海盗鸭

Robinhood 二季度预测市场收入 1.56 亿美元,加密交易收入 1 亿美元,同比掉 38%。公司历史上第一次,赌事件结果的人比炒币的人给它赚得多。事件合约成交 136 亿张,是去年同期的十倍以上。总营收 13.1 亿创纪录,财报出来后股价跌了 4%。

英伟达参投的一只AI股最近出现内部人士持续增持，反常点在于，市场还在争论AI估值是不是透支，最了解公司的那批人却在用真金白银投票。这个信号值得看，但不能只看“英伟达背书”四个字：内部买入要区分公开市场增持、股权激励和低价期权，含金量差很多。对 $NVDAB 来说，参股生态公司的意义也不只是财务收益，更像提前卡位软件、模型和算力需求入口。过去一轮AI行情先涨芯片，后扩散到应用和基础设施，但很多概念股最终没兑现收入。我的理解是，这类消息短线容易拉情绪，真正决定持续性的还是订单、现金流以及内部人士后面有没有继续买，单次动作不宜过度解读。

#AI股票 #英伟达

美国陆军给洛克希德马丁追加 538.6 亿美元的 PAC-3 MSE 拦截弹订单,合同总额到 586.2 亿、周期七年,史上最大的爱国者导弹合同。背景是库存见底:CSIS 估算美军手上的爱国者拦截弹不到 1000 枚。洛马计划 2030 年前把年产能从约 600 枚拉到 2000 枚。

AI 工作负载正在推动 CPU 需求，同时 AMD 又在讨论更长期的中国业务安排，这两个信息放在一起，反映的是机会和限制同时存在。$AMDB 可以受益于数据中心升级以及企业对算力的持续投入，但芯片行业不是只要需求增长就能顺利兑现，产品竞争、供货能力和出口限制都会影响实际收入。中国市场的长期合作如果推进顺利，可能扩大商业空间；但政策变化和合规边界也会让订单预期打折。投资者容易只看 AI 带来的增量，却忽略地缘因素对产品销售路径的影响。接下来要关注的，是 CPU 需求能否持续转化为订单，以及公司对中国业务的表述有没有变得更具体。

#AI芯片 #半导体

For this anti-quantum problem, the hard part isn’t cryptography—it’s coordination. BIP-110 just sets a lower bound on the solution.

First, look at the exposure surface. According to a Google Quantum AI white paper, in early 2026 there will be about 6.7 million BTC sitting in addresses whose public keys have already been exposed on-chain, representing about 34% of the circulating supply. Of these, around 2.3 million are both fragile and dormant—untouched for more than five years.

Next, see how the timeline changes. The same white paper estimates that breaking Bitcoin’s 256-bit elliptic curve would require fewer than 500,000 physical qubits with superconducting hardware. With the prerequisite computations completed, you’d get results in about nine minutes. The estimate in 2019 was 20 million qubits—shrinking by roughly 20 times over seven years. Today, no machine can do it yet; it’s still a few orders of magnitude away. But the rate of shrinkage itself is the information.

The technical path is already moving. In February this year, an IACR paper proposed splitting assets into two categories: public-key-exposed assets use classical/post-quantum hybrid authorization; unexposed assets use STARK-verifiable ownership relationships to bind old addresses to anti-quantum public keys, without disclosing the old elliptic curve public key throughout. There are also solutions that don’t change the main network, instead working at the contract layer.

The hard part is what comes next. The remaining pieces require consensus on three historical issues that have never been resolved: how large the witness data should be, how long the migration window should last, and what to do with coins that are never migrated.

And BIP-110’s goal is only to temporarily restrict non-currency data. Miner signaling rate is 2.53%, with a threshold of 55%. Even small changes can’t get traction.

So among those 2.3 million dormant coins, how many will ultimately never wait for their owners to migrate?

