Coincamps

Translation: Peggy
Editor's Note: On August 19, the U.S. Treasury announced an expansion of long-term Treasury liquidity support repurchase agreements, increasing the single round of repurchase for 10-20 year and 20-30 year nominal coupon Treasuries from a maximum of $20 billion to at least $40 billion. The new arrangement will be effective from September 9. Before the announcement, the 30-year Treasury yield briefly rose to around 5.34%, hitting a post-2007 high; after the announcement, the long-end yield retraced momentarily.
$40 billion is not significant compared to the over $30 trillion U.S. Treasury market, and the repurchase itself does not equate to quantitative easing. What really sparked market discussion is the timing of the announcement: the Treasury had just concluded the quarterly refunding announcement two weeks ago but suddenly increased the size of long-term bond repurchase outside the regular window. This led investors to reassess how willing the Treasury is to actively intervene in the market as long-term yields spike.
The Heisenberg Report cited Nomura Securities' Cross-Asset Strategist Charlie McElligott and Rabobank's Strategist Michael Every's assessment, interpreting this move as a policy signal: the U.S. government may not be willing to let long-term funding costs keep rising, thereby constraining fiscal spending, geopolitical strategy, and private sector financing. This led the market to create the 'Bessent Put,' referring to the 'Bessent Floor Expectation.'
However, there is still a long way to go from expanding repurchases to yield curve control or even restarting quantitative easing. This article is not really discussing whether 'QE is back,' but whether the U.S. policy reaction function is changing: if fiscal pressures, inflation, AI financing, and geopolitical conflicts continue to drive up long-term rates, will the Treasury and the Fed be forced to take stronger actions?
Below is the translation of the original article:
After the U.S. Treasury expanded long-term Treasury repurchases, the market's initial questions were not about the scale but two more direct questions: why now? Does this imply that the U.S. government is starting to set an implicit floor for long-term yields?
Some investors have already dubbed this arrangement the 'Bessent Put,' or the 'Bessent Floor Expectation'; others have called it a 'lite QE' or a new round of 'Twist Operation.' These names are not formal policy concepts but market speculations on the Treasury's policy intentions.
On August 19, the U.S. Department of the Treasury announced that it would increase the liquidity support repurchase size for 10–20-year and 20–30-year Treasury Inflation-Protected Securities (TIPS) from a maximum of $20 billion to at least $40 billion. The official reason given by the Treasury was that the long-term bond repurchase continued to receive a large number of high-quality bids, and therefore, they aimed to provide stronger liquidity support for these tenors.
This explanation did not completely dispel market doubts. A single $40 billion repurchase remains limited, but just before the announcement, long-dated U.S. Treasuries had just experienced a rapid sell-off, with the 30-year yield briefly spiking to around 5.34%. Therefore, investors were more concerned not with how much the Treasury actually bought but with what signal it chose to send at this point in time.
$40 Billion Is Not Large, Unexpected Announcement Itself Is More Important
Nomura Securities' cross-asset strategist Charlie McElligott believes that the specific size of the repurchase is not the key issue. More importantly, Powell seems to be telling the market that the U.S. government cannot accept continued disarray in the long-term Treasury market, and fiscal and monetary authorities may take a more proactive stance than before.
This is an analyst's interpretation of policy intent, not a confirmed yield level target by the Treasury. Officially, the Treasury still defines this adjustment as "liquidity support" and has not announced any yield level they are aiming to support.
However, the timing of the announcement reinforces market speculation. The U.S. Treasury usually communicates funding and debt management arrangements through the Quarterly Refunding Announcement (QRA). This adjustment came just about two weeks after the last QRA, outside of the regular communication window.
According to McElligott, this unconventional timing indicates that the speed of the rise in pressure on the long end of the bond market may have exceeded the policy sector's previous expectations. Consequently, the market interpreted the announcement as a "signaling operation": the Treasury aims to prevent further deterioration of liquidity from amplifying the rise in long-term rates, rather than just routine optimization of the bond structure.
This assessment still needs to be cautious. The subsequent decline in yields after the announcement only indicates that the market reacted immediately to the news, not that the Treasury has successfully lowered long-term funding costs. In fact, the subsequent pressure on long-dated yields also indicates that small-scale repurchases may not be enough to offset deeper factors such as fiscal deficits, inflation, and bond supply.
Long Bond Pressure Does Not Originate From a Single Variable
The article believes that the repurchase behind this is not a single liquidity issue, but that multiple adverse factors are simultaneously squeezing long bond demand.
First is the continued expansion of the US fiscal deficit and debt supply. When investors hold long-term bonds, they usually require an additional return to compensate for inflation, fiscal, and interest rate volatility risks. This part of the return is known as the term premium. The original text's referenced chart shows that the estimated 10-year US Treasury term premium has approached nearly 80 basis points, about twice the peak of the 2023 sell-off in long-end bonds.
Secondly, AI infrastructure development is bringing a large amount of corporate bond financing. Tech companies and data center operators need to raise funds for chip, power, and computing facilities. The increase in corporate credit bond supply will compete with US Treasuries for private sector balance sheets. McElligott summarizes this as a "crowding-out effect": when both government and corporate bonds are issued in large quantities, there is a limit to the long duration risk the market can absorb.
Japanese factors have also added to the uncertainty. Japan is a significant overseas holder of US Treasuries. The depreciation of the yen and its potential intervention needs make the market concerned that Japanese institutions may sell off some US Treasuries to raise dollars. The article links the recent US engagement in the foreign exchange market with the Treasury's expansion of long bond buybacks, suggesting that policymakers may want to avoid reinforcing exchange rate intervention and US Treasury sell-offs.
However, this is still a market interpretation. Public information can confirm that the US Treasury has expanded long-term bond buybacks, and pressure on long bonds, the yen, and corporate financing can be observed. Still, the Treasury has not provided a full explanation of whether these factors directly constitute the reason for this policy adjustment.
「Bessent Put」 Points to a New Policy Reaction Function
What the market is truly repricing is the US government's policy reaction function.
The so-called policy reaction function refers to investors' judgment of what actions policymakers may take under what conditions based on their past behavior. If the market believes that after long-term rates rise to a certain level, the Treasury will increase buybacks, adjust issuance maturities, or enhance coordination with the Fed, investors may begin to factor in this potential intervention into bond prices in advance.
The "Bessent Put" is precisely the market expression of this expectation. It is not an official policy, nor is it a Treasury commitment to support US bond prices. It refers to investors starting to speculate: when long-term yields threaten government financing, economic activity, or other policy objectives, Bessent might take more proactive debt management measures.
Michael Every further explains from a geopolitical strategic perspective that what the US government may focus on is not just "lowering yields" but avoiding long-term funding costs limiting its foreign policy, especially against the backdrop of ongoing tensions with Iran and rising energy supply risks.
Every believes that in the past, the United States could conduct external actions by controlling financing conditions and key supply chain support. However, the current situation is more complex. The United States does not fully control the energy and related physical supply chain, and even though some crude oil can still be transported through the Strait of Hormuz, finished oil supply may not be able to recover concurrently.
McElligott also raised similar risks: if the Gulf situation escalates again, the impact could spread globally through finished oil, manufacturing, and inflation. Crude oil inventories can be released, but refining capacity and finished oil supply cannot be quickly replenished simply by releasing inventories.
This means that policymakers may face two opposite pressures at the same time: geopolitical conflicts pushing up energy prices and inflation, requiring interest rates to remain relatively high; fiscal financing and economic pressure, yet demanding long-term rates not rise indefinitely. Expanding repurchases may alleviate market liquidity, but it cannot eliminate this policy contradiction.
