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 (46%)
• Errors (29%)
• Sign in (25%)

Live Outage Map

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

City	Problem Type	Report Time
Washington	Website Down	2 hours ago
Paris	Website Down	5 hours ago
Reynosa	Website Down	11 hours ago
Marquette	Website Down	22 hours ago
Boston	Sign in	1 day ago
Bordeaux	Errors	1 day 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:

• 
  🚩Deepak Agarwal🚩 (@Deepak020692) reported 5 minutes ago

  @AmazonHelp Absolute pathetic service! ​My return was scheduled for 1st August but picked up on 5th August due to your delay. Still no refund or status! Fix it now or I'm taking this to Consumer Court and Cyber Cell.

• 
  DKM (@SMHCarolinablue) reported 9 minutes ago

  @BellJessbell25 I don't have a problem with Tom Cruise personally, I think he's been milking that Mission Impossible franchise for way to long, stopped watching those after #3 or #4, but he's bankable. The Amazon Prime Reacher is true to the character from Lee Child's book. I like Alan Ritchson too.

• 
  Sloane (@OC_Expat) reported 12 minutes ago

  @ericmmatheny I don't know anyone like that, but then I know seniors on fixed incomes. They're having trouble buying their meds, food, utilities. And Senior nutrition programs were just cut. But Amazon, right? Observation: Blaming poor people for being poor is quite the midterm strategy. 🙄

• 
  🚩Deepak Agarwal🚩 (@Deepak020692) reported 18 minutes ago

  @AmazonHelp Stop giving me automated or canned responses! Your 'Specialist team' has done nothing and the link shared earlier is of no use. My issue is still unresolved, and my money is stuck. Sort this out right now or face legal action."

• 
  Ange (@angejwestwood) reported 20 minutes ago

  @AmazonHelp Your link is broken: "Oops something went wrong". Dreadful customer service if you cannot give a valid link. Already cancelled my subscription to an exceedingly increasing poor delivery service.

• 
  Manas Behera 🇮🇳 (@manas_odin) reported 26 minutes ago

  @AmazonHelp I am trying 8-10 time. But customer executive told that they understand the issue but can't provide a positive resolution and delivery boy delivered not broken eggs. Kindly provide an return or replacement for this damaged items

