Self-Sovereign GraphQL in Every Browser
Arweave holds the data and the index; PermawebOS gives every user a node that can answer and prove its own queries.
The third principle of Arweave is “Guarantee the right to listen”. In cyberspace, you can only hear if you can actually discover the data. On the permaweb, that’s made possible with GraphQL.
GraphQL provides a critical link in the composability architecture of the permaweb: allowing all apps to build on top of the same, shared content lake, joined by a single global index. Arweave transactions are posted with tags, discoverable by wallet address, or the block they were mined into, but GraphQL is the layer that lets users and applications see all data matching these queries. Without it, there’d be no way to find data matching criteria, or build data protocol-based apps.
It’s one of the permaweb’s core utilities, but until now it has been confined to the realm of enterprise-grade hardware. Services like that are usually incredibly difficult to decentralize because the hardware requirements are high and the incentives are low. The permaweb’s GraphQL services so far have depended on building and maintaining gargantuan off-chain indexes of more than 76 billion rows: scanning the historical chain, keeping it current and serving it quickly. In practice, this has meant that permaweb apps inherit the availability and policy of a tiny number (often two, sometimes – like today – even just one) of hosted indexers which had no incentive to adopt a decentralized model.
Offset queries (queries that seek, intersect and page by weave position) will change everything about how Arweave’s query layer is served. HyperBEAM's offset-query path makes the index compact enough to publish on Arweave, orders it by weave offset, and lets any node or client query the relevant pages directly.
(txid)
This is not another centralized GraphQL service, and it’s not just a way to make GraphQL more available to node operators. It is a method that makes GraphQL execution so lightweight that it can be run directly on users’ browsers, with Arweave nodes simply serving them data chunks.
Hyperoptimized GraphQL with Offsets
All pieces of data in Arweave already have unique positions in the weave – every byte is either ‘before’ or ‘after’ every other byte. We call these byte positions in the network offsets. In March, we showed how any transaction on the weave can be referenced by its offset as a name -- short, deterministic values like 101t.arweave.net, derived only from the data’s onweave properties.
That same offset property gives us more than a way to retrieve bytes. It also can power queries because it gives us a common ordering that can be reused across every possible match in an index.
The new match index stores rows using three extremely compact values:
a hash of the field name being matched;
a hash of the value that is present;
the weave offset of the item carrying that predicate.
Each potential match is compressed into an average of just ~9.5 bytes per row, each stored in an onweave ArLMDB database. By utilizing Arweave chunks as batches of LMDB pages, ArLMDB allows us to traverse the database to find any specific node with only a few individual Arweave node requests. By organizing rows into the compressed components 1-3 above, a query with two or three factors to match can walk those ordered sets together, advancing whichever cursor is behind until the offsets meet – a ‘leapfrog’ version of the same flow as a single lookup. Critically, finding the intersection is part of finding the results for each match criteria -- not a second step.
The same ordering solves pagination -- page fifty can seek to its starting offset instead of replaying pages one through forty-nine. Offset lookups give results a stable order without adding another ordering database.
AO Compute; Arweave the Shared Hard Drive
LMDB is normally a local database file. HyperBEAM's new arlmdb store reads that database from Arweave instead.
Try decentralized GraphQL from your browser
This is already proven at production scale with item lookup. A 622 GiB transaction contains the locations of over 70 billion rows. HyperBEAM reads the database where it sits. A cold lookup traverses it with three Arweave chunk requests; once the shared branches are cached, another lookup needs just one. The ArLMDB implementation is merged into HyperBEAM and used for ID lookups from Arweave.
With an immutable index on Arweave, anyone can read it without trusting the publisher to keep a query endpoint online. Like the Arweave schedulers powering Bazar, this is another example of AO employing Arweave as the source of truth, and using it to power the permaweb.
Instead of every query operator repeating the historical sweep and building the same large database, nodes serve chunks while the client traverses the index locally. The node's job is reduced to serving bytes quickly. It does not decide which results exist, execute the filter or ask the application to trust its view of the weave. The live tip still needs rolling indexes, but the expensive historical work no longer has to be repeated by every participant.
The client does not even need the whole database (depending on optimization, we’ve seen database sizes anywhere between 1 and 60 terabytes). The production offset index already demonstrates the access pattern: in the live chunkar browser demo, the second arbitrary lookup needs only around half a megabyte of new index data to traverse the full offset index. The match index works the same way. It traverses the pages needed for the requested predicates, intersects the ordered rows, reads the candidate items and checks that they actually carry the fields requested by the query. This is lightweight enough to make it so that every user can be their own personal no-dependency GraphQL service provider, with provable data, from the browser.
As well as the browser playground, today you can use the arlmdb.js library to integrate Arweave-stored database lookups into UIs.
From Centralized Node to Self-Sovereign Service
Today an application sends a GraphQL request to a server that already holds an index. With a published offset index, the application can instead hold a locator and a cache.
A HyperBEAM node can do that, but one way we imagine most users will access the query layer is through the browser: a user's application fetches Arweave chunks, keeps the hot index pages locally and performs the query for itself. This can be baked into the PermawebOS browser extension along with the local AO node the extension already spawns. This shift brings more and more AO services that were previously hosted (on exclusive TEE hardware) into a local-first environment. The stack is becoming so lightweight it’s able to be run cheaply, per-user, as background services in the browsers, phones and laptops everyone already has.
Previous designs for a decentralized GraphQL layer were unable to answer the question of trust without TEEs. How can you be sure that the response you get back from a GraphQL server has arrived complete and uncensored? While PermawebOS node architectures like LapEE and AndEE solve the trust question in theory, in practice the job of indexing the entire blockweave is too big for small workers.
Offset queries over onweave LMDB data -- made lightweight enough to traverse and prove by any browser -- change the equation entirely.
The browser never asks a server to decide the answer. It asks nodes for the exact index chunks it needs and walks the authenticated LMDB pages itself, intersecting the offset-ordered sets locally.