Repurchase Is Not QE, Further Impact Is Needed for Yield Curve Control
Does expanding government bond repurchases mean that the United States has returned to the path of quantitative easing? The original text suggests that this may open up such a discussion, but it is still too early to draw conclusions.
The Treasury's repurchase is fundamentally different from the Fed's quantitative easing. Treasury repurchases are mainly debt management operations, buying back old securities with poor liquidity, coordinating with the issuance of other maturity bonds to improve market operations or adjust debt structure; while QE involves the Federal Reserve's large-scale purchase of assets and injecting reserves into the banking system, directly expanding the central bank's balance sheet.
Therefore, liquidity repurchases at the $40 billion level cannot directly be called QE, nor are they sufficient to prove that the Treasury is implementing formal yield curve suppression.
McElligott believes that this announcement is more like an "intention statement," prompting further market discussion on the possibility of YCC or QE. YCC refers to yield curve control, where the central bank commits to buying bonds to limit specific maturity yields near the target level; LSAP refers to large-scale asset purchases, which is also a primary form of quantitative easing implementation.
However, he also emphasizes that before these tools become the next policy choice, the market and economic environment must "deteriorate much further." In other words, the "Powell Put" currently changes investors' imagination about the policy boundary, rather than indicating that the United States has launched a new round of QE.
What needs to be observed next is not only whether the Treasury continues to expand the size of single repurchases, but also whether long-term yields can stabilize, term premiums fall back, the Treasury further shortens debt issuance duration, and whether the Fed will adjust its balance sheet policy accordingly.
If these measures continue to escalate, the market's perception of "Treasury backstop" and policy coordination will be strengthened; if long-term rates continue to rise under structural pressure, and the Treasury still limits repurchases to small-scale liquidity operations, then this announcement is more likely just an attempt to stabilize the market in the short term rather than the starting point for QE.
[Original Article]

On August 17, Xiami is back.
Alibaba has released the AI music model HappyShrimp 1.0, Chinese name "快乐虾米" (Happy Xiami). Users only need to write a sentence, and it can generate a complete song.
You don't even need to know what chords, time signature, or key it is in. You can tell it you want to write a song for your recent graduation, best suited for playing in a coffee shop; or you can say, it's almost time to get off work, suddenly raining outside the window, feeling a bit melancholic, but not too heavy. Words that used to describe feelings can now be turned into music commands. HappyShrimp will try to understand and then create accordingly.
Currently, it has launched a desktop web version for testing. Alibaba stated that HappyShrimp adopts end-to-end whole-track generation, incorporating style, emotion, era, and vocals into the same generative process. In actual tests, it is particularly sensitive to Chinese daily expressions. Many times, users are not clear about what kind of music they want, only knowing "lighter," "like a summer ten years ago," "not too bitter," and it can follow these vague descriptions to proceed.
This makes music production very cost-effective. Adding background music to a short video, creating a sample for an advertisement, finding background music for games or podcasts, no longer requires starting a whole production process from scratch. The standard membership price on the test page is about 30 yuan per month, which can generate 150 songs. On average, a song costs about 0.2 yuan.
On the day of its launch, HappyShrimp also announced a collaboration with Taihe Music. Alibaba also plans to have it appear at this year's Anaya Xiami Music Festival. It is also the third member of Alibaba's "Happy + Animal" series, with HappyHorse for video generation, HappyOyster for world models, and now, the third one is music.
Xiami Music took twelve years to become the later nostalgic Xiami.
HappyShrimp can generate a song in about a minute.
The Era of Handcrafting
The original Xiami started very small.
Wang Hao, originally an Alibaba programmer, also played in a band, using the online alias "Pumpkin," taken from his favorite band, Smashing Pumpkins. During college, he played the guitar and was in a band called "Black Water." During his sophomore year summer break, he was bitten by a dog on his finger in the rehearsal room, unable to play the guitar temporarily. With nothing to do, he started learning to code.
It started with HTML, then moved on to PHP. He even translated the PHP manual and built websites for others to earn some pocket money, charging 200 RMB for a static page. Later on, he created his own forum called "SoundNet," dedicated to sharing underground music and event information in Hangzhou. By 2001, the forum had tens of thousands of registered users, with hundreds online every day. A group of young musicians and rock music enthusiasts used the platform to organize gigs, find bandmates, and chat. When bands from out of town came to Hangzhou, Wang Hao would sometimes help them secure venues.
He eventually realized that he might not have the talent to become a professional musician, but he excelled at bringing music enthusiasts together.
After graduating from university, he shared an apartment with Zhu Qi, who would later co-found Xiami. During the week, they focused on their respective jobs, but on the weekends, Hangzhou's musicians and friends would gather at their place to eat, drink, and talk about music. Sometimes, the apartment would be filled with people. Wang Hao made a living selling musical instruments online during the day and helped bands find gig venues at night and on weekends.
In 2003, he believed that e-commerce could be a viable business, so he went to Alibaba to learn about the internet. He started with Java and later transitioned to requirements analysis. After a few years, he started thinking about music again.
Back then, people would use eMule, Kazaa, and BitTorrent to download movies. Wang Hao wondered if a P2P system could also work for music, but with a twist: organizing the messy files and ensuring both uploaders and music creators could earn some money.
In 2007, he left Alibaba. Initially, there were six people in total, with Wang Hao as the general manager, Zhu Qi in charge of content, and three programmers handling operations. Everyone contributed roughly the same amount of capital and held similar shares, paying themselves a monthly salary of 3,000 RMB.
They named their website EMUMO, short for Earn Music & Money. The name clearly reflected their initial ambition: listeners would get music, and music creators would get paid.
It was later rebranded as Xiami. Their explanation was, "We may be small shrimps, but we have big dreams."
In November 2008, Xiami.com officially launched.
The first office also resembled a "small shrimp" company. Located in the Longdoo Building in Gudang, Hangzhou, it was a 70-square-meter mixed residential and commercial space. They didn't do much renovation after the previous tenant left, just moved in a few tables and chairs to start working. With no place to store the servers, they set them up on the balcony. As soon as they entered the office each day, the constant whirring of the fan could be heard. Looking back, Wang Hao recalled one advantage of the office—the view of the Xiaoheshan Mountain from the window, surrounded by lush greenery.
That was 2008. The iPhone 3G had just been released, and smartphones had not yet consolidated all music into one app. In China, when people listened to music, they often searched for song titles on Baidu, navigated between forums, blogs, and resource sites, found an MP3, downloaded it to their computers, and then transferred it to an MP3 player. What they found was purely a matter of luck.
The file names were often garbled, with "320K" written in the name, but upon opening, the quality was blurry; behind one song title were over a dozen download links, and it was unclear which one was genuine. Album information was incomplete and riddled with typos, when multiple singers collaborated, they were simply labeled as "Various Artists," the track order was frequently mixed up, and few people cared about which album or year a song belonged to.
At that time, online music was already one of the most popular internet services among Chinese netizens. According to a 2008 CNNIC survey, 86.6% of internet users had listened to online music in the past six months, and 71.2% had downloaded music. Hundreds of millions of songs were circulating on the internet, most of the time, existing as just an MP3 file.
Xiami, however, aimed to transform these files back into "albums."
Albums had to be arranged in their original track order; for songs with multiple singers, each name had to be spelled out in full rather than simply listed under "Various Artists"; the audio quality aimed for 320K; different versions of the same song had to be separated; and details such as singers, albums, years, record labels, and genres, were all to be filled in if possible.
From its inception, editors made up the largest group of employees at Xiami. Their daily work involved researching, verifying information, and reconstructing complete albums. Even when the website only had 100,000 users, Xiami already employed six editors covering various languages, including Chinese, English, Japanese, Korean, Russian, Thai, and Spanish. Users could even search directly using these languages.