• 
  Dr_Gingerballs (@Dr_Gingerballs) reported 27 minutes ago

  Some important AI context. $NVDA + $AMD have sold a total of about 20 GW of AI compute. There are 1B knowledge workers globally with 20 W brains. We have already deployed the equivalent of all knowledge workers. Despite this, productivity hasn't budged an inch. Profit margins are flat in the most generous terms, and down if you consider AI capex is really opex in disguise. If you recognize that AI is really just a marketing term for applied statistics, and that statistics provide diminishing, logarithmic returns, it's extremely likely that applied statistics will not economically compete with human intelligence. Let's say our uncertainty on productivity measurements is 1%, which seems reasonable. This implies in the best case that AI is 1% as productive as humans today. If you 100x compute as people suggest we should do, you arrive at 2 TW of power at a cost of $3T a year. The chips would cost $20T. Depreciated over 5 years puts us at about $5T a year on compute costs. This is $5000 per knowledge worker. At first this sounds great, given a knowledge worker currently costs around $50-80k a year. But this assumes that compute productivity scales linearly. We already know it scales logarithmically. This means the 100x increase in compute merely doubles the productive output. We go to 2% productivity increase. This is $250,000 per worker. Not competitive. Now obviously we are extrapolating very aggressively here, so it's not surprising the possible outcomes diverge greatly. The question is what should be the base case? Given historically we have seen logarithmic growth, should we expect linear from here on out? Or should we expect exponential like the market is pricing in? From a fundamental stats perspective, uncertainty decreases proportionally to the sqrt(data), and so model value should also follow this trend. This is the diminishing returns with scale that has plagued this field from the start. People are playing tricks with reinforcement learning to try and accelerate the increase in certain areas, with unknown but likely substantial decreases to the general performance (what folks call the "jagged" behavior of statistical models). Nothing has really shown that a general model of all knowledge work can break out of the logarithmic trend. So from a purely technical perspective, one should EXPECT the base case to be an uncompetitive economic replacement for humans. This is my base case. Additionally, systems tend to naturally become inefficient as they become more complex, simply due to entropy. It is rare that a 1000 person corporation outperforms a 100 person corporation simply because of the added complexity of organizing 10x more people. Scale and complexity are not the same thing. AI is not an economy of scale story. Nothing in this field gets cheaper as it gets bigger. It just at best doesn't get more complex, and in reality gets more complex. So size can only be bad for cost and efficiency in these systems (as we have seen to date). One datapoint for linear is the amount of money that OpenAI can charge per watt, which has been flat since inception at around 1.25 $/kWh. This implies that another 10x increase in compute (to 20 GW) would lead to $200B ARR. 200 GW would lead to $2T ARR. Clearly the market is pricing this in as the base case. The fatal flaw of drawing a straight line between revenue and compute is that they are selling the compute below cost. $NVDA charges about $1/kwh for their chips assuming 5 year life at 100% operation. Energy costs eat the rest. Now add in all of the other equipment in the data center, the middle men to operate it, the logistical costs, etc, and they are basically selling the market $3-5 worth of compute for $1.25. The world has converged on this arbitrage at global scales, and yet the AI companies are unwilling or unable to close it. The reason it isn't being closed is for two reasons: 1) There is not enough natural demand for their product absent the free money glitch. 2) There is currently an arms race to create a monopoly in AI compute. this begs the question, if the technical base case is catastrophically unprofitable, and the fantasy market priced base case is also catastrophically unprofitable, what is the value of the monopoly? Sure it solves #2 and gives you pricing power over your consumers, but it doesn't solve #1 in that your consumers will just stop using your product if you try to raise prices. It all ONLY makes sense if this is a war to take down Google and replace them for search. Forget Erdos and Alphafold all of the benchmark nonsense. All of that is marketing slop to get users away from Google and onto their platforms, so they are asking Claude where to buy underwear instead of Google. This is all about an ad selling monopoly. It even then makes sense why Zuck jumped into the fray. He's in the ad selling business, and needs to protect their moat in social media. So Google and Meta are at war with Microsoft and Amazon and the AI startups, where everyone would like to be the last one left standing, or at worst end up in a stalemate where they survive. And what is the best way to signal strength in a war? Money. Just like the US spends 10x other countries on the military to project strength, these companies are all trying to outspend each other to get someone else to buckle first. They are also strategically buying and selling chunks of themselves to each other in an attempt to gain leverage and reinforce weak points. It is at this point it becomes plain that this is a war over existing markets, not the creation of a new one. This is why the new market is a financial mess. Because the AI model market is just the warzone for ownership of the existing ad market. From this lens you also realize that all of the capital expenditures are not investments in the future, but the cost of munitions. It's all being spent on the consumables of war. The long term prudence is irrelevant, and you won't find a satisfactory answer if you go looking for it. It's about survival. Whoever wins takes the entire market. Whoever loses walks away with nothing. So every participant is forced to do whatever it takes to survive. The conclusions: 1) AI as a going concern is a zero. It's not meaningfully improving productivity, and likely never will. 2) Most of the current players will not survive and will have to be completely written off. 3) All of the capital expenditures are not durable investments, but consumables of war that will dissipate in a fireball worthless. 4) Even the winners will be severely impaired by the spending and will need to be repriced to reflect this.

• 
  CiNO (@SeanMancino) reported 33 minutes ago

  @babyat50 @DrJesseMorse From the website, it looks like there isn’t a way to purchase anything. I see a login option but I don’t have an account so I wouldn’t know but I get it from a small health food store. I’m sure there’s many others that carry it or maybe Amazon even carries it

• 
  Julio Carlos (Book reviewer for Indie Authors) (@Julio_Carlos_) reported 33 minutes ago

  For all the readers thinking that I vanished, I didn't. I am just reading a book in Amazon Kindle, and my eyes are slow when it comes to screens.