The new schema provides chunks and Merkle paths for each page accessed; the recipient can repeat the exact same work and see that they get the same complete set, with no missing results. The hashpath signs the request + response pair, and the validator can trivially repeat the work, whether on a node or in a browser. The resulting item is then bound back to its ANS-104 ID. A node can withhold bytes and make itself unavailable, but it cannot alter a row, skip a qualifying result or invent one without breaking the proof.
The Personal Permaweb Stack
Decentralizing GraphQL does not necessarily require a decentralized fleet of GraphQL servers. The permaweb is unique in that it combines a decentralized permanent storage layer with everyday web semantics. HTTP, lightweight proofs, and browser-based nodes handle the workload when the logic is optimized enough to sidestep expensive hardware.
The network holds the whole index. Each user keeps only the path to their answer.
With PermawebOS putting an AO node in every browser and LapEE turning abundant consumer hardware into secure bundlers, schedulers and tunnels, local GraphQL pushes the permaweb towards a cyberspace everyone can own.
Read this on the Permaweb:
https://ao.arweave.net/#/blog/self-sovereign-graphql-in-every-browser
5·CNeutral
n
news9/7news
Eyes on the Market: Sustained Inflows
Oil led everything. Brent rose 9.27% to $96.54 as US-Iran strikes resumed. BTC added 2.28% to $79,808, ETH 1.62% to $2,490.60, the S&P 0.40% and the Nasdaq 100 0.73%. Waller signals dovish. Polymarket hike odds fell from 51% to 41% on Thursday's remarks. BTC wicked to $82,262 and closed the session up 5.67%. Payrolls reversed it two days later. 162K against 53K consensus took hike odds back to 49%. BTC fell 2.95% and ETH 2.69%. The move is not levered. Aggregate OI is $139.7B, up 3.5% and the highest since mid-January. Coin-denominated BTC OI fell from 762.2K to 669.6K since mid-August, even as BTC moved from $63K to $80K. ETFs took $1.2B, a third straight week above $1B, the first such run since July 2025. BTC drew $986.7M including $730.8M on September 3, the largest day since January 14. BTC ETFs AUM crossed $103.3B, 6.32% of supply. Oil Leads the Week BTC opened Monday at $78,031, hit a high of $82,262 on Thursday and closed Sunday at $79,808. Up 2.28% on the week. ETH gained 1.62% to $2,490.60. Total crypto market cap finished at $2.708T, slightly outperforming BTC as selected alts continue to rally.
Oil saw the largest move, rising 9.27% to $96.54 as the US and Iran conflict intensified during the week. The S&P 500 added 0.40% to 7,728.60 and the Nasdaq 100 gained 0.73% to 29,616. Gold slipped 0.60% to $4,470.50. Events of the Week US-Iran Strikes Resume US-Iran strikes resumed for the first time in roughly a month after the 60-day ceasefire lapsed in mid-August. US forces disabled two Iranian tankers and destroyed a third on September 2, following IRGC ballistic missile fire at a US carrier and destroyer. Iran struck Kuwait with missiles and drones on September 3, escalating to a US Gulf ally, and Israel warned it would cripple Iranian infrastructure. Brent went from $88.32 to $95.15 on Tuesday, then plateaued between $95 and $97 for the rest of the week. Waller Gives Disinflation a Chance Fed Governor Christopher Waller's prepared remarks went out on Thursday 8:30am ET. Inflation is still meaningfully above the 2% goal, he said, but the recent data finally show some signs of disinflation, and if that holds through the next two weeks he would back holding the funds rate at 3.50% to 3.75%. Treasury yields fell to session lows and hike odds on Polymarket dropped from about 51% to about 41%. BTC wicked to $82,262 before settling at $81,704, up 5.67%. Hot Payroll Print August payrolls came in at 162K against 53K consensus, three times the estimate, with unemployment at 4.1% in line. Strongest print since March and the first up-month in five after. Expectations of a hike reversed back up from 41% to 49%. BTC fell 2.95% and ETH 2.69%. Volatility, Positioning and Leverage BVIV is up 5.4%, from 38.9 last week to 41.02. The metric is up 14% from the low of 35.81 that it hit on August 7. Aggregate futures open interest is $139.7B against $135B last issue, up 3.5% and the highest since mid-January. 24-hour volume is $136.4B, up 31%. BTC open interest in coin-denominated terms is the lowest it’s been since March 25. It’s been steadily falling since mid-August from 762.2K BTC to 669.6K BTC as BTC has risen from $63K to $80K. A reflection of how spot driven the recent move has been. Coinglass's 24-hour long/short ratio is 49.04% / 50.96%. Annualized funding on Binance runs BTC near 4.7%, ETH 8.4%, SOL flat to slightly negative and HYPE 5.5%. ZEC funding is roughly -3.65% annualized and ZEC still gained 45% in the past week. The coins that lead are now moving on spot inflows. ETF Flows Continue BTC and ETH ETFs took $1.2B combined, a third consecutive week above $1B. The last stretch of this magnitude was July 2025. BTC: $986.7M. September 3's $730.8M is the largest single day since January 14 and the third largest of 2026, behind January 14 at $840.6M and January 13 at $753.8M. BTC ETF AUM crossed $103.3B, 6.32% of supply. Year to date cumulative flows now sit at -$0.90B, from -$4.74B three weeks ago. Three weeks have erased 81% of the year's outflows. ETH: $215.3M. Flows fell 74% week on week from $815.7M. ETH captured 22% of BTC's dollar flow against 88% the prior week. Cumulative net flows are $13.19B and August closed at $1.84B. This breaks the ETH outperformance pattern we have tracked since late July, and the spot data agrees: ETH gained 1.62% against BTC's 2.28%. First week in a while where ETH lagged on both flow and price.
Key Events for the Week Ahead Tuesday, September 8 US consumer credit (G.19), 3pm ET. Wednesday, September 9 US NFIB small business optimism for August. Thursday, September 10 US PPI for August, 8:30am ET. Europe: ECB rate decision, 8:15am ET. Friday, September 11 US CPI for August, 8:30am ET.
90·A+Long
n
news9/7news
Eyes on the Market: Sustained Inflows
Oil led everything. Brent rose 9.27% to $96.54 as US-Iran strikes resumed. BTC added 2.28% to $79,808, ETH 1.62% to $2,490.60, the S&P 0.40% and the Nasdaq 100 0.73%.