When the company's own staff was insufficient, they turned to users for help. Xiami recruited over 300 veteran music enthusiasts from around the world to collectively enrich their music library. Over time, a group of users specializing in data correction emerged, and they referred to themselves as the "Wiki Team."
This name was not an exaggeration at all. They would debate whether an album should be classified as a Studio Album, EP, or Live Album; what naming structure should be used for classical music tracks; whether composers' full names or aliases should be written; and how European and American singles should be cataloged. Some even dedicated themselves to writing rules, reminding editors to consult multiple classical music websites in the future and to "not disappoint the expectations of so many classical music enthusiasts."
For some information that couldn't be found online, they would resort to digging through CDs. Opening the album case, taking out the inner booklet. Who was the producer, who played bass, who arranged the strings, where was the recording studio—line by line, word by word, all entered into Xiami. Later, when someone asked where to find complete production information for a Chinese song apart from buying the original CD, Xiami staff mentioned that oftentimes even the digital materials provided by record companies were lacking, and they had to rely on users to extract data from the inner pages of CDs.
A user named Desperado once single-handedly contributed data for 943 artists, 214 album entries, and also led the Xiami Lyrics Group.
This group of people even established a complete set of informal academic norms for "music listening." Xiami's early community resembled a music BBS. Some curated playlists, some wrote in-depth music reviews, some corrected mistakes, some researched how to compress higher quality 320K MP3s. In the early days, music reviews were required to be at least 300 words. You couldn't just leave a comment like "it sounds good" and leave. Since you specifically clicked to discuss a song, you had to say something meaningful.
When users clicked into a genre, they could continue drilling down further. What was under Dream Pop, where did Shoegaze come from, what was the difference between Post-Rock and Progressive Rock. For many, Xiami was more like an ever-expanding map of music. They would journey from one record to another, from one band to someone they had never heard of, and eventually forget what they were originally looking for.
This classification system grew larger over time. The genre system left behind by Xiami was eventually organized by users into a 436-page document, with 24 primary categories and 548 subcategories. Many of these categories may only be explored by a few people in their lifetime, but Xiami still provided a space for them.
There was also a section called "Desert Island Discs."
This section posed an interesting question: "If one day you were stranded on a desert island and could only bring a few albums, what would you bring?"
Through this question, the editors recommended albums that were not so popular but deserved to be heard. Eventually, the section had its own slogan: "Don't let good music wander." In 2020, they even collaborated with Hema to open a "Desert Island Discs Bakery."
This name actually suits Xiami itself very well. A vast sea, with a small island in the middle. It's noisy outside, but on the island, there are some songs that no one is in a hurry to listen to. Those willing to swim over, just sit for a while.
Xiami's office later became more and more like such a place. Several former employees recalled that there was a small stage in the lounge area, with guitars, pianos, and keyboards nearby. During dinner, someone would get up to play the instruments and sing; someone who had just joined the company recently would pull colleagues together to form a band.
A group of people running a music website, to some extent, made personal use of public property. They brought their hobbies into the office and then found a very legitimate reason to continue listening to music every day.
Even the initial business model carried this idealism. Wang Hao wanted to create a closed P2P system. Users would download a song for 0.8 yuan. According to the early design, a portion would go to the copyright holder, a part to those who uploaded and distributed the music, and the platform would only take the remaining portion. After users downloaded the music, their computers would become nodes in the network, continuing to share the song with others.
Xiami even gave its own cryptocurrency a handy unit, the "Mi".
By uploading original albums and helping to spread music, users have the opportunity to earn Mi. The listeners are also distributors of the music. Wang Hao wanted to shorten the long chain in the record industry, allowing music to more directly reach the audience from the creators.
Looking back, this design seems a bit naive. But in the first few years, Xiami grew little by little following this vision.
Trouble was also embedded right from the start. P2P helped Xiami rapidly accumulate a large music library, but at the same time, it left behind a looming copyright issue. In 2010, dozens of musicians such as Li Zhi, Zhou Yunpeng, Wan Xiaoli, and Zuo Xiaozuzhou publicly protested against their works appearing on Xiami without authorization.
Occasionally
The old Xiami certainly used algorithms too. By 2011, Xiami's own recommendation system had taken shape. Employees later recalled that Wang Hao instructed the backend to create over thirty music tags for each user, while at the same time, Douban FM only recorded about four to five.
However, Xiami did not let the algorithm simply cater to the users. It would first give you mostly familiar things, then intentionally mix in a bit of the unfamiliar. In the early product concept, this ratio was roughly nine to one, with ninety percent closely following the user's established taste, leaving the remaining ten percent for those who hadn't listened to music before.
That 10% later became something many old users missed the most.
There was a user named "Qing" who usually listened to Japanese visual kei and hardcore music. In the daily 30 songs playlist Xiami curated for her, one day a very warm English song, "The High Road," was suddenly added. It didn't seem like something she would like at all. However, after she listened to it, she had it on repeat for three days.
Another user, Ah Cu, was briefly obsessed with black metal. Xiami sneaked in a German band, Empyrium, under "Similar Artists." When she clicked on it, it turned out to be folk music. A week later, she became a fan of Empyrium. Digging deeper, she discovered that this band had indeed done black metal in their early days, then transitioned to unplugged folk later on. The musical form changed, but the reclusive, hermetic quality remained constant. Ah Cu later recalled that her aesthetic sensibilities were influenced by this recommendation for the next year or two.
This was perhaps the old Xiami's understanding of algorithms. It tries to predict what you like while occasionally allowing itself to be a little unconventional.
On Xiami, there is a group of users who like to call themselves "Diggers," and they will click through an album's producer, label, and genre all the way down.
Xiami Music user "Throw Throw" stumbled upon an Italian musician's work in 2014, found the sound to be strange yet fresh. Following the trail, he learned for the first time that this thing had a name, called Vaporwave.
He continued listening, and eventually started creating his own. "Throw Throw" applied to become a Xiami Musician and uploaded his work. Several songs were discovered by editors, appearing on the Xiami homepage multiple times.
Such stories are not uncommon on Xiami. The Xiami editorial team refers to their work as "Music Pilots." Editors listen to a large number of new songs every day, create thematic recommendations, write music reviews and recommendations, and select those sounds that have not yet entered the public eye. In 2020, they were still producing programs such as "Deserted Island Records" and "Global Music Map."
The editor in charge of "Deserted Island Records," Feng Xia, used to drive two hours to the city center Livehouse to attend performances while in university. After joining Xiami, her job became listening to music every day, trying to find ways to pass on what she discovered to others.
In July 2013, the "Xiami Musician" platform was launched. Independent musicians and labels could upload their work, set their own prices for legitimate downloads, and receive 100% of the download income. In the first year, over five thousand musicians joined.
In the second year, Xiami launched the "Seeking Light Plan." Initially, the plan aimed to select musicians from the platform and help them further develop their demos. If they lacked a producer, they would help connect them; if they didn't have a recording studio, Xiami would fund it; from master tapes to album releases, music videos, and tours, Xiami helped with all preparations. In the earliest public plans, they even aimed to produce fifteen albums or EPs in a year, a volume that could match that of a professional record company.
The first season left behind 13 groups of musicians, 14 albums, accumulating about 160 million listens.
Cheng Bi was one of them. She later described Xiami as her "patron." The first three albums were all released on Xiami. At the beginning of the first Seeking Light Plan, she also became a Seeking Light musician. Later, people had the opportunity to get to know Cheng Bi from "Poetry Meets Song" and "I Want to Spend Time with You in Fantasy."
