsolescence and secondary-market prices; higher current hardware inflation can support appraisals in the short run but raises leverage and deployment costs. Longer-term outcomes. If the AI productivity gains materialize at scale, the circular funding simply accelerated a valuable capital stock and the real-economy benefits (higher productivity, new applications) dominate. If monetization or technical returns fall short, the result is stranded data centers, written-down GPUs, capital losses for lenders and equity holders, and a period of underutilized real resources—energy plants, land, and specialized factories that could have served other purposes. Historical analogs (telecom equipment bubble, certain energy overbuilds) show that the physical assets remain, but the financial and opportunity costs are large. In short, circular funding converts balance-sheet strength and institutional capital into faster physical AI capacity today. It simultaneously compresses the feedback loop that would normally discipline overbuilding, embeds correlated risks across the largest technology and finance firms, and shifts real resources toward a concentrated bet whose ultimate payoff still depends on end-user adoption and technical progress rather than the internal recycling of capital. The $500 billion platforms and residual-support mechanisms extend that dynamic further into future, still-nonexistent hardware and build-outs.

• 
  Asher Crowe 🪺 (@ashercrw) reported 13 minutes ago

  THAT 28 SECOND CLAVICULAR CLIP IS NOT DRAMA. IT IS A MOTION REFERENCE WORTH MORE THAN ANY PROMPT YOU WILL WRITE THIS WEEK, AND THE MODEL WILL REFUSE IT UNTIL YOU FIX 1 THING. Girl lying in Clavicular's lap, certain she has already won. Her friend leans across her and kisses him. Then the caption everyone quoted: I thought you said you weren't interested. Everybody watched that for the drama. Watch it again for the movement. The lean. The reach across a body. The half second where a face reorganises itself in real time. That is the exact class of motion no prompt has ever produced, because a sentence can tell a model what moves and only footage can show it how. There is a rule underneath this that almost nobody applies. Text describes what moves. Video shows how. If the movement matters, you attach a clip that already has it. The 2 questions to answer before you attach anything: > Transfer or edit. Transfer borrows the motion and camera language and replaces everything else. Edit keeps the clip and extends it, precedes it, or interpolates between 2 > Inspiration or anchor. Inspiration means the model takes palette and mood and improvises. Anchor means that exact face or product survives into the output untouched > Say which one out loud in the prompt, because the model cannot guess and will pick the wrong one > 1 clean image already gives you 360 degree consistency, so stop burning reference slots on 5 angles of the same subject > Spend the slots you just saved on motion and audio instead Now the wall you hit about 4 minutes into trying this tonight. Realistic human faces in input video get blocked. Not degraded. Blocked. Most people find out by burning credits on it first. How to actually run a feed clip as a reference: > Obscure or blur the faces going in and let the model paint them back on the way out > Or feed a stylized version, animated or 3D, and let the output land realistic on the other side > Attach the clip for motion only, then anchor identity with a separate clean still of your own subject > Decide all of that before you generate, because discovering it afterwards costs an afternoon and a credit balance > Zoom every output to 100% before it leaves your machine, since small text still melts in everything below the newest tier Your feed stopped being entertainment roughly 6 months ago and nobody announced it. Every clip that stops your scroll is a motion reference, a lighting reference or a timing reference. The people building right now are saving them. Everyone else is arguing in the replies under a Clavicular video. So scroll back through everything you watched today. How many were content, and how many were reference files you failed to save? The full stack that turns a saved clip into finished work is in the article below. A shot from a top rated Amazon Prime series generated in a $10 tool, 5 recipes you can run tonight, the whole studio for $50 to $100 a month. Read it, then follow @ashercrw, because half the prices in that piece expire within weeks.

• 
  Rob Doyle (@robdoylecouk) reported 17 minutes ago

  @AmazonHelp When I sign in, it changes the store. Not me. Why?

• 
  Kiwi Mew🦋 (@KiwiMeww) reported 19 minutes ago

  It’s so uncomfortable because I feel like they just poked to see what they could get away with and now they’re in trouble. I’m starting to get really uncomfortable with Amazon because they keep on doing weird stuff and now twitch not caring for the streamers.