• 
  The Analyst (@MMatters22596) reported 33 minutes ago

  $AMZN is literally telling you where to invest. 4 years ago, they acquiered strategic positions in $ALAB and $MRVL. Since then: $ALAB + 350% $MRVL +170% Today they discloced a new stock position. They own 23% of a SMR company. This secures Amazon massive amounts of energy capacity for AWS Data centers. While it grants the SMR company strong backing and a huge order book. That company is $XE | X-Energy They already have a backlog for 11GW of power. But here's what set's them apart from $OKLO and $SMR. $XE is vertically integrated. They are producing their own fuel, which solves one of the biggest problems for SMR's. Security of supply. Fuel is the biggest bottleneck for the entire nuclear sector. $XE is not affected. But it also means that they can generate recurring revenue by selling their Fuel. Over all, this may be the best positioned SMR company right now. First commercial delivery is panned for 2030. Construction permits are expected for January 2027.

• 
  Mario Gordillo (Better Together) (@MarioGordilloR) reported 33 minutes ago

  Daily Markets & Business Briefing — August 7, 2026 1. July payrolls shock sharply lowers the probability of another Fed hike. The U.S. economy unexpectedly lost 23,000 jobs in July, versus expectations for an 80,000 gain. May and June were also revised down by a combined 103,000 jobs. Unemployment slipped to 4.1%, but largely because labor-force participation fell to 61.4%. Market impact: Treasury yields fell, the dollar weakened and the Nasdaq rose about 0.7% as traders cut the implied probability of a September Fed hike to roughly 40–44%. Watch next: next week’s inflation data. A soft CPI would be particularly supportive for long-duration growth stocks. 2. This is a meaningful positive for Big Tech valuations. Lower expected rates mechanically improve the present value of future earnings, which is especially relevant to NVIDIA, Amazon, Microsoft, Alphabet, Meta and Tesla. The immediate equity reaction was positive after the jobs report. Market analysis: for your core tech exposure, today’s weak labor data is initially bullish because it reduces rate pressure. The risk is that investors eventually interpret weak employment as a genuine growth slowdown rather than simply a reason for the Fed to pause. Watch next: whether falling yields continue to lift Nasdaq leadership without a simultaneous deterioration in earnings expectations. 3. Tesla and SpaceX deepen vertical integration with a US$16.8 billion AI-chip plant. Tesla and SpaceX announced an initial US$16.8 billion investment in a massive Texas semiconductor complex called Terafab, designed to manufacture, package and test AI chips for Optimus, Cybercab and SpaceX’s future space-based data centers. Total investment could eventually rise substantially beyond the initial phase. Market analysis: strategically positive because it reduces reliance on external chip capacity, but it also materially raises capital intensity. For Tesla, the investment increases the importance of successfully monetizing robotics and autonomy. Watch next: project timing, funding requirements, Intel’s participation and whether the spending starts affecting Tesla/SpaceX free cash flow. 4. The AI trade remains strong, but investors are punishing anything short of exceptional execution. This week AMD fell about 7% after earnings despite a revenue beat, while Sandisk and Western Digital also dropped following strong results. Reuters notes that S&P 500 earnings growth is tracking toward nearly 50% year over year this quarter, yet several high-growth technology names sold off after reporting. Market analysis: this is the key signal for NVIDIA: underlying AI demand remains powerful, but valuation sensitivity is rising. “Beat” is no longer sufficient; investors want accelerating guidance and evidence of return on capital. Watch next: NVIDIA customer CAPEX guidance and any sign that hyperscalers are slowing AI infrastructure commitments. 5. Energy remains the biggest macro wildcard. Brent crude traded around US$82 as negotiations over the Strait of Hormuz competed with renewed Middle East tensions, including attacks linked to the Houthis. Market analysis: lower oil plus weaker employment would be an ideal combination for Big Tech because both reduce inflationary pressure and Fed tightening risk. A renewed oil spike would reverse part of today’s valuation benefit. Watch next: any U.S.-Iran agreement affecting Hormuz access and crude prices over the weekend.