Such stories became more and more common. Eventually, Xiami attracted over 40,000 original musicians. The music library contains 30 million songs, over a thousand genres, and users have created over 500 million playlists. In industry data from February 2018, Xiami had 9.685 million daily active users. The scale was already considerable, but compared to the largest music platforms, it still resembled that small initial shrimp.
The Last Night
Then the times changed.
The copyright war escalated, with Tencent Music and Netease Cloud Music spending big money to secure exclusives, causing batches of Xiami's music library to turn gray. Music became the ammunition of a traffic war. In 2015, the "Iron Triangle" composed of Gao Xiaosong, Song Ke, and He Jiong took over Ali Music, transforming Tingting into Ali Planet the next year to venture into fan economy, transferring personnel from Xiami, and slowing down product updates.
In January 2016, Wang Hao left Xiami, which he had founded, and moved to Ali Dingtalk. He said the music industry had become outrageously absurd.
When he made that statement, Xiami was still operational. Five years later, it shut down. By 2020, Xiami had around one million daily active users. From 9.68 million to 1 million in just a few years.
On January 5, 2021, Xiami announced that it would shut down one month later, giving users time to export their playlists, which could be saved as static web pages or Excel files.
Two days after the announcement, Wang Hao was interviewed by PingWest and said:
"I don't think there's anything worth reminiscing about."
On February 4, 2021, Xiami had its final day.
On this day, the personalized daily recommendation feature lost its personal touch for the first time. When users opened Xiami, they all received the same playlist. Over the past twelve years, this section curated songs for individuals every day; on the last day, the editorial team selected thirty songs for themselves.
"Long Time No See," "You Have Always Been There," "I Don't Want to Leave You Alone," "When Love Has Become the Past," "I Finally Lost You," "Goodbye." The song titles were already blunt enough, and above each song was a short recommendation. These thirty phrases were connected in sequence and formed the last letter Xiami wrote to its users. It even included a very Xiami-like joke: "Although we still didn't manage to get Jay Chou's copyright (laugh)."
The real farewell began at 8 p.m.
At that time, the Xiami app had a feature called "Live Room," which could be understood as a music live broadcast room where a host played songs, and others came in to listen, request songs, or chat on the screen. Several Xiami user groups connected in advance and agreed to host a live event that night, naming the room "The Last Night of Xiami Planet." The event description read: "We may not be able to change fate, but we can accompany you in the countdown until the very last second."
Thousands of people crowded in.
Some requested songs, some left messages, some did nothing but kept the app open. People from Xiami's editorial team also joined. Editor Agu bid farewell one by one in the live room, thanking users and then thanking the editors who had worked at Xiami. He said that over the years, there had been dozens of editors at Xiami. Many users, for the first time, were in the same room with the people who had created those playlists.
What everyone was waiting for was midnight on February 5.
The official announcement had made it clear that from this moment on, song previews, downloads, comments, and profile exports would all cease. Some took the day off in advance, some left work and closed the door behind them, listening alone all night; and some refused to exit the app, thinking that as long as the page was still open, the music might continue playing a little longer.
Midnight arrived, but Xiami did not fall silent instantly like pulling the plug.
Some people's songs were still playing. So, instead of the expected farewell, excitement rose among users, and messages kept popping up on the screen:
“Do you still have sound?”
“I can still hear it.”
Some tried disconnecting from the internet, while others cautioned everyone not to close the app. Everyone was doing something a bit absurd, scattered across different cities, different phones, confirming whether a piece of software still showed signs of life. As long as one person could hear sound through their headphones, Xiami seemed as though it hadn't completely died.
Eventually, it did stop.
Songs that some were listening to suddenly cut off, and others' export progress halted midway. An hour later, there was no music playing, yet the screen kept flashing “someone has joined.”
In the month leading up to the shutdown, users had actually been moving out.
Xiami had enabled profile exports, while QQ Music and NetEase Cloud Music had provided song playlist migration. Some spent six hours rescuing over six thousand songs they had collected; some took screenshots of their personal page to save the exact number of minutes and songs they had listened to; and some meticulously sorted Xiami's almost paranoid-level complex music genres into an Excel sheet, giving this document a very solemn name: “Saving Xiami's Legacy.”
Some took more direct action. After the shutdown announcement, several users rushed to the Alibaba office in Wangjing, Beijing, unfurling two banners:
“Xiami, please don't go.”
“Xiami, please don't close.”
Xiami's founding team member and former Product Director Zhangstraw (稻草) only posted one sentence on social media: “The best farewell now is to say nothing at all.” Founder Wang Hao posted five Weibo posts that day, not one of which mentioned Xiami. He had left Alibaba several years earlier, long completing his own goodbye.
But the users hadn't.
Two years later, in February 2023, someone even wrote a “Xiami Music two-year Anniversary Memorial.” An app that no longer existed was still commemorated by people in a human way, anniversaries, two-year milestones, death anniversaries, eulogies.
The shrimp that people later fondly remembered had actually died many times before it was finally shut down. Batch after batch of copyright takedowns, the founder leaving, management changes, the entrance being moved, and the user base shrinking. The shrimp that was still alive in the end only had one million people opening it every day, which was no longer a significant player in the copyright war.
But when it did die for real, those one million people suddenly had a name.
They called themselves "music refugees."
After the shrimp died, the two words "Xiami Music" did not disappear. Seven months after the player was shut down, Damai established "Xiami Music Entertainment." The previous slogan "Hear a Different Voice" was changed to "See a Different World"; the shrimp that used to be in headphones began appearing in theaters, on stages, and at music festivals.
In 2022, the first Anaya Xiami Music Festival was held on the beach in Qinhuangdao, with the theme "Keep Swimming Until the Sea Turns Blue."
The crowd gathered again under the name "Xiami," but this time, they were not just listening to music through a player but were actually standing by the sea, watching the band on stage, watching the sun set. By the fourth annual music festival in 2025, six international musicians had made their mainland China debut here.
So, Xiami had a strange "afterlife." The product died, but the name remained. The original team disbanded, but the new business continued to use it. The small shrimp that used to be hidden in the player icon was plucked out from one place by Alibaba and put into another.
By 2026, this name appeared for the third time. This time, there was no player or stage; it was placed on an AI music model.
HappyShrimp, a joyful Xiami.
Ten years ago, the person who founded Xiami left Alibaba and even went to Phuket; ten years later, Alibaba picked up the name he left behind. However, this time, a new group of people were running Xiami, and even the concept of "music" had changed.
New Body
HappyShrimp is still very young. In version 1.0, the vocals occasionally had a robotic sound, singing complex Chinese lyrics in a muddled way, and the paragraph controls were not as precise as professional production software. But these are probably just a matter of time. What today requires drawing cards repeatedly more than ten times to get, the next version may be achieved by changing a few words.
What is truly noteworthy is that the act of writing a song is suddenly becoming too easy.
Suno can now generate over 7 million songs per day. In Deezer's peak in June this year, the platform received close to 90,000 fully AI-generated songs per day, surpassing half of all new music added to the platform for the first time.
The music industry used to rarely worry about "too many songs."
When the old Xiami was born, the internet certainly wasn't lacking in MP3s, but finding a song you really wanted to listen to was still not easy. It might be hidden in an obscure album, under a band no one knew, or you might not even know the genre. So Xiami spent a lot of effort doing one thing: searching.
Back then, music platforms were afraid of good content sinking, but the world of Joyful Xiami was just the opposite.
If you wanted a song to listen to on a rainy day, it could be made on the spot; if you wanted a female voice, a slower tempo, reminiscent of a summer ten years ago, that could also be arranged. Short videos lacking background music, ads needing a sample, games missing a soundtrack—previously, you had to search in a music library or find someone to compose, but now you can generate a song instantly.