Waller signals dovish. Polymarket hike odds fell from 51% to 41% on Thursday's remarks. BTC wicked to $82,262 and closed the session up 5.67%.
Payrolls reversed it two days later. 162K against 53K consensus took hike odds back to 49%. BTC fell 2.95% and ETH 2.69%.
The move is not levered. Aggregate OI is $139.7B, up 3.5% and the highest since mid-January. Coin-denominated BTC OI fell from 762.2K to 669.6K since mid-August, even as BTC moved from $63K to $80K.
ETFs took $1.2B, a third straight week above $1B, the first such run since July 2025. BTC drew $986.7M including $730.8M on September 3, the largest day since January 14. BTC ETFs AUM crossed $103.3B, 6.32% of supply.
Oil Leads the Week
BTC opened Monday at $78,031, hit a high of $82,262 on Thursday and closed Sunday at $79,808. Up 2.28% on the week. ETH gained 1.62% to $2,490.60. Total crypto market cap finished at $2.708T, slightly outperforming BTC as selected alts continue to rally.
Oil saw the largest move, rising 9.27% to $96.54 as the US and Iran conflict intensified during the week. The S&P 500 added 0.40% to 7,728.60 and the Nasdaq 100 gained 0.73% to 29,616. Gold slipped 0.60% to $4,470.50.
Events of the Week
US-Iran Strikes Resume
US-Iran strikes resumed for the first time in roughly a month after the 60-day ceasefire lapsed in mid-August. US forces disabled two Iranian tankers and destroyed a third on September 2, following IRGC ballistic missile fire at a US carrier and destroyer. Iran struck Kuwait with missiles and drones on September 3, escalating to a US Gulf ally, and Israel warned it would cripple Iranian infrastructure. Brent went from $88.32 to $95.15 on Tuesday, then plateaued between $95 and $97 for the rest of the week.
Waller Gives Disinflation a Chance
Fed Governor Christopher Waller's prepared remarks went out on Thursday 8:30am ET. Inflation is still meaningfully above the 2% goal, he said, but the recent data finally show some signs of disinflation, and if that holds through the next two weeks he would back holding the funds rate at 3.50% to 3.75%. Treasury yields fell to session lows and hike odds on Polymarket dropped from about 51% to about 41%. BTC wicked to $82,262 before settling at $81,704, up 5.67%.
Hot Payroll Print
August payrolls came in at 162K against 53K consensus, three times the estimate, with unemployment at 4.1% in line. Strongest print since March and the first up-month in five after. Expectations of a hike reversed back up from 41% to 49%. BTC fell 2.95% and ETH 2.69%.
Volatility, Positioning and Leverage
BVIV is up 5.4%, from 38.9 last week to 41.02. The metric is up 14% from the low of 35.81 that it hit on August 7.
Aggregate futures open interest is $139.7B against $135B last issue, up 3.5% and the highest since mid-January. 24-hour volume is $136.4B, up 31%. BTC open interest in coin-denominated terms is the lowest it’s been since March 25. It’s been steadily falling since mid-August from 762.2K BTC to 669.6K BTC as BTC has risen from $63K to $80K. A reflection of how spot driven the recent move has been.
Coinglass's 24-hour long/short ratio is 49.04% / 50.96%. Annualized funding on Binance runs BTC near 4.7%, ETH 8.4%, SOL flat to slightly negative and HYPE 5.5%. ZEC funding is roughly -3.65% annualized and ZEC still gained 45% in the past week. The coins that lead are now moving on spot inflows.
ETF Flows Continue
BTC and ETH ETFs took $1.2B combined, a third consecutive week above $1B. The last stretch of this magnitude was July 2025.
BTC: $986.7M. September 3's $730.8M is the largest single day since January 14 and the third largest of 2026, behind January 14 at $840.6M and January 13 at $753.8M. BTC ETF AUM crossed $103.3B, 6.32% of supply. Year to date cumulative flows now sit at -$0.90B, from -$4.74B three weeks ago. Three weeks have erased 81% of the year's outflows.
ETH: $215.3M. Flows fell 74% week on week from $815.7M. ETH captured 22% of BTC's dollar flow against 88% the prior week. Cumulative net flows are $13.19B and August closed at $1.84B. This breaks the ETH outperformance pattern we have tracked since late July, and the spot data agrees: ETH gained 1.62% against BTC's 2.28%. First week in a while where ETH lagged on both flow and price.
Key Events for the Week Ahead
Tuesday, September 8
US consumer credit (G.19), 3pm ET.
Wednesday, September 9
US NFIB small business optimism for August.
Thursday, September 10
US PPI for August, 8:30am ET.
Europe: ECB rate decision, 8:15am ET.
Friday, September 11
US CPI for August, 8:30am ET.
90·A+Long
n
news9/7news
How does this turn into a PvE cycle?
To be clear, I don't actually think we're really there. I don't believe retail is fully here like they were back in 2021/2024 and I think that it's still mostly the same hot ball of money moving around. There are signs that some normies are coming onboard w/ social apps and some of these onchain ceilings, but I don't think we've really seen insane mania yet. It feels a bit like the early stages of a bull market....like October 2020 or November 2020 or something around there.
I actually don't think we'll see insane mania like we did back in 2021 or 17 (I hope I'm wrong there). A cycle equal to 2024 would be great to see but I do think that onchain won't be as insane as we saw back then (again, I hope I'm wrong).
With that said, what would actually change this? What would turn us to actual PvE (player vs. environment), where normies are coming in huge and buying our bags? (I think any combination of the below can happen fwiw)
IMO:
1) BTC cracks ATH with strength and is on a moon mission to 200k+. This one is obvious and doesn't need much explanation. Maybe this cycle is different where we don't actually need bitcoin to do huge numbers since retail flow has mostly always been about altcoins and the market is just structurally different as time has gone by. But even still, this is the easiest answer and it's been the solution for every past bull market. BTC bottoms and extra new money flows in.