• 
  🚩Deepak Agarwal🚩 (@Deepak020692) reported 34 minutes ago

  @amazonIN @AmazonHelp Stop giving me automated or canned responses! Your 'Specialist team' has done nothing and the link shared earlier is of no use. My issue is still unresolved, and my money is stuck. Sort this out right now or face legal action."

• 
  Fraze lator (@FrazeLator) reported 36 minutes ago

  @bruce4nv @Spotify I used Amazon for a while but had issues playing back downloaded songs. Switched to Apple and have it as benefit through my credit card now too. I like the interface more(probably just cause I’m used to it now). Haven’t had issues with Ai slop.

• 
  Reddio 4aaaa.eth (@RichardCantwel9) reported 37 minutes ago

  They didn't hack these routers. They built them this way. A security firm called VulnCheck bought twenty Chinese-made routers off Amazon, AliExpress, and Alibaba. Every single one arrived with a remote-control implant already installed and running — as root, with no encryption and no authentication of any kind. Nobody broke in. The devices came off the factory line phoning home to a server in China, waiting for orders. The manufacturer is Zbtlink, a brand of Shenzhen Zhibotong Electronics. VulnCheck's research team named the implant ENDLESSDOORS. Here's the clever, insulting part: the implant disguises itself as "kworker," the name of a normal Linux kernel process. In a process list it hides in a crowd of legitimate-looking entries. But it isn't a kernel thread — it's an ordinary userland process running with full root privileges, launched at every boot by the vendor's own init script. Underneath, it's a customized build of rctl, an obscure remote-control tool that's been sitting untouched on GitHub since 2015. The mechanics are almost embarrassingly simple. The router dials out to a command-and-control server and sends a fixed 39-byte "hello" — a short device label plus its LAN MAC address. That's the entire registration. No handshake, no key exchange, no verification of the client or the server. After that, anything the server sends back is executed as root. There's no allow-list and no sandbox. One special keyword tells the implant to open a second connection and hand over a live, interactive root shell. The whole protocol has a vocabulary of two phrases: run this as root, and give me a root shell. The reason this is so dangerous is the direction of the connection. The router calls out. That means there's no open port to scan for and no inbound firewall rule to punch through. The connection originates inside your network and slips through NAT and typical egress filtering the way any normal outbound traffic does. A router sitting behind three layers of firewall in a hotel back office is exactly as reachable as one with a public IP address — as long as it can reach the internet at all. Whoever controls the C2 domain, or simply sits somewhere along the network path, controls the device completely. And this isn't a hypothetical. VulnCheck intercepted the outbound connection from their own test router, told it to hand over a shell, and it did — uid=0(root). Full control, in seconds. This isn't one bad batch or one model. Every firmware image on Zbtlink's own download page — roughly two dozen builds spanning more than two years of releases — embeds the implant and starts it at boot. All of them phone home. All of them can be hijacked the same way. The flaw has been assigned CVE-2026-66747 with a critical severity of 9.3 out of 10, and it's classified as CWE-506: embedded malicious code — not a coding mistake, but code that was put there on purpose. The scale is probably worse than twenty models. Zbtlink openly advertises OEM and ODM services, meaning it will build and rebrand this exact hardware and firmware for anyone who wants to slap their own logo on it. A "Wiflyer" unit can be the same affected device as its Zbtlink twin. The advice from VulnCheck is blunt: match on the model number, not the brand on the case. The true number of affected devices in the wild is impossible to count. VulnCheck also made a pointed choice not to notify the vendor first. Coordinated disclosure exists to give a manufacturer time to fix an accidental defect — but that logic only holds if the behavior was accidental. Here, the vendor's own boot script launches the implant, across twenty models and years of firmware. There's no bug to patch, because the backdoor is the product. Warning the shipper that they shipped it, they argued, would only tip off whoever operates the infrastructure. If you own one of these, the uncomfortable truth is that there's no fixed firmware coming. This is a device-trust problem, not a patching problem. You can technically disable the init script if you have shell access — but then you're trusting the rest of an image that shipped a root backdoor in the first place. For anything carrying real traffic, the honest answer is to replace the device. Or, put more plainly: if you've got one of these miracle boxes on your desk, take it for a walk to the nearest dumpster.