So, for the first time, music shifted from "finding a suitable song" to slowly becoming "creating a suitable song."
These two actions may seem different by just one word, but the worlds behind them are completely different.
When searching for a song, there was always a stranger at the other end. It might be an Italian musician you've never heard of, a college student who just uploaded their first demo, or a German band that disbanded decades ago. Xiami was responsible for taking you there.
When generating music, human presence is not required at all. You describe the mood, and the model provides the melody; if you're not satisfied, you make another one. Music has become something that can be instantly produced as needed. It doesn't have to belong to anyone first, nor does it have to wait to be discovered by anyone.
That's why the name Joyful Xiami seems a bit strange here.
Eighteen years ago, Wang Hao named the website EMUMO, short for Earn Music & Money. One of the problems he initially wanted to solve was how music should be monetized in the internet age, how to help music creators earn more.
Today, Joyful Xiami is facing another problem. Whether good music already exists and is no longer even important.
As a result, copyright issues have reappeared, just in a different form. The previous dispute was about a song being uploaded to the platform without permission, and who should receive the money; now the issue has turned into which songs the model has learned from, and to whom does the new music it generates belong.
On August 17, the day Joyful Xiami was launched, the U.S. independent music publisher Round Hill Music sued Suno, accusing it of using hundreds of song lyrics without authorization to train its model.
The old Xia Mi helped people find songs, while the happy Xia Mi created songs for others.
In the former world, a strange song could change a person's aesthetic taste. Listeners would follow it all the way, and eventually, some even became musicians themselves. In the latter world, all a person needs to do is say what they want to listen to.
Then, wait for a minute, and the song arrives.
Swimming Until the Sea Turns Blue
Author Yu Hua lived in Haiyan, Zhejiang when he was a child, and the sea he saw was always yellow. However, textbooks said the sea is blue.
He wondered where the blue was exactly, so he swam into the sea. As the shore gradually drifted away and the water became deeper, he kept swimming forward. Children tend to believe in many specific things, such as reaching that kind of blue described in textbooks as long as they swim far enough.
In the end, of course, he did not find it. But many years later, he still remembered that he once swam in that direction.
I always feel that what the old Xia Mi left behind is a bit like this.
In the era it existed, many things were still distant. A song that one truly liked might not be liked at first listen. An unfamiliar name would not immediately become part of your aesthetic taste. People would continue on because of a faint sound they heard by chance, not understanding it today, but coming back in a few months; moving from one person to another, from one type of music to another.
In retrospect, it is realized that many of one's preferences developed during these detours.
Over the years, the Internet and AI have been shortening this distance. Search makes a song easier to find, recommendations bring unfamiliar music to your ears. With generative AI, this process has taken a step forward: a person can even describe a mood, have a suitable song written out without waiting.
One minute later, the music is here.
This has certainly changed many things.
In the past, there was a long road between the audience and the creators. Someone wrote songs, someone recorded, someone distributed, someone organized, someone recommended, and finally, someone stumbled upon it. Xia Mi spent twelve years, setting up many signposts along this road, making it easier for people to discover music they wouldn't have encountered otherwise.
Today, this road is getting shorter. Sometimes, even the audience can cross to the other side of the road. Someone who can't play an instrument or understand music theory can try turning the vague emotions in their mind into a song for the first time. Perhaps it's not perfect, maybe there are still clear machine traces, but the act of "creating music" has never been closer to ordinary people.
This may also be the interesting point between the Happy Shrimp and the Old Shrimp.
Eighteen years ago, Shrimp tried to make more music heard. Eighteen years later, another shrimp began trying to make more people make music. They are facing not the same music world anymore, nor is there a need to answer the same question.
Yu Hua didn't swim to the blue sea when he was a child. Later generations may not need to swim that far anymore. Technology keeps bringing distant things closer, putting in front of more people what only a few could do in the past.
But people still have to decide where to swim.
HappyShrimp can generate a song in about a minute.
But some things may just happen beyond that minute.
Original Article Link

TL;DR
Bernstein used South Korea's export data as an indicator of HBM revenue, estimating Samsung's third-quarter HBM revenue to reach $12 billion, about 30% higher than its original forecast.
In July, Samsung's related region's exports increased by 122% compared to April, with unit prices doubling, indicating that the growth may be primarily driven by the high-price HBM4 volume.
SK Hynix's related exports declined by about 27% compared to April, and the baseline regression model indicates a 20% quarterly decline in HBM revenue for the third quarter. However, single-month data and seasonal differences introduce significant uncertainty into the results.
Bernstein attributes SK Hynix's weakness to delayed shipments of HBM4 related to Rubin, but this is still an analyst's speculation and not a confirmed causal relationship by NVIDIA or SK Hynix.
HBM prices have not skyrocketed in sync with traditional DRAM prices. Samsung's increase in unit value reflects more of a product mix shift to HBM4 rather than a significant price hike for similar-spec products.
Bernstein remains optimistic about Samsung, SK Hynix, and Micron, believing that the next round of earnings revisions is more likely to come from the 2027 HBM contract prices rather than relying solely on this year's market share changes.
South Korea's July memory export data provided the first set of leading signals for Samsung and SK Hynix's third-quarter HBM business.
Bernstein has long been tracking South Korea's multi-chip memory exports to China, Taiwan, and Malaysia. The report suggests that these exports have a strong correlation with Samsung and SK Hynix's quarterly HBM revenue: Taiwan is a key location for CoWoS advanced packaging, and Malaysia has Intel's EMIB packaging facilities.
In July, South Korea's multi-chip memory exports to China, Taiwan, and Malaysia declined by 32% from the previous historic high in June. However, the report attributes this change mainly to intra-quarter seasonality. Compared to the first month of the previous quarter in April, July exports still increased by 13%, with a year-on-year increase of 64%, indicating that overall HBM demand has not significantly weakened.
While maintaining strong overall volume, the export trends of the two South Korean memory manufacturers have diverged: Samsung's related exports continue to rise, while SK Hynix's related data are significantly below Bernstein's expectations.
Samsung's Third-Quarter HBM Revenue May Exceed Expectations by 30%
Bernstein used Chungcheongnam-do's exports as a proxy indicator for Samsung's HBM shipments, as Samsung's related backend production and packaging facilities are mainly located in that region.
In July, Chungcheongnam-do's exports of multi-chip memory to China, Taiwan, and Malaysia reached $2.2 billion. Despite a 35% decline from the seasonal peak in June, this amount represented a 122% growth from April. Samsung's related exports have also exceeded SK Hynix's for two consecutive months.
According to the regression analysis based on historical export data and HBM revenue, Bernstein estimates that Samsung's HBM revenue in the third quarter could reach around $12 billion, representing a sequential increase of about 80% and exceeding its original forecast of around $9.3 billion by about 30%.
If only recent data since the first quarter of 2025 is used for regression, the forecast value would further rise to $12.6 billion, corresponding to a sequential growth of about 90%.
South Korea's monthly export value of multi-chip memory to China, Taiwan, and Malaysia (in billion dollars) and year-on-year growth rate. Although the July export value declined sequentially due to seasonal factors, it still increased by 64% year-on-year, indicating that the overall HBM demand remains robust.
However, this number is still a model estimate and not company guidance. Samsung's exports usually concentrate in the last two months of the quarter, with the first month of the quarter averaging less than 20% of the total quarter's exports since 2025. Therefore, whether the third-quarter revenue can ultimately reach the model's forecast will depend on the actual shipments in August and September.
Bernstein believes that the July data is at least consistent with Samsung's previous direction: the company expects third-quarter HBM4 sales to more than triple sequentially and account for over 60% of HBM sales in the second half of 2026.