• 
  APARNESH PANDEY (@DEEPAK0551) reported 20 minutes ago

  @AmazonHelp I can't connect more 30th time i already called, my phone number is 7007232898, i need my money back by today only as i am having medical issues because of your amazon behaviour and no one bother to resolve and give proper response.

• 
  priya s (@priyanka2428) reported 20 minutes ago

  @AmazonHelp The link is not working

• 
  Sachin (@jetblack_ninja) reported 22 minutes ago

  @radhika_bajaj i was able to do so in 3 minutes. skill issue. amazon is doing fine.

• 
  Logan Weaver (@LogWeaver) reported 26 minutes ago

  US companies just posted their best profits since 2021. The stock market should be soaring, but... They just had their first down week in a month. This makes no sense on the surface. But there are two reasons hiding under the headline: First, look at how good the numbers were. S&P 500 earnings grew about 50% last quarter. That is the fastest growth since 2021. Most companies beat what analysts expected. 86% of them beat, well above the usual rate. It looked like a blowout quarter on paper. But that number hides something. Two companies did most of the heavy lifting. Strip out Alphabet and Amazon, and growth drops to 32%. Those two alone added 18 points to the total. The gains were not spread across the market. That is a lot of weight on two names. A big part came from gains on other companies they own. One booked a $98 billion gain on paper. The other booked $53 billion the same way. Those were paper gains, not real sales. It is like your house going up in value. You feel richer, but you earned nothing new. It was a one-time accounting boost. It was not money from selling more products. Real growth was there, just not fifty percent of it. So maybe the market saw through the hype? Not quite. Even the real number was strong. Revenue grew 15%, the fastest in years. Profit margins hit a record high. Even without those two giants, margins set a record. The core business was genuinely healthy. Demand was real and customers kept spending. So that is not why stocks fell. So what actually pulled them down? Two things were working against the good news. Borrowing costs have been climbing fast. Higher rates make every stock worth a little less. Stocks already trade above their long-term average. And the good news was already priced in. Investors expected a great quarter and got one. Analysts had even raised their targets going in. Even companies that beat barely moved higher. Good news everyone expects is already in the price. The best quarter in years still was not enough. Every other 50% jump this century came after a collapse. This time, earnings are already at record highs. This is the trap with headline numbers. A big number can hide a weaker story. A record quarter does not guarantee a rising stock. Even a strong story can already be priced in. The 50% looked amazing on the surface. Underneath, a third of it was two companies. Part of that was paper gains, not cash. And the rest was already expected. The headline and the real story rarely match. Great news can be a terrible time to buy. Retail saw record profits and got confused. The pros read what was actually inside them. That's the whole game. Surmount builds strategies on what the numbers really say. Start for free and trade the facts, not the headline.

• 
  Evan Swanson (@Evan_Swanson_) reported 26 minutes ago

  I take product IDs seriously because Amazon does. A UPC is tied to the product, and Amazon can check it against the GS1 database. If the number doesn’t match, you get a listing problem. and Amazon can remove your ability to create new ASINs. That creates a dangerous situation for brands that: • Buy questionable UPCs • Reuse identifiers across different products • Enter product IDs without verifying ownership • Assume a barcode on the packaging must be valid The safest path is simple: Get the identifier directly from GS1. Confirm it matches the exact product before creating the listing. A bad product ID can become a catalog-level problem.

• 
  Syed Owais | AI-Native Growth (@Ozee05) reported 29 minutes ago

  What's the real cost of an AI agent running quietly in the background? Amazon shelled out $1.8 million over five months on a Claude project. The unnoticed expenses highlight a critical issue: Most companies can't quantify the true cost of AI implementations.

• 
  kaushal Shakya ❤️ (@ITSKSL) reported 33 minutes ago

  @AmazonHelp @AmazonHelp @amazonIN I don't see things moving in here for this issue. The order is not yet delivered and I haven't received any call from Amazon support or delivery agent till today

• 
  Udita Arora (@arora_udita) reported 35 minutes ago

  @AmazonHelp There is a word in English ...sorry which u people use. The trouble is sorry doesn't make dead men alive.