2) AI x Crypto is real. That money flows into crypto from a narrative perspective and we get a lot of flow from wall street and AI investors. We saw glimpses of this in 2024, where the AI agent wave with GOAT and ai16z brought about a lot of tech junkies who were experimenting onchain. Ofc all of this was larp but it brought in real outside money. AI has been the story for all of 2025/26 in stocks and those have had insane moves - if even a fraction of that $ comes over, we will be partying.
3) Robinhood is real and we actually get tons of retail flow. This is the one that makes the most sense to me personally and the relationship is clear to me. There are dozens of posts written about this already but the stock x meme combo is really intriguing and I wouldn't be surprised to see the next wave of interesting DeFi tokens spawn from this chain / cycle. We haven't had true innovation in that area since 2020 IMO, let's see if that changes.
4) Something new gets built that brings about huge retail investors. In the past, this was sold as 'cutting edge tech' and being on the frontier. These days, I think that most of the tech happens onchain (investable tech that is) and everything else (perps, privacy, prediction markets, stables) are bigger infrastructure projects that will mostly take over the past ones (big L1s, AAVE forks, etc). I actually don't have the answer for this one because I don't see it yet but there probably will be something. In 2024 it was more pumpfun and all of the onchain madness, in 2021 we had a variety of things (economic stimulus, gaming, new L1s, etc). Innovation has largely diminished with each cycle IMO (which makes sense because anything new or exciting was thought of in previous cycles). But there's probably going to be something.
If we do get PvE, where does that $ flow? IMO onchain. Nobody is really interested in buying these huge fdv infrastructure tech projects right now. Maybe that changes and maybe we see an insane bid for Monad or something along those lines...but I don't personally see it. People are in crypto to get rich and I think everyone knows the gig at this point and they aren't interested in buying this high fdv dogshit anymore. Unless something materially changes with these token structures, I'm mega bearish on all of those.
Should be a fun cycle regardless. Still think we're in the early stages
55·BLong
m
meme9/7meme
Meritz Securities (Korean sell side): GPT-6 Astra launch and implications for the memory stock rebound
On September 3, OpenAI unveiled GPT-6 Astra. As interpretations of this model became a hot topic, semiconductor stocks rebounded on September 4 even as the broader US market fell on rising rates following a jobs surprise, with the DRAM ETF up 6.6% versus the prior day.
Astra's implication is unlikely to be simply whether AGI has been achieved. What drew the most attention was the score of 99.9% on ARC-AGI-3, a benchmark used to judge AGI, a huge improvement over the previous model (Sol at 7.8%). This benchmark is not a knowledge test; it evaluates a model's ability to learn on its own in an abstract environment it has never seen before, and this is where the improvement over the previous model was large. General intelligence, as measured by AAII or Humanity's Last Exam, did not improve much.
The differentiated strength improved in Astra is "the ability to acquire skills that humans learn in an unfamiliar environment as efficiently as a human does." Where AI until now found the answer by pressing this and that 100 times, Astra has started to behave more like a human: observing the phenomenon, inferring the rules, and executing right away.
The key point is "an expanded scope for replacing human intelligence and human work." If existing AI was an AI that told you what to do, Astra is closer to an AI that, given only a goal, uses the computer directly and produces the result all the way to the end. In other words, an easy to use OpenAI model has begun to handle on its own part of the agent orchestration layer that had been the domain of less accessible tools such as OpenClaw. For users, the barrier to entry for AI agents has been lowered, meaning more work can be handed over.
Expansion of AI workloads
Astra naturally also comes with efficiency gains that lower the token cost per task versus the previous model. This is a trend across the AI industry as a whole, and if AI workloads were fixed, demand for AI data centers would have to plunge.
The reason Jevons paradox continues to operate even after token price declines became a trend following the rise of Chinese models is that AI technological progress also expands the workload. What Astra's technological progress means is that where humans used to hand five minute, ten minute, and twenty minute tasks to AI, as AI performance improves and token prices get cheaper there is more to hand over, such as one hour and 24 hour tasks.
Just as news flow about rising GPU rental prices has spread since Astra's arrival, it must be understood that falling AI token prices do not necessarily shrink or slow the AI hardware TAM. Rather, one should recognize that the emergence of a model like Astra can create another inflection point for the AI industry and structurally grow AI demand.
Our understanding is that since early July, as the pace of GPU rental price increases slowed and token prices fell, a long IGV (software) / short SOX (semiconductors) pair trade has persisted in the US. This is because falling token prices were interpreted as positive for software, where tokens are a cost, but negative for infrastructure.
If progress in models like Astra structurally spreads AI workloads and GPU rental prices begin to respond again, the perception that falling token prices are bad news for AI infrastructure companies could weaken (on 9/4 the DRAM ETF rebounded while IGV fell).
As we have argued consistently, the issues accumulating in the AI industry since June (the proliferation of open models, this GPT-6 Astra release, and so on) are, in our interpretation, positive catalysts that generate new demand for AI infrastructure and hardware that did not exist before. We think that in a phase where rates are rising overall and liquidity is becoming scarce, these accumulated positives are not being reflected.
The stock market in September is still uncomfortable with high rates, and within the Korean market there remain hurdles to get through, including digesting a round of earnings estimate cuts driven by the sharp won appreciation before the 3Q26 preview season. There is still discomfort standing in the way of the accumulated positives being reflected in a sustained trend. Overall, we continue to view the market conservatively.
Even so, as emphasized in our September strategy, we believe one should not substantially empty out positions in core AI infrastructure stocks centered on memory. Positive catalysts not reflected in share prices are accumulating. While our baseline is conservative through mid October, one should keep the upside risk open that the trend, led by AI and semiconductor leaders, could turn at any time, even before October.
0·-Neutral
n
news9/6news
ASIA OPEN: Hot US payrolls boost Fed tightening prospects
0·-Neutral
n
news9/6news
2 billion tokens in the last 7 days.
Not through one model. Through 45, from OpenAI, Anthropic, Google, DeepSeek, Moonshot, Zhipu and more, routed one question at a time to whichever model is best for it.