Doubling Unit Value, HBM4 Becomes Samsung's Growth Driver
In addition to the export volume, the export unit value of Samsung's related products has also significantly increased.
Export value of multi-chip memory from Chungcheongnam-do (Samsung's HBM packaging location) to China, Taiwan, and Malaysia (in million dollars) compared to Samsung's quarterly HBM revenue (average per quarter). The July export value reached $2.2 billion, a 122% increase from April, marking the second consecutive month higher than SK Hynix. Right Chart: Based on historical data regression analysis, the July export data suggests that Samsung's third-quarter 2026 HBM revenue could reach $12 billion, with a sequential growth of about 80%, exceeding Bernstein's original forecast by about 30%.
In July, the export unit value of multi-chip memory from Chungcheongnam-do increased by another 21% sequentially, more than doubling from the April level and approaching four times that of SK Hynix's related region. Due to the higher capacity, stacking layers, and technical complexity of HBM4, Bernstein believes that this change likely reflects a rapid increase in Samsung's HBM4 shipment share.
This does not mean that Samsung has doubled the price of HBM with the same specifications within a few months. The unit export price will also be affected by product structure, capacity specifications, and packaging combination, and can only serve as a directional indicator of the average selling price. A more reasonable explanation is that Samsung is replacing some HBM3 and HBM3E with a higher-priced HBM4, thereby increasing the overall export value.
This is also a key opportunity for Samsung to regain market share. Previously, SK Hynix took the lead with HBM3E and a closer supply relationship with NVIDIA; as we enter the HBM4 cycle, the competition is shifting towards who can complete validation earlier, improve yield rates, and achieve scale delivery.
SK Hynix’s Export Performance Weakens, But Monthly Data Does Not Directly Equate to Market Share Reversal
In contrast to Samsung, SK Hynix's related exports notably weakened in July.
Bernstein used exports from Chungcheongbuk-do and Icheon as indicators of SK Hynix's HBM shipments. In July, the export value in these two regions decreased by 28% compared to June and by about 27% compared to April.
Based on this, the baseline regression model estimates that SK Hynix's HBM revenue for the third quarter may be only $5.6 billion, a decrease of about 20% from the previous quarter, which is 55% lower than Bernstein's original forecast. If this prediction holds true, Samsung's HBM revenue in the third quarter will significantly surpass that of SK Hynix.
However, the uncertainty of this prediction is much higher for SK Hynix than for Samsung. The report also points out that if SK Hynix follows its historical seasonal pattern with shipments concentrated in the latter part of the quarter, its HBM revenue for the third quarter could still reach $12 billion, close to Bernstein's original forecast.
In other words, based on the same set of July data under different seasonal assumptions, there could be a huge range of $5.6 billion to $12 billion. At this stage, a more cautious assessment is not that "SK Hynix has already lost its lead," but rather that its performance at the beginning of the third quarter was weaker than expected, and whether it can catch up in August and September will determine the final market share.
Bernstein speculates that the weakness may come from delays in Rubin-related HBM4 shipments. SK Hynix had previously stated that HBM4 began mass production in the second quarter and will ramp up fully in the second half of the year. However, attributing the July export decline directly to NVIDIA's Rubin progress is still an analyst's inference based on industry dynamics and has not yet been explicitly confirmed by the companies involved.
HBM Did Not Follow the Traditional DRAM Price Increase Trend
The report also points out that the price trends of HBM and traditional memory are diverging.
Since the third quarter of 2025, the price of traditional memory has increased by about five times cumulatively, but the export unit price of SK Hynix's related products has remained generally stable. Samsung's unit price has approximately doubled, mainly driven by the increased proportion of HBM4, rather than a significant simultaneous price hike of existing products like HBM3E.
This means that the contract pricing, supply relationships, and product iteration cycle of HBM are partially detaching it from the traditional DRAM spot price fluctuations. The short-term price increase of traditional DRAM does not necessarily proportionally transmit to HBM, as the profit variations of the latter depend more on the proportion of new-generation products and annual customer contracts.
Bernstein believes that suppliers and customers have already begun negotiating 2027 HBM contracts, and expects that a price increase will drive upward revisions to Samsung's and SK Hynix's profit outlook next year. Compared to the share changes in a single quarter of the third quarter, this may be a more critical pricing variable for the next phase of the memory sector.
Malaysia's Exports Surge, Remains an Unresolved Variable in the Supply Chain
In July, South Korea's exports of multi-chip memory to Malaysia increased by 20% month-on-month to around $1.3 billion, continuing the rapid growth of the past several quarters. Samsung still accounts for the majority of this, but SK Hynix's exports have also increased significantly.
Bernstein speculates that this HBM may be flowing to Intel's EMIB advanced packaging facility in Malaysia, to equip server chips with HBM. However, the report also acknowledges that it is currently difficult to explain why the related exports are growing so early before product mass production.
Therefore, Malaysia's exports are more suitable as a follow-up supply chain clue, rather than directly equivalent to Intel's customer orders or specific product volume.
Overall, the July data strengthens the logic of Samsung catching up in the HBM4 cycle, but it is not yet enough to confirm that the long-term market share pattern has reversed. What really needs to be observed next is whether Samsung can continue the export growth in August and September, and whether SK Hynix can catch up on the shipment gap from the beginning of the third quarter as HBM4 ramps up.
Market Share Reversal Still Awaits August and September Data
The July export data reinforces the logic of Samsung accelerating its catch-up in the HBM4 cycle: the related export scale has significantly increased, unit value has rapidly risen, and it corroborates the company's provided HBM4 sales guidance.
However, this data is still insufficient to confirm that Samsung has already regained its leading position in the HBM market on an annual basis.
First, Bernstein's model uses regional exports and company revenue for regression, not actual order data from the companies. Second, if HBM packaging is shifted overseas from South Korea, or if some products are further packaged domestically in South Korea, customs data may not fully capture this. Finally, monthly exports are also subject to customer acceptance, shipping arrangements, and quarterly seasonal effects.
Next, we need to observe three variables: whether Samsung's August and September exports can continue to grow, whether SK Hynix's exports and unit value will rebound with the ramp-up of HBM4, and whether the contract price for HBM in 2027 for both companies can be substantially increased.
If Samsung's exports continue to grow while SK Hynix fails to catch up, HBM4 may drive a more substantial change in market share. If SK Hynix concentrates its deliveries in the last two months of the quarter, the divergence in July is more likely just a timing difference in shipments.
Therefore, the current data is not so much about the HBM competition having reached a conclusion, but rather about Samsung re-entering the race for market share at a pace that may be faster than previously anticipated by the market.

Translation: Peggy
Editor's Note: On August 19, the U.S. Treasury unexpectedly announced that it would increase the single-day liquidity support repurchase cap for 10-20 year and 20-30 year nominal Treasury securities from $20 billion to at least $40 billion, with the new arrangement set to take effect from September 9. Following the news, the long-end Treasury yields briefly declined by around 9 basis points, leading to a notable flattening of the yield curve.
However, the market quickly reverted to selling. The next day, the 10-year Treasury yield rose to 4.71%, and the 30-year yield approached its previous high. This prompted the market to question: if the repurchase size is relatively limited and has not yet been actually implemented, why did the Treasury choose to make an interim adjustment to the plan just two weeks after the quarterly refunding announcement?
ZeroHedge cited a report from J.P. Morgan's rate strategist Jay Barry, suggesting that the Treasury may not be addressing market liquidity dysfunction but rather expressing concerns about the rise in long-term yields. J.P. Morgan is genuinely worried not about the $40 billion repurchase itself, but whether the Treasury is deviating from "conventional and predictable" debt management principles to a more opportunistic approach to tenor and issuance management.