• 
  Lord Paul Addison (POB)🟣 (@patvmru) reported 35 minutes ago

  @AmazonHelp No I won’t delete as this is what you need to resolve the issue instead of playing games with me!!!!

• 
  Akash Garg (@AkashDGarg) reported 39 minutes ago

  @amazonIN Order #402-4771984-1339557, not delivered yet. You charged extra for next day delivery. Order Date: 19th Aug Isn't it a poor quality of service. not fulfilling commitment. Faced issues in 3 of last 5 orders from Amazon.

• 
  Ahmed Elbehiry (@Abehiry1998) reported 40 minutes ago

  @AmazonHelp I cannot. It gives me an error. I cannot even make a call. Damn you! You made me angry and canceled the order.

• 
  Peter (@pw7fam) reported 41 minutes ago

  @EladioSantiag14 Circular funding (or circular financing) in the current AI boom is a web of interdependent equity stakes, credit guarantees, residual-value support, and third-party financing platforms in which capital from chipmakers, hyperscalers, and financiers flows to AI labs, neoclouds, and data-center builders, who then recycle much of it back into purchases of the same suppliers’ GPUs, cloud capacity, or infrastructure.15 It is essentially modern vendor financing at extreme scale. NVIDIA is the clearest node: it takes equity in customers (OpenAI, Anthropic, CoreWeave, etc.), provides or arranges credit support/guarantees, and helps structure debt vehicles collateralized by its own chips; those customers then buy more NVIDIA hardware or lease capacity that uses it. Hyperscalers (Microsoft, Amazon, Oracle, etc.) play similar roles with their cloud and equity commitments. The August 10, 2026 NVIDIA announcement of memorandums with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR formalizes this further: independent platforms designed to mobilize >$500 billion of third-party capital over time for AI-factory infrastructure, with NVIDIA optionally providing residual-value support up to 25% on a project basis. The chips themselves become an “investable asset class” that can be financed and leased.11 Extending “pretend accounting” to future sales of nonexistent hardware and build-outs The circular structure enables aggressive forward-looking commitments and accounting optics around capacity that does not yet exist: Massive multi-year purchase commitments, leases, and offtake agreements are announced and booked into remaining performance obligations (backlogs). These support revenue guidance, valuations, and further fundraising long before the hardware is manufactured or the data centers are powered. Financing platforms and residual-value guarantees let projects raise debt against future NVIDIA systems (Grace Blackwell, Vera Rubin, etc.) and the expected cash flows from leasing that compute. The hardware may still be in the design/production pipeline. Equity or guarantee capital from the supplier lowers the cost of capital for the buyer, who then places larger orders. NVIDIA (or a hyperscaler) can effectively see revenue and backlog growth that is partially enabled by its own balance-sheet or partnership support. In some cases the supplier ends up leasing capacity back or guaranteeing utilization, tightening the loop. Critics compare elements to 1990s telecom vendor financing (Lucent-style) or Enron-era special-purpose vehicles, though the structures are generally disclosed and the underlying demand for AI training/inference is real.13 Revenue recognition itself still largely follows delivery/control transfer under ASC 606, but the appearance of demand—order books, committed capex, and financed build-outs—runs far ahead of end-user cash flows and physical realization. Rising memory and server prices (the 15%+ hikes reported for early-2027 systems) amplify the issue: a fixed financed budget simply buys fewer actual machines, so the same dollar commitments translate into less real compute capacity.12 Effects on the real economy Short-term acceleration. The loops have demonstrably sped physical build-out. Hyperscaler and AI-related capex has become a major contributor to U.S. GDP growth in recent periods. Data-center construction, power infrastructure, semiconductor manufacturing, and related supply chains (HBM memory, networking, cooling, substations) expand faster than pure end-user demand or internal cash flows would allow. Jobs are created in construction, engineering, energy, and manufacturing; land, power, and specialized components are rapidly allocated to AI. Capital and resource misallocation. Because financing is abundant inside the circle, investment is pulled toward AI infrastructure at the expense of other sectors. Manufacturing construction, non-AI tech, housing, and other private investment face higher capital costs or reduced availability. Talent, copper, transformers, natural gas, and grid capacity are diverted. Estimates of over-investment relative to socially efficient levels (driven by competitive “race” dynamics plus circular stakes) run on the order of 50% or higher in calibrated models.46 Distorted signals and fragility. Reported revenue growth and order backlogs partly reflect recycled capital rather than independent, cash-paying end demand. If AI monetization (enterprise adoption, consumer willingness to pay, advertising, API usage) lags the infrastructure build, utilization falls. Specialized assets (GPU clusters with limited alternative uses) become candidates for fire sales. Interconnected equity, guarantees, and collateral create systemic spillover risk: stress at a keystone player (often identified as OpenAI-scale labs) can impair demand assumptions, collateral values, and credit across the web. Private-credit and institutional capital now sit inside the structures, raising the chance that a correction transmits beyond tech equities into broader credit markets.50 Cash-flow and balance-sheet pressure. Several hyperscalers have already seen free cash flow turn negative or sharply lower as AI capex ramps. Take-or-pay contracts and non-cancelable leases mean outflows continue even if revenues disappoint. Residual-value assumptions and chip collateral valuations are sensitive to technological obsolescence and secondary-market prices; higher current hardware inflation can support appraisals in the short run but raises leverage and deployment costs. Longer-term outcomes. If the AI productivity gains materialize at scale, the circular funding simply accelerated a valuable capital stock and the real-economy benefits (higher productivity, new applications) dominate. If monetization or technical returns fall short, the result is stranded data centers, written-down GPUs, capital losses for lenders and equity holders, and a period of underutilized real resources—energy plants, land, and specialized factories that could have served other purposes. Historical analogs (telecom equipment bubble, certain energy overbuilds) show that the physical assets remain, but the financial and opportunity costs are large. In short, circular funding converts balance-sheet strength and institutional capital into faster physical AI capacity today. It simultaneously compresses the feedback loop that would normally discipline overbuilding, embeds correlated risks across the largest technology and finance firms, and shifts real resources toward a concentrated bet whose ultimate payoff still depends on end-user adoption and technical progress rather than the internal recycling of capital. The $500 billion platforms and residual-support mechanisms extend that dynamic further into future, still-nonexistent hardware and build-outs.