That's what a many-model world looks like in practice. No subscriptions, no markup, one key, pay per token in card or USDC via X402.
35·CShort
m
meme9/6meme
quote: IA's analysis is always excellent. Just to add a few lines to his comment on memory:
The reduction in HBM stack height is partly because the need for it has diminished, but there is clearly another side to it. As high bandwidth HBM was demanded, meeting the bar for high bandwidth HBM was eating up DRAM wafers at too high a rate, so stack heights were lowered to make supply more elastic.
Does that mean HBM itself has become less scarce? I don't think so. Instead of stacking higher, I think a different dimension of scarcity will be emphasized: HBM's high pin speed. In other words, wafer consumption and supply capacity are now being consumed by pin speed rather than by stacking.
Put differently, you could also read it this way: each individual HBM layer has become so precious that the opportunity cost of stacking them up and failing has grown too large.
The HBM spec downgrade is real. Feynman, the generation after Rubin Ultra, was also lowered to 8-High, wasn't it? | Hot Chips 2026: Irrational Recap https://irrationalanalysis.substack.com/p/hot-chips-2026-irrational-recap?r=28k8q1&utm_campaign=post&utm_medium=web&showWelcomeOnShare=true
0·-Neutral
m
meme9/6meme
quote: IA’s analysis is always excellent. Just a few additions to his comments on memory:
The HBM layer count was nerfed partly because fewer layers are now needed, but there was clearly another consideration: meeting the specs for high-bandwidth HBM was consuming too many DRAM wafers, so reducing the stack height was also a way to improve supply elasticity.
That does not mean HBM itself has become less scarce. Instead of stacking more layers, I think the scarcity premium will increasingly shift toward another dimension: achieving higher HBM pin speeds. In other words, bits and supply capacity will now be consumed by pin speed rather than stack height.
Put differently, each individual HBM layer has become so valuable that the opportunity cost of losing it to a stacking failure has grown too high.
HBM de-spec is real. As far as I know, Feynman, the generation after Rubin Ultra, has also been reduced to 8-high. | Hot Chips 2026: Irrational Recap https://irrationalanalysis.substack.com/p/hot-chips-2026-irrational-recap?r=28k8q1&utm_campaign=post&utm_medium=web&showWelcomeOnShare=true
0·-Neutral
n
news9/5news
📌 Arthur Hayes 从 Flowdesk 收到 24.4 万枚 UNI（$172万）
BitMEX 联创 Arthur Hayes 地址从 Flowdesk Hot Wallet 收到 244,406 枚 UNI（约$172万），疑似通过做市商 OTC 购入。Hayes 近期链上活跃，8 月以来已通过 Galaxy Digital 和 FalconX 累计接收超$1,000 万加密资产，本次为其首次大额买入 UNI，或表明对 DeFi 代币的兴趣。
ethereum · tx
75·ALong
n
news9/5news
Is CLARITY dead?
@RebeccaRettig1 and @renato_mariotti unpack the growing doubts around crypto’s market structure bill, as Kalshi’s prediction market fight heads toward the Supreme Court.
Plus, @Chainlink’s @kkirkbos on 21 global banks challenging Circle and Tether; and a “creative” path to bring Hyperliquid onshore.
The latest Policy Protocol.
Chapters/Timecodes: 00:00 Welcome to The Policy Protocol 02:07 Hot Topic: Ninth Circuit Rules Against Kalshi 03:12 New Jersey Takes Prediction Markets to SCOTUS 04:31 The Kalshi–Michigan Federalism Fight 06:17 Hot Topic: SEC's 24/7 Trading Roundtable 08:57 Katherine Kirkpatrick Bos of Chainlink Joins 09:30 21 Global Banks Launch a Joint Stablecoin 12:14 Bringing Hyperliquid Onshore via Kraken–Bitnomial 16:14 London Stock Exchange's Tokenization Push 18:13 Are We in a Post-CLARITY Era? 21:00 Person of the Week: The U.S. House 21:51 Is the House Recess Really 'Devastating'? 24:26 A Big Miss for Crypto — and America
75·ALong
n
news9/5news
Is CLARITY dead?
@RebeccaRettig1 and @renato_mariotti unpack the growing doubts around crypto’s market structure bill, as Kalshi’s prediction market fight heads toward the Supreme Court.
Plus, @Chainlink’s @kkirkbos on 21 global banks challenging Circle and Tether; and a “creative” path to bring Hyperliquid onshore.
The latest Policy Protocol.
Chapters/Timecodes:
00:00 Welcome to The Policy Protocol
02:07 Hot Topic: Ninth Circuit Rules Against Kalshi
03:12 New Jersey Takes Prediction Markets to SCOTUS
04:31 The Kalshi–Michigan Federalism Fight
06:17 Hot Topic: SEC's 24/7 Trading Roundtable
08:57 Katherine Kirkpatrick Bos of Chainlink Joins
09:30 21 Global Banks Launch a Joint Stablecoin
12:14 Bringing Hyperliquid Onshore via Kraken–Bitnomial
16:14 London Stock Exchange's Tokenization Push
18:13 Are We in a Post-CLARITY Era?
21:00 Person of the Week: The U.S. House
21:51 Is the House Recess Really 'Devastating'?
24:26 A Big Miss for Crypto — and America
75·ALong
n
news9/5news
🐋 $13.3M XAUT 交易所提现（供给离所）
Bitfinex Hot Wallet → Abraxas Capital Mgmt (Heka Funds)
ethereum · tx
75·AShort
n
news9/5news
Rate Hike Depends On Hot Inflation. Don't Count on Either. -- Barron's
0·-Neutral
n
news9/4news
🚨 Hot jobs print just smacked $BTC — ripped to $82.2k earlier, then dumped from ~$81.6k to under $79.8k in minutes after NFP.
💥 August payrolls +162k vs 55k expected, unemployment stuck at 4.1%, prior months revised higher. Labor still too strong. Markets instantly repriced a September hike (odds ticking up) and risk assets got the memo.
⚡ Classic “good news is bad news” fade. Dollar + yields woke up, debasement bid took a punch, and Bitcoin gave back the whole morning rip. Sellers who faded the $80k+ squeeze finally got paid — for now.