This distinction is crucial for the long-term pricing of Treasuries. If investors believe the Treasury is trying to suppress financing costs through repurchases or reducing long-dated supply without simultaneously improving the fiscal deficit, the temporary downward pressure on yields may not be sustainable. Instead, it may lead to an increase in term premiums, raising the cost of long-term borrowing.
Translation of the original text:
After the U.S. Treasury expanded its long-term Treasury bond repurchase, the market initially responded positively.
The Treasury announced that the single-day liquidity support repurchase cap for 10-20 year and 20-30 year nominal Treasury securities would be increased from a maximum of $20 billion to at least $40 billion. According to the Treasury's announcement, the new size will take effect from September 9, rather than entering the market immediately on the day of the announcement.
Following the news, long-end Treasury yields fell by about 9 basis points, and the yield curve also showed a similar level of flattening. However, this market trend did not last long. The next day, the 10-year Treasury yield rose to 4.71%, essentially reversing the post-announcement downturn.
JPMorgan Chase believes that the most noteworthy aspect of this repurchase adjustment is not the size, but the timing: the Treasury Department had just released a tentative repurchase schedule for the next three months in the August 5 quarterly refunding announcement, when the single-day repurchase limit for 10-20-year and 20-30-year bonds was still $20 billion.
Why Did the Treasury Department Increase the Size on Short Notice When the Market Was Functioning Normally?
US Treasury bond repurchases are mainly divided into two categories: cash management repurchases and liquidity support repurchases.
This adjustment targeted the latter. The Treasury Department repurchases less liquid off-the-run securities, i.e., bonds no longer from the most recent issuance, to improve the trading efficiency between different securities and provide market participants with predictable exit options.
Under this mechanism, the key consideration for increasing the repurchase size should typically be whether market liquidity is deteriorating.
Referring to the evaluation framework proposed earlier by the Treasury Borrowing Advisory Committee, JPMorgan Chase examined the repurchase auction sizes, the Treasury curve dislocation, and the valuation gaps between on-the-run and off-the-run bonds. The conclusion was that the relevant indicators for 10-20-year and 20-30-year bonds remained close to the average levels of the past year, showing no significant market dysfunction.
The report stated that the pricing bias of off-the-run bonds relative to the fitted yield curve has remained stable, significantly lower than the extreme levels of the past five years; the asset swap spreads between on-the-run and off-the-run bonds have not experienced significant dislocations. Overall, the operation of the US Treasury market this year has even improved.
Therefore, JPMorgan Chase interprets this temporary adjustment as a policy signal: the Treasury Department's concern may not be liquidity, but rather the long-term yield itself.
The Treasury Department Might Be Developing a New Long-End "Reaction Function"
JPMorgan Chase believes that there may be a common theme in recent policy actions—the Treasury Department showing increased sensitivity to rising long-term yields.
This involves a common market concept: reaction function, which is an informal rule where investors infer what actions policymakers may take under certain conditions based on their past statements and actions.
In JPMorgan Chase's view, the Treasury Department's decision to announce the repurchase adjustment a few hours before the 20-year bond auction and a day before the 30-year Treasury Inflation-Protected Securities auction may indicate its desire to alleviate long-end financing pressures. However, this is still an analyst's interpretation of policy intent, not a confirmed policy objective by the Treasury Department.
The report also points out that this year, U.S. bond yields have risen, with a significant portion of this move explained by the market's hawkish repricing of the Fed's policy path. According to J.P. Morgan's fair value model, the 10-year yield has not significantly diverged from fundamentals.
What has truly deviated is the longer end of the curve. Global long-term bond yields have generally risen, especially with the surge in Japanese long-term government bond yields, diminishing the relative attractiveness of U.S. bonds to some overseas investors.
During Japan's implementation of a negative interest rate policy and yield curve control, U.S. bond yields, after currency hedging, were more attractive than Japanese government bonds, helping to suppress U.S. long-end rates. Now, this mechanism is partially reversing: as Japanese long-term rates rise, it may reduce Japan's incentive to allocate funds to U.S. bonds and amplify upward pressure on the U.S. yield curve's longer end.
Repo Operations Address Symptoms, but Deficit Is the Core of Term Premium
J.P. Morgan's main criticism of the Treasury's strategy is that while repo operations can alleviate short-term pressures in the long-bond market, they cannot alter the fiscal backdrop of continuously increasing U.S. debt supply.
The report projects that the U.S. financing gap over the next several fiscal years could exceed $3.5 trillion. In this environment, the Treasury may ultimately need to offer more duration to the market, not less. Even if the Treasury reduces the size of long-bond auctions, it only shifts the funding requirement to other tenors without eliminating the overall borrowing need.
J.P. Morgan also notes that there has not been a collapse in demand at long-bond auctions. End-investor participation in 30-year Treasury bonds is at a record high for the year, and 20-year bond demand is also near a historical peak. This further weakens the argument that "temporary ramp-up through repos is necessary to improve market functioning."
The underlying issue remains the fiscal deficit. J.P. Morgan describes the current fiscal situation as a deficit of about 6% of GDP; the Congressional Budget Office's February baseline forecast for the 2026 fiscal year deficit is around $1.9 trillion, about 5.8% of GDP, with both estimates aligning closely.
The report suggests that without substantial fiscal consolidation, the market may view more flexible and opportunistic debt management as lacking credibility. Should the Treasury further deviate from "conventional and predictable" issuance practices, investors may demand a higher term premium to compensate for future supply, inflation, and policy uncertainty.
This implies that an operation aimed at lowering long-end rates may, in the long term, carry the risk of raising yields. However, this remains a risk scenario presented by J.P. Morgan rather than an outcome that has already occurred.
UK Experience: Adjusting Long-Bond Supply with Diminishing Impact
To assess whether the continued reduction in long-term debt supply can sustainably depress yields, J.P. Morgan referred to the UK experience.
The UK has lowered the proportion of long-term gilts to the net issuance from an expected 28.4% in April 2022 to the current 9.1%. Over the past four years, the UK DMO has adjusted the long gilt issuance share 12 times.
These adjustments often manage to flatten the yield curve in the short term. J.P. Morgan's analysis shows that in the five days following the announcement, the spread between UK 5-year and 30-year gilt yields narrows by an average of around 2 basis points; looking at a ten-day window around the announcement, the average reduction is around 5 basis points.
However, this impact is not enduring, and with repeated similar operations, each announcement's uplifting effect on long-end gilts gradually diminishes. Despite the UK base rate falling 175 basis points from its cyclical peak, long UK gilt yields remain near multi-decade highs.
Based on this assessment, J.P. Morgan believes that the US expanding repo or reducing long-dated issuance may also temporarily lower long-end yields but could struggle to alter the longer-term trajectory. Without a simultaneous narrowing of the fiscal deficit, structural supply adjustments are unlikely to serve as a sustainable tool for lowering funding costs.
Going forward, the market will need to observe not only whether the Treasury continues to increase repo sizes but also whether it trims the auction sizes of 20-year and 30-year bonds, and if there are substantive changes in the fiscal deficit, foreign demand, and term premium.
If long-end yields continue to rise post-repo actualization or if the market's reaction to each policy tweak becomes increasingly short-lived, it would bolster J.P. Morgan's view that "debt management cannot substitute fiscal rectitude." Conversely, if market liquidity metrics significantly deteriorate while repos continue to enhance trading efficiency, this adjustment is more likely to be deemed a technical operation rather than the Treasury's direct attempt to control long-term rates.