Gary Black recently gave Uber the biggest chance to popularize Robotaxi, not Tesla or Waymo. The unusual part of this judgment is that he’s not betting on the strongest players in autonomous driving technology; he’s betting on platforms that already have drivers, passengers, and dispatch networks. Uber can connect to multiple fleets at the same time and doesn’t have to single-handedly bear the costs of building cars, sensors, and operating across the entire city; however, $TSLAB and $GOOGLB , which correspond to Google’s parent company, hold core technology. Once scaling drives costs down, they may also bypass the platform and operate directly. My view is that Robotaxi competition early on is about technology, but later on it’s more like a battle over traffic gateways and fleet utilization rates. Uber’s advantage lies in its light-asset model and demand density; its weakness is that its bargaining power may be taken by the technology providers. In the end, whoever can control profit per kilometer wins.
#Robotaxi #自动驾驶 #US stock tech

Carvana 二季度卖出 197325 台车,同比增 38%,营收 73.8 亿美元,调整后 EBITDA 7.69 亿,全是纪录。然后它第一次给出全年 EBITDA 指引:27 亿到 30 亿美元。此前德银的预期是 30 至 32 亿,摩根士丹利是 44.5 亿。股价盘后跌了 6.6%。

Robinhood delivers a great earnings report, yet the stock price still falls. Second-quarter revenue of $1.31 billion hits a record high, up 32% year over year; earnings per share of $0.62 beat analysts’ expectations around $0.42. The stock dropped 3.15% on the day and continued to fall the next day. Crypto revenue fell nearly 40% year over year, which is the main reason. Needham, Deutsche Bank, Goldman Sachs, and Barclays all cut their price targets, although their ratings remain moderately bullish.

Clarity 法案里吵得最凶的伦理条款,7 月 22 日版本带两个限制:2029 年 1 月 20 日到期,且只管总统和配偶「发行」或「赞助」特定数字资产,不覆盖其他家庭成员。背景是特朗普今年申报,家族加密业务去年进账超 14 亿美元。

Three demand pipelines all went out at the same time—who ultimately ends up holding the support worth 64,000?

Let’s first break down the structure. Spot ETFs have seen net outflows for four straight trading days, totaling about $527 million; on July 24 alone, it was $240 million. Perpetual contract buy pressure is weakening. On-chain new capital has stalled. All three legs are soft at the same time.

What’s truly worth watching is the remaining leg. Long-term holders’ supply is set to climb to around 16.64 million BTC by the end of July, representing about 83% of circulating supply—an all-time high—then it starts to decline. There’s a slight twist here: LTH distribution from the peak has historically been treated both as a “bottoming” signal and as real, concrete selling pressure. The same action, two interpretations.

Fidelity’s Yardstick is on the optimistic side. It takes market cap divided by computing power to produce a Z-score—essentially giving Bitcoin a “P/E ratio” based on market cap relative to safety costs. Over the past 92 days, 83% of the time it has stayed in the undervalued zone. But they themselves also flagged a thorn: the realized market cap of long- and short-term holders is currently 3.9, while historically, a more certain bottom requires 4.0 or higher. The gap is 0.1.

What I care about more is that the nature of the ETF pipeline is being repriced. In a bull market, everyone treats it as structural buying. Now it’s becoming clearer that it’s a two-way door: the same channel turns into an amplifier when there’s a pullback. Custodians don’t make judgments—they merely execute subscriptions and redemptions.

So the real question is: does a channel that only keeps capital when prices are rising actually count as institutionalization?

美联储按兵不动当天,30 年期美债收益率盘中冲到 5.244%,2007 年 7 月以来最高;10 年期升 8 个基点到 4.68%,2 年期反倒跌到 4.24%。曲线是以熊陡的方式撕开的——不加息,但市场对通胀这条不买账。7 月 31 日 30 年期一度摸到 5.28%。

Robinhood Q2 财报:事件合约收入 1.56 亿美元,同比涨超 10 倍,第一次把加密交易收入甩在身后——后者 1 亿。总营收 13.1 亿,净利 5.73 亿,都是纪录。管理层在电话会上主动把监管列为事件合约的风险项。财报当天股价跌 3.3%。

On July 29, the Federal Reserve held steady and kept interest rates at 3.5%–3.75%. In the crypto world, it was basically treated as “the bad news is finally out.” But the real information from this meeting isn’t in the outcome—it’s in the voting pattern.