• 
  Connor Abene (@ConnorAbene) reported 39 minutes ago

  DTC founders: don’t celebrate revenue growth until you check this... A client sells through 3 channels: 1. Direct to consumer 2. Wholesale 3. Amazon Right now, around two-thirds of revenue comes from DTC, with wholesale making up most of the rest. That matters because wholesale runs at a lower margin than DTC for this business. Today, that is not a problem. It is just a fact sitting in the numbers. But if wholesale keeps growing as a share of revenue, the business can add topline every month while margin erodes underneath it. Revenue grows. Sales keep climbing. But the profit profile is changing. And by the time the margin hit becomes obvious, the company may have already made hiring, inventory, and spending decisions based on a margin that no longer exists. Don’t just track revenue by channel. Track the mix. Know what percentage of revenue each channel represents every month. Then model what happens to margin if that mix shifts 10 points in either direction. Revenue growth is only good news if you understand what kind of revenue is growing.

• 
  Zed (@Zed94K) reported 40 minutes ago

  @ThyGamblerArmy It would really help my situation. Literally the most depressed ive ever been but hey we’re trying. Paralyzed dad for over 19 years that is 78, mom is 74. A car that doesnt work. Overdue rent bills and medicine everyday. Amazon layoffs due to AI. Broke my arm no money to fix it. When i tell you i thought of ending it i mean it. Stake: ZKEclipse Photo of my dad incase people think im lying

• 
  Ross Gerber (@GerberKawasaki) reported 42 minutes ago

  Uber has many new competitors that don't have to deal with human management and issues. Uber isn't close to building an AV system of their own. And it eats margins and hurts their current drivers to do that. Amazon, tesla and Google don't need ubers software or customer data... sure they can serve lots of cities without AVs. But they are just getting attacked from all sides. Tech will beat the human solution for cabs for sure.... $UBER

• 
  ItsJeffSDigital (@highertrust) reported 42 minutes ago

  @torontobaghead @SportsnetPR @hockeynight They fail to realize they are only making fragmentation of their market worse. As bad as watching hockey on Amazon or Apple is - at least those platforms are improving and stick to the sport. Reminds me of all MSM that lose more and more of their market each year - double down.