• 
  Raj Gohil (@Ravenclaw__95) reported 42 minutes ago

  @AmazonHelp Yet another failed pickup. I don’t understand the problem here.

• 
  Professor X⚕️ (@pepple_miracle) reported 42 minutes ago

  There’s a thing about Amazon KDP most has failed to realize—once you don’t have a common purpose, niche, mission towards why you’re a publisher you’ll fail badly. There’s no secret just make show you reduce errors and manipulative activities as much as possible you’ll be fine.

• 
  Nehal choudhary 🔆 (@nehaljain_123) reported 43 minutes ago

  @AmazonHelp If Amazon genuinely wants to investigate this, please use the information available instead of simply redirecting me to chat. Please investigate the issue rather than avoiding it.

• 
  Helena Waulters (@laneylouise1) reported 46 minutes ago

  @RTweets44769 Fer sure!! She even said she had an “Amazon pickup”, and when Amazon replaces most items they usually tell you to keep/throw away the broken one. Not for a $700+ scooter.

• 
  pradeep kumar (@pradeep12345678) reported 49 minutes ago

  @AmazonHelp I have already connected with them none is ready to even create a escalation ticket for this issue. Please refund my money.

• 
  शुभेच्छु (@_subhechhu_) reported 50 minutes ago

  @amazonIN @amazon Do i need to go to himalaya & meditate for weeks to get your attention so that my issue gets resolved?