9 to 3. Three dissenting votes—Hammack of Cleveland, Kashkari of Minneapolis, and Logan of Dallas—each called for a 25-basis-point rate hike. The direction of dissent was completely aligned among all three, the first time this has happened since September 2016.

More importantly, there have been structural changes since Chair Warsh took office: the forward guidance has been removed. Previously, the FOMC would tell you what its reaction function looked like, and markets would price future meetings accordingly, smoothing out volatility. Now, that line has been cut.

The consequence is that the interest-rate path shifts from “a guided curve” to “opening blind boxes one meeting at a time.” Uncertainty before the meeting is no longer worked into expectations in advance; instead, it all piles up on the day of the meeting. It’s essentially adding an extra volatility risk premium to all duration-sensitive assets.

Bitcoin is the longest-duration asset in this room—no cash flows, no maturity date, and its valuation depends almost entirely on discount rates and risk appetite. With guidance gone, it’s the most affected.

The backdrop also doesn’t stand with the bulls: core PCE rose from 3.0% in December last year to 3.4% in May. The market now expects one to two more rate hikes before year-end—not rate cuts.

“Not hiking” doesn’t mean “bullish.” It only pushes the answer to the next meeting.

Are you still using the same positioning framework as the rate-cut cycle?

Robinhood Q2 总收入 13.1 亿美元创纪录,加密交易收入只有 1 亿,同比降 38%,是七个季度里最差的一季,比市场预期还少 2500 万。App 内加密成交量掉 35% 到 180 亿美元。财报出来后股价盘后跌约 4%。公司另外提了一句:7 月加密日均成交比 Q2 还要慢。

It’s also about throwing money at AI. Microsoft is up 7%, while Meta is down 10%. The difference isn’t spending—it’s accounting and storytelling.

First, how Microsoft’s 7% came about. It shifted future data center lease commitments from finance leases to operating leases, meaning they no longer count as capital expenditures. At the same time, it extended depreciation life from 15 years to 25 years. As a result, Microsoft’s calendar-year 2026 capital expenditure guidance dropped from about $190 billion to about $175 billion. But CFO Amy Hood was very clear: the total amount of actual spending hasn’t changed.

What changes is the accounting. The same pot of money, categorized differently, and the market response differs by an entire order of magnitude.

Meta, on the other hand, doesn’t have that buffer. In Q2, revenue was $60.8 billion, up 28% year over year and a record. But free cash flow plunged by about 91% to $784 million; it was $8.5 billion in the prior-year period. Single-quarter capital expenditures exceeded $30 billion, up 83% year over year, and the company also issued roughly $25 billion of debt to keep this construction going.

What the market is truly punishing is “can’t be booked.” Microsoft can point to Azure growth of 43% and annualized revenue topping $100 billion. Meta’s AI value is embedded in ad conversion rates, with no standalone revenue line it can price.

This quarter, the combined free cash flow of the four mega-scale vendors fell to about $7 billion—its lowest level in a decade. For people in the industry, this isn’t noise. The liquidity layer that underwrites risk assets is being drained to build server rooms.

In the same week, the Federal Reserve kept rates unchanged at 3.5%–3.75%, 9 out of 3 votes—while those three dissenting votes wanted rate hikes.

Money is getting more expensive, and it’s all going to server rooms. Where will the marginal buyer of risk assets come from?