• 
  🌵Sentinel Base (@SlShuo) reported 43 minutes ago

  On August 6, the U.S. stock market was split in two. On one side, big short Michael Burry was shorting AI again. On the other side, money was still pouring into AI at an almost absurd pace. So is this a revolution, or is it a bubble? Let’s break it down piece by piece. First, the short side. According to reports, Michael Burry opened massive short positions in Nebius $NBIS at $211.77 and Oracle $ORCL at $144.63. This was also his first disclosed position in $NBIS. His core logic is straightforward: the boom in AI infrastructure may be backed by heavy leverage and a pile of off-balance-sheet obligations, including backstop support, leases that have not yet started, and purchase commitments. He is not betting that AI has no future. He is betting that part of this AI infrastructure boom may be supported by “future debts” that have not been fully priced in. But here is the problem: the data on the other side is also extremely strong. Applied Optoelectronics ($AAOI ) reported second-quarter revenue of $191.9 million, up 86% year over year, marking its fifth consecutive quarter of record revenue. The company also returned to non-GAAP profitability. Management expects demand to remain ahead of production capacity through mid-2027. Now look at AI capital spending at the macro level. In the second quarter of 2026, private-sector investment in AI-related categories in the U.S. surged to an annualized rate of $1.5 trillion. Over the past two years, AI-related business investment has increased by roughly $500 billion, and direct AI investment may now account for 25% to 33% of recent U.S. GDP growth. Data centers are even more extreme. In June, U.S. data center construction spending jumped 46% year over year to a record annualized rate of $68 billion. Since January 2024, spending has surged 158%. Today, spending on data centers is about $25 billion higher than spending on office buildings. In 2022, it was the opposite: office construction spending was $57 billion higher than data center construction. It is very clear where the money of this era is going. Even Google ( $GOOG ) is loading up on ammunition. According to reports, Google plans to raise funds through the U.S. investment-grade bond market, marketing as many as 10 tranches of notes with maturities ranging from 2 years to 40 years. Initial pricing discussions for the longest-dated tranche were around 1.55 percentage points above U.S. Treasuries. So Burry is warning about risk, while industry capital is still pouring in aggressively. That is the core contradiction in the market right now: AI fundamentals are still hot, but the trade itself is already very crowded. The clearest example is the memory ETF $DRAM. It only launched in April this year, but trading activity has already reached extreme levels. Daily turnover once approached $8 billion, higher than ARKK at the peak of its 2020-2021 frenzy. Cumulative inflows into $DRAM have climbed to around $27 billion, also above ARKK’s previous peak. But do not forget: $DRAM has already fallen nearly 40% from its June high. What does that tell us? It does not mean the industry thesis is dead. It means that once a crowded trade starts unwinding, the speed can be very fast. For individual stocks, that means names like $MU in the memory space are not suitable for blind short-term accumulation. First, we need to see whether the position unwind is actually over. Now, look at the cold wind from last night. On the Federal Reserve side, reports related to Kevin Warsh sent an important signal: if upcoming inflation data runs hot and the market starts pricing in higher borrowing costs again, he would be open to a September rate hike. That sentence matters. The market had been trading around rate cuts and easier liquidity. But if the probability of a September rate hike starts rising again, tech valuations will come under pressure. The labor data also did not give the market a clean dovish signal. June JOLTS job openings fell to 7.359 million, below expectations, showing that labor demand is cooling. But hiring increased, quits rose to nearly a one-year high, and layoffs remained largely stable. This is not a collapse. It is “cooling, but not stalling.” The labor market is still not weak enough for the Fed to comfortably turn dovish. So the real focus today, August 7, comes down to one data point: the nonfarm payrolls report, released at 8:30 a.m. Eastern Time. Scenario one: non-farm payrolls come in strong. A hot labor market means rate-hike expectations rise, and tech stocks come under pressure. In that case, Palantir $PLTR may first test support around $148. Microsoft $MSFT needs to see whether buyers step in around $460. SpaceX $SPCX had the second-highest options volume yesterday at 1.7 million contracts, and with post-unlock volatility amplified by speculative capital, it could fall harder than the broader market if risk appetite weakens. But there is one key detail with $PLTR: yesterday, volume expanded, but the price went sideways. High volume with no price movement is not “no signal.” It is actually an important signal. Technically, this is called high-turnover equilibrium: a large number of shares are changing hands, but buyers and sellers are temporarily balanced. $PLTR was "stuck" at the $154 to $155 area yesterday. There might be two possible interpretations. One is accumulation: buyers are stepping in during fear, and the stock refuses to fall. The other is distribution: profit-taking is still happening, but it is being absorbed for now. The options market confirms this tug-of-war. $PLTR saw 477,000 options contracts traded yesterday, putting it in the top 10 across the market. Both bulls and bears are placing heavy bets here. So, for $PLTR today, the key is not how much it shakes intraday. The key is where it closes. If a hawkish shock pushes it below $143 and it closes underneath that level, then it is no longer ordinary volatility. It becomes a short-term breakdown, and the technical discipline is to prioritize reducing risk. But if it only dips intraday and then recovers into the close, that means support is still present. Scenario two: nonfarm payrolls come in weak. A cooler labor market means rate-hike expectations ease, and tech stocks get room to rebound. In that case, $PLTR, $SPCX, and $AMZN all have a chance to rebound as risk appetite recovers. If $TSM continues moving higher, watch the resistance around $450 and whether volume confirms the move. Whether the rebound can continue is not about one green day. The key is whether volume follows through. There were also a few headlines from last night that can easily mislead traders, so they are worth addressing separately. First, Bezos filed to sell about 1.2096 million shares of Amazon at $286.41 per share, for pre-tax proceeds of roughly $346.5 million. But this type of 10b5-1 planned sale is more about rules-based execution. Do not simply interpret it as “the founder is bearish on Amazon.” The headline may look scary, but it is not the core signal for trading $AMZN today. Second, Burry’s logic for shorting $NBIS and $ORCL is fascinating, but it is a monthly-level industry narrative, not an overnight candlestick trigger. It reminds us to watch for off-balance-sheet obligations and leverage in AI infrastructure, but it should not be treated as proof that all AI stocks must immediately collapse today. Third, the options market remains extremely hot. By contract volume, yesterday’s top 10 most active names included $NVDA, $SPCX, $AAPL, $TSLA, $MU, $MSFT, $INTC, $HTZ, $PLTR, and $ET. $NVDA led by a wide margin with roughly 2.9 million options contracts traded, showing that the AI theme remains the center of market speculation. Finally, look at two signals from asset allocation. Tether bought another 14 tons of gold in the second quarter, bringing its total holdings to 146 tons, worth about $18.8 billion. It is now the largest known private holder of gold outside central banks and governments. In the first half of the year, only four central banks bought more gold than Tether: Poland, Uzbekistan, China, and Kazakhstan. This shows that outside the AI frenzy, safe-haven capital has not been idle. On the other side, Chinese AI companies are starting to monetize more directly. According to Reuters, Alibaba plans to seek revenue-sharing arrangements for the next version of its open-source Qwen model, while Moonshot’s Kimi is also asking partners for revenue sharing of up to 30%. After two years of land-grabbing, AI models are finally entering the stage of “who can actually get paid.” So last night’s U.S. market was essentially a tug-of-war: On one side, there is an AI boom backed by real money and real spending. On the other side, there are bubble warnings from rates, valuations, leverage, and crowded trades. Before today’s open, remember one line: nonfarm payrolls decide the direction, but the close confirms the truth. Strong payrolls: watch tech pressure and key support levels first. Weak payrolls: watch the strength of the AI rebound. When everyone is going crazy, remember to keep one part of yourself clear-headed. [Disclaimer: This article is for market information sharing only and does not constitute investment advice. Investing involves risk. Please enter the market with caution.]