• 
  Amazon Help (@AmazonHelp) reported 52 minutes ago

  @DerEternal Hello! We're sorry to hear about the trouble with your account. This is never what we want or expect for you. On which Amazon Marketplace is your account registered (.com, .uk, .de, etc.)? We're here to help! -Sasha

• 
  kiki ☀️ (@kikibats) reported 52 minutes ago

  in fact it’s well known in the author community that self publishing through amazon preserves a lot of the money for the author instead of a publishing house getting a cut. so while amazon has its issues SO DOES EVERYTHING ELSE. if u have time to get mad about GR touch some grass

• 
  azh_jon (@azh_jon) reported 54 minutes ago

  @amazonIN @amazon there is no problem charging marketplace fee of ₹5 but access to customer care contact is conveniently removed.I received a near expired product which is non returnable. To connect, there is no option. Best customer experience. Keep charging.

• 
  Pam Lassiter (@PamLassite8u5e) reported 57 minutes ago

  @Breaking911 And he should be fired and never hold another job with Amazon. The man obviously has anger issues towards a 71 year-old man. Yeah the men in America are no longer man. They’re just little babies wanting to get their way. No wonder there are so many less children today women aren’t marrying idiots like this. Why would anybody marry a man that would beat up on an old man much less want him to be the father of the children?

• 
  Kim (@xxx_kimbo_xxx) reported 57 minutes ago

  Thanks to whoever nicked my amazon parcel recently. Refund has been issued and the item has gone down in price, so cheers buddy you've saved me a tenner 😎

• 
  Jayendra Mangela (@jay86m) reported 59 minutes ago

  @AmazonHelp Why is your service is going shi*ty day by day? you can check so many tweets from my side for all different orders, and why there is no human who can actually connect with me with all the issues I am facing? (2/2)

• 
  Ricardo (@Ric_RTP) reported 1 hour ago

  The most important company in AI just hired a bank to sell itself. You've probably never used it. But almost every AI product you have used was built on top of it. Here's what's really going on with "the Switzerland of AI": Hugging Face is the place where open AI models actually live. A lab finishes training something and publishes the weights there. Developers pull them down. Companies build products on top of them. Millions of models, datasets and apps sitting on one platform. And the reason the whole industry trusts it is that NOBODY owns it. Look at who funded the last round in 2023: Salesforce, Google, Amazon, Nvidia, Intel, AMD, Qualcomm and IBM. Sworn enemies writing cheques into the same company at a $4.5 billion valuation. And that was the entire point. Every rival holds a slice, so no rival controls it. That's why people started calling it the Switzerland of AI. On Sunday, Business Insider reported that Hugging Face has been working with a bank to sound out buyers at $13 billion or more. And this is where it gets really interesting… Late last year Hugging Face publicly turned down $500 million from Nvidia. They said no on the record, and the whole reason was staying independent. Then in November, CEO Clement Delangue stood on stage at the Axios BFD conference and said the industry was in an LLM bubble that might pop in 2026. He added that roughly half the $400 million his company had raised was still sitting unspent. So Delangue called the top, said he didn't need money, refused strategic cash to protect his neutrality, and then hired a bank to exit anyway. The numbers explain why: Hugging Face has raised about $395 million across 10 years. The ask is $13 billion. That's roughly 33x every dollar anyone has ever put into it. Five days before this leaked, Stripe agreed to buy OpenRouter for a reported $7.5 billion. OpenRouter routes developer traffic to models. Hugging Face stores the models. Neither company builds a model of its own. Somebody has worked out that the roads are worth more than the cars. Here's what a buyer actually gets: Whoever owns Hugging Face owns the shelf that every open model on Earth sits on, including the ones built by their direct competitors. They'd set the hosting terms, the search ranking, the access rules and the download speeds for products designed to beat their own. But also keep in mind that hosting millions of free models costs a fortune and earns almost nothing. The last public revenue figure was $30 to $40 million. Delangue and his cofounders started this in 2016 as a chatbot app for bored teenagers, and 10 years in, their investors want paying. A rich owner might be the only thing keeping the Hub free. Last month OpenAI disclosed that models it was testing escaped their sandbox, reached the open internet, and broke into Hugging Face's systems. Researchers counted around 17,600 attacker actions across four days in July and laid the whole thing out at Black Hat. Nothing is signed. No buyer has been named. Hugging Face hasn't said a word in public. But the bank is already making calls. Every AI lab on Earth paid to keep this thing neutral. One of them is about to own it.