One week, Bitcoin ETFs were bleeding money. The next, they pulled in $626 million in three days.

According to SoSoValue data, US spot Bitcoin ETFs had a net outflow of $61.53 million in the week ending July 27, marking the first time in three weeks that the ETFs posted net outflows.

In contrast, spot Ether ETFs saw $27.42 million of inflows during the same week, adding to a string of four weeks of positive inflows.

The Bitcoin ETFs compounded three days in a row of inflows, reaching a total of $626 million, with the largest inflow of $244.4 million on a single Wednesday.

BlackRock's IBIT did most of the work, gathering $479 million over the period and now taking the lead in terms of overall net inflows with cumulative net inflows of $61 billion, far ahead of any other fund.

It's a common occurrence for ETF flow data, but a nice reminder for traders eyeing a single headline that a weekly snapshot can be misleading.

Isolated one-week outflows seemed to be a sign of waning institutional interest. The same product category experienced one of its best outflows in months three trading days later.

Ethereum's ETFs followed suit, totaling $60.9 million over that same Wednesday to raise a total of $114.6 million over the previous two days. XRP ETFs had the opposite trend, with $3.58 million in outflows despite cumulative net inflows over $1.5 billion.

The lesson learned: One week of ETF data is not necessarily reliable. The trends in cumulative flows, and which fund is truly collecting the cash, are difficult to overlook here, and are far more significant when viewed as a whole than when considered on an individual basis.

Following the flow of money to the Bitcoin price, BTC saw a quick spike to above $64,900 before dropping near $64,700. The price action was mild in comparison to the flow swing that it represented.

The $687 million net swing over the course of about a week only made a slight dent in spot price, as ETF flows tend to impact sentiment far more than to singlehandedly drive short-term moves in the market of this size.

If you step back, you can see Bitcoin ETFs are still in their infancy as a product category and are still getting the hang of things. Weeks of outflow alerts will continue to trigger alarms, and 3-day inflow streaks will continue to create hype. Traders who are only paying attention to one data point, rather than the trend, could be prone to being “whipsawed” in both directions.

Do ETF flows still move your trading decisions, or has the signal gotten noisier over time?

Not financial advice.

A five-year-old bug just cost Bitcoin holders over $100 million. Nobody touched their hardware wallet to lose it.

Coinkite servers for their Coldcard Mk3 devices are supplying weakened software-based randomness functions rather than the built-in security chip to create the seeds. The bug was dormant since 2021.

Researchers observed more rounds of attacks start flowing in late July, the wallets losing significant amounts of money, and losses have topped $100 million at hundreds of addresses so far.

This is a mechanics thing. The seeds with entropy contained about 40 bits, which is well below the 128-bit goal of a secure 12-word phrase. It's not a weakness; it's a difference that an attacker could exploit without ever touching the device itself, simply by brute-forcing it offline.

Affected are Mk2 and Mk3 units that are running the affected firmware. There is a less serious but actual risk with Mk4, Mk5, and Q devices prior to their solid releases.

Wallets that are constructed using the dice-roll entropy option with physical randomness added on top are considered safe.

If you have an affected device, there's no hard part about the fix, but it is urgent. Check your firmware version, confirm to the patched version, create a new seed on the new firmware, and transfer the funds in increments, testing with a small amount before transferring the larger amounts. Don't delete the old backup until the new wallet is completely verified.

Again, there is a debate self-custodians rarely like having to engage in, and that is that exchange failure or custodian insolvency is replaced by other risks with the purchase of a hardware wallet.

Even if your keys never connect to the internet, firmware bugs, supply-chain tampering, and phishing don't go away. Both have the effect of moving exposure, but not eliminating it.

The incident is the third significant seed generation failure of the type in recent years, according to researchers tracking the incident.

The bug was present at wallet creation time, which users could not practically check after they created their wallet.

The uneasy part for those who are not holding BTC wrapped in an ETF or exchange is that it would seem to be a sign of a deeper market crash. No longer is it a mere one-off job to check firmware or create new seeds; it's more like a maintenance job.

But for those who don't want to make wallet security a full-time duty, the regulated custodians and spot ETNs are increasingly portrayed as the compromise solution, albeit it's losing some of the element of self-custody.

Self-custody or trusted custodian, where do you land after this one?

Not financial advice.

Bitcoin is trading well above the price most long-term holders paid. That's usually good news. Right now it's a warning sign.

The on-chain data shows that the realized price for long-term holders (BTC held for 155 days or more) is around $49,700.

Bitcoin is trading around $64,300 this week, and those who have held on to their coins for the long haul have decent unrealized gains. But the same group is being identified as showing capitulation signals.

Here's the tension. Long-term investors are expected to be the "strong hands", those who put their money in at low prices and trade with style when the market is volatile. If they begin selling when they're profitable, it's usually due to weak conviction, not a lack of cash.