• 
  GraceLynn (@FullOfGraceUS) reported 43 minutes ago

  Welp, Walmart's surge pricing has hit their website. So now I have no reason to shop with them at all. I haven't stepped inside one since delivery started. So now, if you buy something often, they will raise the price on you. I accidentally found this out by searching for puff pastry before I logged in. I logged in, and they raised the price by .30 cents. I logged out, and it went back down. Fuuuuck you, Walmart. I would rather pay double at Publix than give you one more red cent. Raise prices as high as you want; that's your prerogative, but don't be sneaky with it. This is Amazon-level stupidity.

• 
  reji koshy (@rejipkoshy) reported 45 minutes ago

  @AmazonHelp It was a glitch on the screen with relation to your website that didn’t show up. I’ve sorted it out.

• 
  Nidhish Dubey (@NidhishDubey_) reported 45 minutes ago

  @AmazonHelp @amazonIN Suddenly all large appliances are showing unavailable for delivery at my pincode 487551 for the last 30 days. Tested multiple accounts. Nearby pincodes work. This seems like a backend glitch. Don't want to switch to Flipkart. Please help

• 
  MarMar Labs (@MarMarLabs) reported 47 minutes ago

  The most useful agent release this week might be a two-line manifest. Google just joined Amazon, Cursor, Microsoft, OpenAI, and Vercel as a core maintainer of Agent Plugins 1.0.0, a working-draft format for packaging skills + MCP servers into one portable directory. The compatibility page lists VS Code, Cursor, GitHub Copilot, ChatGPT & Codex, and Kiro, with the components and transports each supports. The package is intentionally boring: plugin.json → identity + spec version skills/ → instructions, scripts, references mcp.json → tool connections com.yourclient/ → client-specific extras The part I like: the spec stops at packaging. It does not define installation, distribution, permissions, sandboxing, trust/provenance, or UX. Those stay with each client. A capability can travel without pretending every runtime has the same security obligations. My rule for builders: • one skill? ship a skill • one MCP server? keep it simple • skill + tools that only make sense together? make a plugin • client-specific behavior? isolate it instead of forking the whole package We spent the last year standardizing how agents call tools. Now the distribution layer is starting to standardize too. If you ship agent capabilities today, what still causes the most wrapper drift: config, auth, or hooks?

• 
  carp (@DepasMarki92257) reported 48 minutes ago

  An Amazon senior engineer stopped feeding the model everything and built a pipeline that decides what it sees. Four stages: Rank files by relevance to the task, not by last commit date. Route the task to a context budget — a typo fix gets three files, a module rewrite gets the module. Compress what survives to the lines that matter. Ship that, and nothing else. The result is obvious in hindsight: token cost per finished task drops as soon as the model isn't reading half the repo to fix one function. Full breakdown of the four stages below.

• 
  Lily (@lily_choudhury_) reported 49 minutes ago

  A guy had Prime for 9 years. He used it for exactly one thing: free 2-day shipping. That's it. $139 a year, every year, just to skip the shipping fee on protein powder and phone chargers. His roommate — who used to work logistics for a third-party Amazon seller — watched him pay full price for groceries at the store down the street, the same week Prime members were getting 10% off and double points on that exact grocery chain through Amazon. "You know Prime does way more than shipping, right?" "It's just shipping and the TV shows." "You're paying $139 a year for a membership that includes free grocery delivery, prescription discounts, a photo storage vault, exclusive deals, and a credit card that pays you 5% back on everything you buy here — and you've been using it like a coupon for phone chargers." He walked him through his account in 15 minutes. Same membership. Same $139. Completely different value. Here's everything he showed him 🧵

• 
  Michael Patrón (@michaelpatron0) reported 50 minutes ago

  Just another great day with Amazon - they broke my variations on tons of ASINs. Been live for years, no abuse but now Amazon decided to f***k them all - will take my team hours and hours to fix this. Never have this issue with Target, Walmart, Shopify, Tiktok or Ebay.

• 
  Sourabha Shakya (@sourabha4u) reported 51 minutes ago

  @AmazonHelp They are not helping. They are saying the laptop is not found in their warehouse and refused to issue refund.

• 
  Wicked Freckled Trull (@WickedFreckledT) reported 53 minutes ago

  @DanNEO_SS Length is probably less relevant than getting reviews at all so the algorithm is mollified. Are you fixing the right problem? I am reminded of the alternatives to GoodReads, Amazon, bookstores, and libraries that readers are not checking so posting there is irrelevant.

• 
  jason25 (@Joe09459005) reported 53 minutes ago

  @AmazonHelp "Hi, my Prime delivery dates have suddenly become delayed only for my saved address in Pattabiram (PIN 600072). Other nearby addresses receive faster Prime delivery. This address worked normally before. Could you please check if there's an address mapping issue?"

• 
  Sourabha Shakya (@sourabha4u) reported 54 minutes ago

  @AmazonHelp @JeffBezos Your team in @amazonIN is doing suspicious. They pick up product and do not issue refund. Can you investigate - how someone pick the laptop, match the seial number and update status - empty box?