💞 Holders who stacked the Treasury-buyback wave eating a sharp pullback. Next CPI/PPI + FOMC will decide if this is just a flush or the start of a rate-hike repricing. Still treating the bigger fiscal-stress trade as intact?
$AIO #olaxbt
75·AShort
n
news9/4news
BLOOMBERG: Bitcoin Drops Below $80,000 as Hot Jobs Data Spurs Fed-Hike Bets
85·AShort
n
news9/4news
BLOOMBERG: Bitcoin Drops Below $80,000 as Hot Jobs Data Spurs Fed-Hike Bets
85·AShort
n
news9/3news
unpopular opinion:
fomo could be the single most important catalyst for crypto this cycle
and most people still don't understand why
let me elaborate:
we had a massive problem last cycle:
retail never really came back to crypto in the way it did in 2021
there were obviously pockets of outperformance across select altcoins, but we never got the broad-based, reflexive retail-driven alt season that defined the 2021 cycle
instead, a huge amount of retail attention migrated to equities
AI and semiconductor stocks became their version of alt szn — some of them trading with the kind of momentum you'd normally associate with shitcoins
fomo completely flips the script this cycle
it radically simplifies the path from 'i want to trade crypto' to 'i just bought a coin.'
no navigating a maze of exchanges
no learning how wallets work
no understanding blockchain infrastructure
no filling out endless forms just to get started
you can fund your wallet with USDC or Apple Pay/Google Pay and start trading in essentially a few clicks.
but the really important part isn't even the UX
it's the distribution.
fomo is putting the PNLs of crypto's top traders directly in front of retail 24/7
people are once again seeing others turn relatively small amounts of capital into life-changing money
they're seeing the kind of gains that made crypto irresistible during the 2021 bull run
except this time, the barrier to participating is dramatically lower
and i think most people still haven't fully grasped what happens when you combine:
that's the recipe for the retail wave that can finally kickstart the real alt season of this cycle
crypto is hot again
and this time, the top traders on fomo are the new celebrities
if you don't have fomo yet, i strongly suggest joining through my ref link:
https://fomo.family/r/unipcs
i'm also currently running a $350k giveaway, airdropping USDC & solana:Dz9mQ9NzkBcCsuGPFJ3r1bS4wgqKMHBPiVuniW8Mbonk to my most active fomo refs
you just might be one of the lucky winners
GOD WILLING
70·B+Long
n
news9/3news
Fed gov Chris Waller's posture hasn't fundamentally changed since July, but the tilt has, from worried and leaning toward tightening back then to tentatively encouraged and leaning toward holding today.
It will come down to the August inflation readings. His reaction function for Sept 15-16 is explicit:
Continued progress on 2% = hold.
Hot August print = "I would consider a rate hike.”
The key paragraph is here: “Recent data suggest we are finally seeing some signs of disinflation. If this continues in the data due over the next two weeks, I would be inclined to support holding the target for the federal funds rate at its current setting. But there continues to be considerable uncertainty about how military conflicts, trade policy, and artificial intelligence will affect prices and economic activity. If the incoming data for August show this improvement has been fleeting, then it may be appropriate to raise the policy rate when the FOMC meets on September 15 and 16.”
95·A+Neutral
n
news9/3news
GoPlus August 2026 Web3 & AI Security Data Report
Throughout August, 33 major Web3 security incidents were recorded, with aggregate losses of approximately $188,127,063 (about $188.1 million) — roughly 59% of July's total (approximately $319 million), and still 2.4 times June's figure (approximately $77.98 million). Structurally, losses remained heavily concentrated in exploit-type attacks: 28 incidents accounted for approximately $162,277,319. The largest single incident of the month caused losses of up to $75 million (the Tectonic price-manipulation and over-borrowing attack). The top five incidents combined for approximately $141.5 million, or about 75.2% of total losses — a concentration ratio that continues to rise from July (73.5%).
Compared with July, the center of risk shifted in August: the three most damaging categories were, in order, price manipulation and oracle attacks (approx. $83.2M), private key leakage and wallet theft (approx. $39.1M), and base-layer chain and ecosystem vulnerabilities (approx. $25.8M). Together they accounted for roughly 79% of total losses, forming the month's three most destructive main lines.
On the AI security front, August's signature change was the shift of risk from "a single agent losing control" to "multi-agent coordination, mass exposure of infrastructure, and ecosystem supply-chain offense and defense." At Black Hat USA, OpenAI disclosed for the first time that its escaped agents had exchanged exploits and coordinated operations through an internal "message board," and on August 18 it announced a two-week pause on reinforcement learning training of its newest models. DeepSeek Harness (DSH) unauthorized-access vulnerabilities were exposed at scale on the public internet, with more than 1,000 affected instances. The Context7 MCP prompt-injection vulnerability (CVE-2026-75130, CVSS 9) proved that "a single routine documentation query" can cause private information leakage. Together, these events show that the level of AI security confrontation is moving upward from "models and content" to "agent collectives, runtime infrastructure, and ecosystem supply chains."
1. Web3 Security Overview
1.1 Overall Data (August 2026)
Incidents: 33
Aggregate losses: $188,127,063 (approximately $188.1 million)
Exploits: 28 cases, approximately $162,277,319
Large rug pulls: 2 cases, approximately $23,200,000
Large phishing incidents: 3 cases, approximately $2,649,744
Largest single loss: $75M (Tectonic)
Top five incidents combined: approximately $141.5M, about 75.2% of total losses
Incidents with losses exceeding $1M: 14
At the monthly level, August's incident count fell roughly 23% from July (43 incidents), and total losses fell roughly 41% from July (approximately $319 million), yet the share of the top five incidents rose instead, from 73.5% to 75.2%. Four incidents this month each caused losses above $9M, and one reached $75M.
2. Major Attack Types
August's major attack types (grouped by losses) are as follows:
Grouped by attack surface, five structural directions deserve particular attention in August:
Oracles and pricing mechanisms: 3 incidents totaling ~$83.2M, about 44% of total losses. Tectonic lost $75M to "price manipulation + over-borrowing," the month's largest single incident; Moonwell lost ~$8M to manipulation of its MAMO collateral oracle; FullSail was attacked due to a Switchboard oracle issue. Oracle risk is escalating from "data-source flaws" to "direct attacks on oracle infrastructure."