预测市场在降温。Kalshi 加 Polymarket 合计周成交环比跌 19.8% 到 97.4 亿美元,6 月底以来第一次掉破 100 亿。4 月的月度峰值是 240 亿左右。另一个数:Kalshi 今年累计成交 1480 亿,占它历史总量 85%——这条赛道基本全部发生在 2026 年。

On July 29, on the same day, Meta and Microsoft both released their earnings reports. Capital expenditures hit record highs for both companies—one stock fell nearly 10%, while the other rose by about 7%. The difference isn’t how much money they spent, but which line item on the statements that money ends up in.

Meta: Revenue was $60.8 billion, up 28% year over year and beating expectations. But quarterly capital expenditures were $31.1 billion, and operating cash flow was $31.9 billion—almost all of it got eaten up. Free cash flow came in at just $784 million; a year earlier it was $8.5 billion.

Microsoft: Revenue was $90 billion, up 18%; quarterly capital expenditures were $41 billion; free cash flow was $19.6 billion, down about 23% year over year. Azure’s full-year revenue broke $100 billion.

What you really should look at is Microsoft’s remark about accounting changes. CFO Amy Hood said that starting in FY27, the estimated useful lives for data centers and office buildings will be extended from 15 years to 25 years. She spelled out the implication: this is mainly not a profit issue (it has very little impact on operating profit in FY27), but a classification issue—more future data center leases will shift from finance leases to operating leases. Finance leases are counted as capital expenditures, while operating leases are not.

The same server rooms, the same power—once the accounting treatment changes, the capital expenditures line suddenly looks good.

But the market’s judgment on the day was more straightforward: can you name which revenue line this money corresponds to? Microsoft can point to Azure. Meta can only say that ad efficiency has improved, along with a hint that it wants to build a cloud.

Depreciation life and lease classification are the two most underestimated variables in this round of AI capital expenditures. When you read the reports, will you actually turn to these two lines?

7 月 FOMC 按兵不动,联邦基金利率维持 3.50%-3.75%,但 Logan、Hammack、Kashkari 三位地方联储主席投票主张加息 25 个基点。9 比 3,是 2016 年以来分歧最大的一次决议。Warsh 在发布会上继续拒绝给前瞻指引。

The problem with XRP isn’t that the long side is too crowded—it’s that nobody is taking the other side

The common saying about XRP is “leveraged longs are overcrowded, whales are dumping.” The data basically doesn’t support this version.

Funding rates have stayed near zero, long and short positions are roughly balanced, and CryptoQuant’s reading is a slow, cautious repositioning—not a leveraged frenzy. And on the whales’ side, it’s even more counterintuitive: the amount of large transfers into exchanges has fallen from the cycle peak of 583 million XRP (about $1.36 billion) to 25.3 million XRP (about $23 million). The 30-day inflow has dropped to a two-month low.

Selling pressure is decreasing. So why is the price still lying there?

Look at these ratios: daily futures volume is about $1.8 billion, spot is about $180 million—close to 10:1.

The market is priced almost entirely by derivatives; the spot side is empty. Less sell pressure and new buying pressure are two different things—the former can only slow down the drop, not lift the price. A futures-driven rebound has no spot “foundation”; if it rises, it’s hollow. This is exactly why XRP has been kept down all year.

By the way, let’s correct a number that’s been quoted everywhere: many places write “down -67% from the all-time high,” but what they’re really referring to is the drop over the past year. Using the $3.65 historical high recorded by CoinGecko, the current level of about $1.02 corresponds to roughly -72%.

Holders’ average cost basis is around $0.75—most wallets aren’t losing money yet. There’s no motivation to panic-sell and cut. But that also means there’s been no chance to flush out a low.

This is a contraction in volume, not a breakdown. The solutions are completely different: a breakdown waits for sell pressure to run out; a low-volume move waits for demand to show up. The latter doesn’t come with a timetable.

美国陆军把洛克希德的爱国者 PAC-3 MSE 合同加到 586.2 亿美元,7 年期,这款拦截弹史上最大单。做法是在 4 月那份 47 亿的一年期合同上追加 538.6 亿。合同尚未定型,金额和数量还会变。洛马说产能要从年 600 枚提到 2000 枚。