It is a bit rougher to read the short-term accountings. Their actual cost is around $69,000, and that means that new investors are in negative equity.

It's a well-established late cycle trend that we've seen in the past: when the market has started to correct, those who have been in for a while are offloading, and those who have joined more recently are in fear.

The lesson of history is important here. Bitcoin had been through such sell-offs before, but it didn't really lead to a short-term low. When the rangebound price occurs for months, and on-chain metrics take time to process, that's exactly what's seen throughout this summer.

Traders: keep an eye on whether the current cost basis around $69,000 is retrieved. That has been a resistance point, and every time, analysts are talking about it as a point where a real attempt at recovery is going to begin or stop.

There's another track to keep an eye on, too. Cost basis gap tends to narrow prior to big moves.

This can occur when the price of the coins meets up with the short term holders, or because the price of the long-term holders' own average cost increases as the coins age into the group.

That's an unusually large spread right now, and on-chain analysts see this as proof that the market isn't fully processed through this correction.

In the past, extended periods of strong paper gains by long-term investors have been followed by either false or true bottoms, and capitulation rates remained high.

The key is that it usually ebbs and flows over a period of weeks instead of months. This is still a young one where either scenario is still possible.

There is no capitulation among believers that approaches what is happening among tourists. The question is which one do you watch right now?

Are you reading this as a bottoming signal or more downside ahead? Let's hear your chart.

Not financial advice.

Vanguard didn't mean to buy Bitcoin exposure. It happened anyway.

The asset manager has just purchased an additional 269,200 shares of Strive Inc., a Bitcoin treasury company, to raise its holdings to approximately 1.98 million shares valued at approximately $23.7 million. This wasn't a deliberate bet on Bitcoin.

It was mechanical index rebalancing: mechanical, that is, because they are done without human intervention when a stock's market cap reaches a size significant enough to warrant a rebalancing.

As a result, Strive currently has Bitcoin in its corporate treasury as part of its overall corporate treasury management program, along with preferred stock holdings and other holdings.

Each time Strive increases its equity holdings to acquire additional BTC, its market value changes and passive index funds such as the VTSAX must increase their holdings of BTC to reflect the market. Vanguard does not have to trust in Bitcoin. The flywheel works regardless of the flywheel.

This isn't the only indicator of quiet institutional creep into Bitcoin treasury names.

A Canadian pension fund revealed its multi-million dollar investment in Strive earlier this year in a routine filing known as a 13F, which is paperwork that generally doesn't garner much attention but eventually makes itself known in the numbers.

Just three weeks prior to the Vanguard filing, the company announced the first-ever role for a "Head of Digital Assets.

There's the passive exposure and the internal hires for the digital asset infrastructure, which would seem like more than a coincidence, though no one at Vanguard has confirmed that it is a Bitcoin bet in public.

This is how institutional adoption really works – not through a press release, but through index mechanics and a quiet org chart. The signs are slower and happen so much that retail investors looking for the latest news don't even pick them up.

The takeaway from traders isn't that Vanguard is bullish on BTC. This has to do with the fact that there are now large enough Bitcoin treasury companies to move passive fund flows on their own, which wasn't even the case two years ago.

Do passive inflows like this matter more or less than direct spot ETF buying? What's your read?

Not financial advice.

Google just revealed a $94.1 billion stake in SpaceX. It can't sell a single share of it.

The value of the position swelled in Alphabet's Q2 filing, with about $80 billion of it locked up under common IPO restrictions and another $14.1 billion set to remain frozen until late 2027. The gain exists on paper — no shares have actually been sold.

The truth lies not in the stock stake. This is what's to pay for it. Alphabet also increased its capex outlook for 2026 to $195–205 billion, from the range of $180–190 billion in April. That's mainly chips, data centers and power.

In June, Google made a deal with SpaceX for $920 million monthly for AI computing power, which includes over 110,000 Nvidia GPUs that will be available until mid-2029.

This is the trend that crypto traders need to observe. Big Tech is not only purchasing GPUs, but it's also purchasing entire compute ecosystems, and the capital intensity is changing the next destination of money.

DePIN projects that the lease out decentralized compute, storage, and bandwidth exist because there exists a supply gap where smaller players can accommodate centralized capex at this scale.

AI-token narratives have calmed down from the 2024 hype cycle, but the idea that the demand for AI requires more compute than hyperscalers alone can provide has received additional backing.

For a company as large as Alphabet, spending nine figures per month to gain access to GPUs makes decentralized alternatives seem less and less like an exotic gamble and more like infrastructure arbitrage.