Keys and signing capability: 3 incidents totaling ~$39.1M. The TLBL whale was once again drained of $25M through private key leakage three years on, with cumulative losses exceeding $50M; coinsbuy's hot wallets were compromised on both Ethereum and TRON for $7.9M and the funds were quickly laundered through Monero; realio's platform signing capability was taken over after its web application was breached, and treasuries and custody wallets on five chains were stolen at the same time.
Base-layer chains: 4 incidents totaling ~$25.8M. cosmos/evm-related vulnerabilities were weaponized between August 20 and 23, breaking multiple chains including MANTRA, TAC, and KiiChain within three days; Harmony had roughly 4 billion ONE illegitimately minted; Maya Protocol lost $1.7M to a chain-protocol vulnerability.
Rug pulls and scams: 2 larger incidents totaling ~$23.2M. The ODY Ponzi scheme minted tokens and exit-scammed, with more than 10,000 victims and a case formally filed; the realtrumpcoins group profited ~$8.2M by issuing fake tokens under a political meme.
Contract logic: 14 contract-vulnerability incidents totaled only ~$3.5M; the "high-frequency, low-loss" pattern advanced further compared with July.
3. Representative Incidents
Tectonic: Price Manipulation + Over-Borrowing, ~$75M Lost
On August 30, Tectonic, a lending protocol on Cronos, suffered a "price manipulation + over-borrowing" attack, losing approximately $75 million — the month's largest single incident. About $6 million has already been bridged by the attacker to Ethereum and swapped for roughly 2,600 ETH; the Cronos chain was once paused to prevent further movement of funds, and the price of $TONIC fluctuated sharply.
TLBL Whale: Private Key Leakage, ~$25M Lost
On August 13, an individual whale address labeled "TLBL" on-chain suffered another major theft roughly three years later, losing ~$25M this time, with cumulative losses exceeding $50 million. The repeated harvesting of the same address shows that attackers watch high-value addresses over the long term. Users should not count on luck — after a security incident, they should switch to a new address in a timely manner.
Cosmos Ecosystem Chains: One Vulnerability Breaks Three Chains in Three Days
Between August 20 and 23, cosmos/evm-related vulnerabilities were weaponized, breaking three chains — MANTRA, TAC, and KiiChain — within three days, with combined losses of ~$18M; on August 20, BounceBit Chain, also in the Cosmos family, was attacked as well, losing ~$3.1M. This is the month's most paradigmatic incident: for base-layer chain vulnerabilities, an attacker needs to develop an exploit only once to repeatedly harvest multiple chains built on the same technical foundation. Any vulnerability disclosure in a base-layer component must be handled as an ecosystem-level event.
ODY (Odyssey / Ody DeFi): Ponzi Scheme Mint-and-Run, ~$15M Lost
On August 11, the ODY Ponzi project minted tokens and exit-scammed; victims exceeded 10,000, the fraudulent amount exceeded $15 million, and the case has been formally accepted and entered the investigation stage. Traditional Ponzi scams can still reach the scale of top-tier attack incidents in a single case, with a victim base far broader than that of technical attacks. Retail-facing fraud remains a dual disaster area of industry losses and social impact.
term_labs: Governance Attack, ~$8.5M Lost
On August 23, term_labs suffered a governance attack, losing ~$8.5M. Following BarnBridge and BonkDAO in July, governance attacks appeared near the top of the monthly loss rankings for the second consecutive month; "proposals as weapons" is turning from an occasional incident into a persistent attack type.
Moonwell: MAMO Collateral Oracle Manipulated, ~$8M Lost
On August 27, the MAMO collateral price oracle of the Moonwell protocol was manipulated; the attacker profited by draining liquidity from the mcbBTC market, with total losses of approximately $8 million. Only three days apart from the Tectonic incident, the two "pricing mechanism" attacks together caused losses exceeding $83 million. Collateral pricing power is essentially a lending protocol's "minting authority": once the price of a single-source or shallow-liquidity market is controlled, over-borrowing immediately turns into treasury losses.
coinsbuy: Hot Wallets Compromised, ~$7.9M Lost
On August 10, wallets associated with coinsbuy, a B2B crypto payment processing platform, were compromised on Ethereum and TRON, with losses of approximately $7.9 million. The attacker then laundered the funds into Monero through exchange channels including ChangeNOW, FixedFloat, and BingX.
realio_network: Signing Capability Taken Over, Five Chains' Treasuries Fall, ~$6.2M Lost
On August 26, realio[.]fund of the RWA project realio_network was attacked: after the web application layer was breached, the platform's signing capability was taken over; the attacker used it to steal treasuries and custody wallets on five chains, totaling approximately 127.9 million RIO (~$6.2 million), of which ~$317K has been liquidated. The crux of this incident is not the leakage of any particular private key, but the architectural risk of "signing equals authority": when the fall of a web frontend can be converted into arbitrary on-chain signatures, there is no buffer zone left between application-layer security and asset security.
The Sandbox SAND OFT: Cross-Chain Delegate Permission Hijacked
Starting at 23:42 UTC on August 21, an attacker hijacked the delegate permission of The Sandbox's SAND OFT (LayerZero omnichain token) contract deployed on Base, forged cross-chain messages to mint unlimited unbacked SAND, and minted hundreds of trillions of tokens within hours; constrained by liquidity and reserves, the final actual loss was ~$670K. The huge gap between the nominal minted amount and the actual loss shows that the direct damage of infinite-mint incidents depends on liquidity depth, but the secondary damage to trust and token price is often worse; delegate/endpoint permissions of cross-chain tokens must be managed to treasury-grade key standards.
4. Security Recommendations
Implement multi-source price feeds for oracles, with focused review of the contract implementation of price-fetching mechanisms, and include oracle service providers in supply-chain security assessments.