The risk-off twist is there as well. Alphabet reported a profit of $99 billion, which was largely on the back of unrealized gains in equity, while the company's core ad or cloud business is responsible for only a fraction of the company's earnings. Such paper gains get their value discounted by markets.

As sentiment has become increasingly correlated between the valuation of AI-adjacent stocks and BTC/ETH over the last two years, any wobble in the former inevitably bleeds over into the risk asset space more broadly, and that includes crypto.

Be mindful of capex guidance season. In the past, the number of companies in this list has always been a precursor to the renewed surge in interest around AI and DePIN tokens in the following few weeks.

So, is Big Tech's AI spending spree bullish or bearish for the AI-token story of crypto? Make your guess below.

Not financial advice.

China's Chip Equipment Demand Keeps Climbing — What It Signals for Decentralized Compute

A Shanghai-listed equipment maker's latest earnings report offers a microcosm of the larger AI hardware story

Shengmei Shanghai, a maker of cleaning equipment for semiconductors, said its annual bottom line grew to about 6.79 billion yuan, around 21% higher than a year before, and its net profit attributable to shareholders also increased about 21%, in company filings.

The result in its first quarter of 2026 was more of a mixed bag: revenue grew 13.06% year-on-year to approximately 1.48 billion yuan, but net profit dipped 57.66% year-on-year.

The company, which blamed foreign-exchange losses on currency fluctuations, primarily for its foreign investments, as the main cause for the decline, blamed the slowdown not on any loss of demand for its equipment, but on foreign exchange.

New order volume was a different story, as the company's shipments in the first quarter of 2026 rose 48.65% year-on-year and new signed orders were roughly 65% higher than those seen in the same period of 2025, signaling continued robust momentum for the remainder of the year.

The larger story is the domesticization of the semiconductor equipment industry as China looks to bring more semiconductor manufacturing, in particular cleaning machines and plating and advanced-packaging equipment, back to domestic shores.

The year ahead is likely to continue a trend of growth in the sector for global semiconductor equipment sales, as demand for AI chips remains strong, according to analysts.

That demand chain eventually arrives at decentralized compute networks as well. Such as Akash Network, it creates marketplaces directly on-chain for leasing GPU capacities.

Any increase or decrease in global chip-equipment supply will trickle down to the amount of compute global networks will eventually be able to access or provide.

According to CoinGecko, AKT is trading at almost $0.47, which is approximately 94% lower than its 2021 all-time high of $8.07, with a market cap of approximately $142 million.

Order books for equipment and token prices are moving on different timetables, but at the same underlying trend, whether it be the factory floor or the blockchain.

This is not financial advice.

Do you watch semiconductor-equipment earnings as a leading indicator for AI-linked crypto tokens?

AI Just Scored Under 1% on a New Test — Here's Why That's the Whole Point

The most sophisticated AI models in the world nearly failed a test, and the creators are rejoicing.

The ARC Prize team will launch ARC-AGI-3 in March 2026, which creates AI agents and lets them enter into an interactive game space without any instructions, goals, or rules to follow.

They are easily solved by humans. The Frontier models, on the other hand, don't fare as well as Google's Gemini, which achieved a score of 0.37%, OpenAI's GPT with 0.26%, or Anthropic's Opus with 0.25% in preliminary tests, as reported by ARC Prize data.

That's not a coincidence. ARC-AGI creator François Chollet has pointed out that the actual problem with AI is not that the language models are large — it is agentic reasoning: figuring out a new environment, crafting a goal, and adjusting a plan based on the environment, without being directed in what to do.

ARC Prize 2026 has more than $2 million in awards to be distributed, with the top open-source agent winning the $700,000 grand prize for achieving 100% on the benchmark.

This is important for crypto's AI-agent space, as it's a rehashing of the rivalry. Projects with a scale focused on static language models might be more successful in the long run than those simply hawking the "AI" label but not shipping real agentic functionality.

Bittensor is among the few networks based on this concept: miners and validators receive TAO for the value of the models they bring to specific subnets.

According to CoinGecko, the current price of TAO is approximately $191, which is approximately 75% off its all-time high of $757.60, and the market cap is currently hovering around $1.8 billion.

It is undecided whether decentralized subnets can be able to crack problems like ARC-AGI-3 before centralized labs.

However, the benchmark is a reminder worth noting: agentic intelligence remains largely unsolved, and that's where agent-economy tokens are looking to make their mark.

This is not financial advice.

Do you think decentralized AI networks have a real shot at cracking problems like this before the big labs do?

Ready Is Moving Its Card Program — What Self-Custody Users Need to Know

Self-custodial wallet Ready is transitioning to another network, and existing wallet users must look out for a deadline.

In mid-August 2026, the wallet, previously known as Argent, confirmed that it will shift its mobile app from Starknet to a prominent Ethereum layer-2 network. The new version will prioritize holding and spending digital dollars, instead of all of Starknet's DeFi features.

The Ready Card is currently being phased out, and the company is replacing the card issuer. This doesn't depend on the network migration, says Ready.

It will provide more information about card availability and eligibility nearer to the launch. Those who experienced a change in issuers earlier in the year and were banned from using services outside the UK and EEA should be alert for changes and an update on whether they will be able to use services in their region again.

If a user has already deployed assets on Starknet, there will be no automatic transfer. Ready is self-custodial, with only a user being able to authorize a transfer.

When the app gets updated, a new account will be automatically generated on the L2 network, but will not have any Starknet tokens.

Users holding NFTs, DeFi assets, or other non-translatable assets are encouraged to transfer their holdings before August 15, as Ready X is a self-custody wallet that will continue to work on Starknet indefinitely. That's the quickest and easiest route, Ready says.

Ready suggests creating a separate email account to be used for the Ready X account. It also cautions people not to accidentally send Starknet tokens to any other L2 or Ethereum address, as these transactions can be irrevocable and will result in the loss of funds.

NFA.

If you hold assets on Ready's Starknet account, have you already moved them to Ready X, or are you waiting until closer to the deadline?

The US Just Restricted New Chinese Humanoid Robots — Why AI-Agent Tokens Should Care

Washington just drew a line around humanoid robots, and the crypto AI-agent business is standing right next to it.

The FCC on July 28, 2026, placed foreign-made humanoid and quadruped robots on its “Covered List,” and denied permission to authorize new devices on the grounds of national security.

The rule only applies to robots that haven't been approved for US sale—those that have been approved, including many from market leader Unitree, can continue to be sold. This doesn't impact existing owners either.

According to Counterpoint Research, Unitree has a market share of almost 20% in the global humanoid robot market. While you don't have to worry about the number of humanoids deployed in the US, in TrendForce's next forecast, it is estimated that China will ramp up its humanoid manufacturing by 94% in 2026, which means that the on-ramp is also relatively small.

Why do AI-token traders need to be concerned with the rules governing the import of these AI tokens? Humanoid robotics are joining on-chain AI agents—it is because of this that the two are converging.

What crypto "agent" networks are attempting to construct are systems that can perceive, then make decisions, take actions, without constant human involvement, and coordinate with other systems to accomplish a task – which is what physical robots are expecting to do.

At the heart of that agent-economy story is the token of the Artificial Superintelligence Alliance (FET), which is the result of the merger between Fetch.ai, SingularityNET and Ocean Protocol.

Per CoinGecko, FET trades in the $0.14–$0.16 range, with a market cap near $320–$415 million, well off its March 2024 high of $3.45.

A robot import restriction does not affect the flow of FET's code or activity in FET's subnet.

The point is, though, that it's not just about hardware; it's a live sign of the seriousness governments are taking towards autonomous physical AI as strategic infrastructure — and infrastructure-level attention is sure to make its way to the tokens funding the software side of the same trend.

This is not financial advice.

Do you see humanoid robotics and on-chain AI agents as one trend or two unrelated stories?

Rare Earths, Real Stakes: What the Minerals Push Means for AI-Compute Crypto

The group of mining CEOs convening with US authorities is no crypto story — unless you dig into the backstories of their supply chain to trace the GPUs behind every AI token you follow in your watchlist.

On August 7, 2026, the State Department hosted a roundtable with executives from Rio Tinto, BHP, Freeport-McMoRan, MP Materials and other critical-minerals producers.

The declared objective: to reduce the U.S. dependence on China's dominant rare earth and antimony supplies for use in weapons, energy and advanced chip-making.

This is a secret that connection traders are missing. Rare earths and specialty metals are not limited to magnets and batteries; they are used as inputs for semiconductors that fuel AI training and inference.

Every point of turbulence in that pipe ends up in the GPU supply chain, and the whole business of a decentralized compute network is about supplying GPUs.

The most obvious one is Render Network. It operates a market for GPU owners to be paid in RENDER for rendering and AI compute work.

According to CoinGecko, RENDER is currently trading in the $1.35–$1.40 range, falling from its all-time high of $13.53 in March 2024, and has a market cap of approximately $690 million.

This is a token that depends on the throughput of the GPUs, and is vulnerable to anything that squeezes the supply of those chips, such as a slower, more expensive route from the mine to the chip.

A long trade stand-off with China would do just that; quick domestic mineral processing would relieve it over time.

None of this is any justification to run on a chart. A reminder that “crypto” and “critical minerals” are now in the same supply chain, whether they realize it or not.

Mineral policy watching is slowly moving into the fold of the fundamental watching of AI-tokens.

This is not financial advice.

Do you track commodity or supply-chain news when researching AI tokens, or does that feel like a stretch?