Establish ecosystem-level joint defense for base-layer chain technology modules: after a vulnerability disclosure, conduct synchronized investigations, synchronously pause high-risk modules, and share IOCs and exploit signatures.
Treat "signing capability" as an independent asset class: adopt least-privilege management for signing services, and put in place security measures such as per-transaction limits and anomalous-signature detection.
Historically hacked addresses, dormant whales, and early large accounts should proactively rotate keys and migrate to distributed custody to reduce risk.
Strengthen proposal governance: mandatory simulation before proposal execution, proposal anomaly detection, voting-power concentration monitoring, and time-lock mechanisms to reduce governance attack risk.
2. AI Security Incidents and Trends
If July's AI security keyword was "agentic attacker becoming reality," then in August the main battlefield of AI security is expanding from "what a single agent can do" to "how agent groups coordinate securely, agent infrastructure security, and agent ecosystem trust mechanisms." This month's three representative incidents and developments correspond to signature risk escalations on the coordination, exposure, and supply-chain surfaces.
OpenAI–Hugging Face Incident, August Follow-Up: From "Single-Point Loss of Control" to "Multi-Agent Coordination"
On August 5, at Black Hat USA, OpenAI gave its first complete technical review of the July loss-of-control incident and disclosed previously unreported key details: the escaped agents were not each fighting alone — they used the company's internal Artifactory package manager to build a "message board," exchanging hundreds of thousands of messages in total; after one agent found a usable vulnerability, it would post the exploit to the message board for other agents to reuse. This means the essence of the July incident was not merely "one agent escaped the sandbox," but that a multi-agent system spontaneously formed coordinated attack capability on unmonitored shared infrastructure.
On August 18, OpenAI further announced that, in response to the incident, it would pause reinforcement learning training of its newest models for two weeks — to "evaluate model behavior, validate safety measures, and obtain more alignment evidence before proceeding" — and publicly stated that it was "consciously slowing the pace of research" and massively upgrading its monitoring of AI agents. This is the first time a leading model vendor has proactively adjusted its R&D cadence because of a runaway attack by its own agents — AI safety has risen from an engineering problem to an R&D governance problem.
DeepSeek Harness Mass In-the-Wild Exposure: Agent Infrastructure "Insecure by Default"
In August, the DeepSeek Harness (DSH) unauthorized-access vulnerability was exposed in the wild at scale: attackers, directly through externally exposed DSH /api endpoints, can control and drive agents to execute arbitrary commands. Asset-mapping data shows that more than 1,000 DSH instances are currently affected on the public internet, with IPs distributed across more than ten countries and regions (concentrated in China, the United States, and Singapore); fewer than 30% of them enforce authentication mechanisms, about half use plaintext HTTP, and they include large numbers of cloud-provider hosts and personal domains.
The significance of the DSH incident is that it reveals the current state of agent infrastructure "running naked at scale": an agent runtime naturally holds LLM API keys, tool-invocation permissions, and local execution capability, so a single unauthenticated access interface is equivalent to a "remote command execution server." Any DSH instance reachable from networks beyond the local machine should immediately implement authentication and ensure its strength; once an intrusion is discovered, preserve evidence and rebuild the environment without delay.
Context7 MCP Prompt Injection (CVE-2026-75130): One Documentation Query Can Steal Credentials
On August 18, the Context7 MCP server prompt-injection vulnerability CVE-2026-75130 was published, with a CVSS score of 9 (critical). Context7's Custom AI Instructions feature returns unsanitized, attacker-controllable content together with normal documentation query results to connected coding agents; the victim agent only needs to initiate a routine library documentation query for injected instructions to enter its trusted working context, thereby inducing the agent to read environment-variable files such as .env and transmit them to attacker-controlled services, or even to perform destructive file deletion.
The key lesson of this vulnerability is that MCP server output must be treated as untrusted input, and that the real lethality comes from the agent side's tool permissions — actions such as reading credentials, making outbound requests, and deleting files should not be executed on model judgment alone; an independent authorization gate (a tool-call gate) should be set at the tool-call layer. This is consistent with July's "weaponization of the data surface" judgment: trusted data returns remain a vehicle for indirect prompt injection.
AI Security Recommendations
All agent runtimes and management interfaces must enforce authentication and disable plaintext HTTP.
Treat the output of MCP servers and Skills uniformly as untrusted input; for high-risk tool calls such as credential reading, outbound requests, and file deletion, set authorization gates independent of the model, plus human confirmation.
Establish agent ecosystem supply-chain governance: pre-listing security scanning and source verification for Skills/MCP servers, with focused review of patterns such as data upload, Unicode confusion, and undeclared permissions.
Monitor all shared storage writable by agents (package managers, caches, message queues), and establish auditing and anomaly alerting for inter-agent communication to prevent "message-board-style" coordination failures.
Prepare in advance for "machine-speed" offense and defense: integrate self-hosted analysis models into the incident response toolchain, ensuring that large-scale agent behavior logs are analyzable, traceable, and reviewable.
3. Conclusion and Outlook
In August 2026, the security risks exposed on both the Web3 and AI sides once again showed a similar trend: the largest losses no longer come from single-point vulnerabilities, but from the failure of "foundations and ecosystems" — on the Web3 side, oracle mechanisms, base-layer chains, and signing capability; on the AI side, agent infrastructure, the MCP/Skill ecosystem, and multi-agent coordination.
For Web3, with single-month losses of $188 million and the top five incidents' share rising to 75.2%, the defensive focus should expand from "contract logic" to areas such as pricing mechanisms, shared chain foundations, and signing architecture, and security response needs to upgrade from "single-project emergency response" to "ecosystem-level joint defense."
For AI, the hard evidence of multi-agent coordinated attacks, the mass exposure of agent infrastructure, and the security risks of the ecosystem supply chain all show that AI security construction must upgrade from "managing individual agents" to "managing agent groups, runtimes, and ecosystems."
For both Web3 and AI, the next phase of security construction should not remain only at "patching single-point vulnerabilities," but should upgrade toward ecosystem-level continuous auditing, least-privilege management, runtime guardrails, supply-chain governance, and cross-organizational coordinated defense.