КРАТКО
·Фаза III исследования INTerpath-001 успешно подтвердила способность мРНК-платформы Moderna выходить за рамки респираторных вакцин, однако полные данные об эффективности не были раскрыты.
·Необычайный рост Moderna на 177% за один день трудно объяснить одной лишь ценностью платформы: свою роль сыграли переоценка платформы, короткое покрытие и нарратив о «раковой вакцине», которые в совокупности подпитывали ралли.
·Успех при меланоме нельзя напрямую экстраполировать на другие показания, такие как рак легкого и рак почки, что подчеркивает важность воспроизведения результатов на разных типах опухолей для реализации ценности платформы.

Ранее мы потратили много времени на обсуждение логики роста долгосрочных доходностей гособлигаций в Европе и Соединенных Штатах, а также различных методов, которые центральные банки и министерства финансов используют для решения этой проблемы. Наконец, прошлой ночью мы увидели, как Министерство финансов США напрямую использует соглашения об обратном выкупе, чтобы подавлять доходности долгосрочных государственных облигаций. Глава Минфина заявил: «Текущий максимальный размер в 2 млрд долларов на операцию будет как минимум 4 млрд долларов на операцию». Думаю, я хотел бы поделиться несколькими моментами: 1. Хотя это не в традиционном смысле то, что известно как Контроль за кривой доходности (YCC), с точки зрения определения и операционного субъекта, вы можете найти множество различий, но подобные придирки не имеют смысла; по сути, правительство напрямую вмешивается в собственные расходы по финансированию. И я также думаю, что нет необходимости спорить о том, будет ли эта практика иметь долгосрочный эффект; ключевой вопрос — насколько далеко правительство готово зайти и какую цену оно готово заплатить. 2. Через несколько дней замечания ФРС или Варша на Jackson Hole стали очень тонкими. Известно, что он часто взаимодействует с Бенсоном, и рынок всегда говорил, что ФРС необходимо повысить ставки и усилить коммуникацию, чтобы снизить неопределенность политики. Однако вы можете видеть, что Варш не так ястребин: в последних протоколах он пытается продолжить уменьшать частоту коммуникаций. Это различие очень заметно. Я думаю, в конечном итоге это вопрос и краткосрочного, и долгосрочного горизонта.

Во второй половине дня 19 августа Казначейство США изменило ряд данных на своем официальном веб-сайте.
Разовый лимит для сделок обратного выкупа долгосрочных казначейских облигаций был увеличен с 20 миллиардов долларов до 40 миллиардов долларов.
В тот день не было новых сделок по обратному выкупу. Казначейство не потратило ни дополнительного доллара. Тем не менее доходность 30-летних казначейских облигаций быстро снизилась примерно на 9 базисных пунктов.
Здесь базисный пункт — это сотая доля процента в процентной ставке. 9 базисных пунктов равны 0,09 процентного пункта. Доходность облигации — это процентная ставка, под которую США предлагают занять деньги, поэтому растущая доходность означает более высокую стоимость заимствований.

Предлагаемое масштабное долговое финансирование Broadcom выводит на первый план ограничение по денежным потокам для инфраструктуры ИИ. Местные одобрения и развертывание в компаниях одновременно доказывают, что помимо финансирования границами расширения остаются электроснабжение, лицензирование и управление данными.
1｜Broadcom планирует более $600 млрд долгового финансирования, расширяя кредитную цепочку для инфраструктуры ИИ
Reuters сообщило 21 августа, что Broadcom ведет переговоры с группой кредиторов о привлечении более $600 млрд долгового финансирования для компаний вроде чипов ИИ Anthropic. Bloomberg заявил, что план может включать примерно $300 млрд субординированного долга и $600–$700 млрд в части старшего обеспеченного долга при общей величине до $1 трлн. Сделка пока не завершена: Broadcom и Apollo не ответили на запросы о комментариях, а Blackstone отказалась комментировать.

Перевод: Пегги
Примечание редактора: 19 августа Министерство финансов США объявило о расширении программ обратного выкупа долгосрочных облигаций, увеличив разовый максимальный объем выкупа для казначейских бумаг со сроком 10–20 лет и 20–30 лет с $20 млрд до как минимум $40 млрд. Ранее доходность 30-летних облигаций резко выросла примерно до 5,34%, достигнув максимального уровня с 2007 года. После объявления новости доходность по «длинному» участку быстро откатилась назад.
Это дало рынку четкий бычий сигнал: Казначейство США активно улучшает ликвидность по долгосрочным облигациям, возможно, даже формируя ожидания какого-то механизма "страховки Казначейства".

HYPE взлетел примерно с $58,5 до $73,7 за последние два дня, показав рост на 26%. Сейчас он находится всего примерно в 4% от своего исторического максимума $77.
По мере приближения цены к новому максимуму Hyperliquid также наблюдает позитивные события.
Договор pre-IPO поступает в SEC
Первая новость поступает из Hyperliquid Policy Center. Он сотрудничает с trade[XYZ], чтобы направить в SEC письмо с комментариями в ответ на публичные консультации регулятора по реформированию процесса IPO, предлагая включить при включении в листинг бессрочные контракты в обсуждение на уровне регулирования.

Перевод: Пегги
Примечание редактора: На этой неделе доходность 30-летних казначейских облигаций США выросла примерно до 5,34%, достигнув максимума, не наблюдавшегося с 2007 года. Затем министр финансов США Бесент объявила о расширении выкупа долгосрочных облигаций: максимальный размер одного выкупа для облигаций со сроком 10–30 лет увеличен с $20 млрд до как минимум $40 млрд, а соответствующие меры должны быть реализованы в период с 9 сентября по 4 ноября. После объявления долгосрочные доходности откатились назад, доллар ослаб, а рынки рисковых активов получили некоторую поддержку.

Перевод: Пегги
Примечание редактора: 19 августа Министерство финансов США объявило о расширении программы выкупа долгосрочных облигаций, увеличив размер разовой дневной поддержки ликвидности для некоторых облигаций со сроком 10–30 лет с 20 млрд долларов до как минимум 40 млрд долларов. Ранее доходность 30-летних казначейских облигаций ненадолго поднялась до 5,337%, достигнув нового максимума с 2007 года; после выхода новости долгосрочная доходность быстро отступила.
Рынок сразу начал задаваться вопросом: показывает ли Министерство финансов более явную чувствительность к быстрому росту долгосрочных ставок?

Согласно новостям PolyBeats, 19 августа по пекинскому времени один человек смог получить доход в 120% всего за 89 секунд, предсказав «сможет ли Россия занять определённую территорию на поле боя».
Этот человек вложил $4 000, чтобы купить прогноз «Россия захватит Константиновку в этом месяце» на платформе под названием предиктивный рынок. Вероятность того, что это событие случится, на тот момент составляла лишь 43%. Спустя 89 секунд эта вероятность выросла до 92%.
На этой платформе «вероятность-цена» если событие произойдёт, вероятность будет зафиксирована на 100%, и стоимость позиции соответствующим образом увеличится. Таинственный человек разместил ордер, стал свидетелем резкого всплеска, а затем вышел по тейк-профиту менее чем за две минуты. В эпоху, когда существует множество торговых стратегий, удача в трейдинге подобного рода — не редкость.

Письма, переписка в чатах, проектные документы, наряды на выполнение работ — раньше это были просто цифровые остатки, ожидавшие очистки после закрытия компании. Теперь их заново оценивают, упаковывают, продают и направляют в тренировочный конвейер AI-компании.
Похоронный агент сохраняет последнее достоинство покойного. Постмортальное достоинство предприятия — доказать, что то, что оно оставило после себя, по-прежнему имеет ценность.
17 августа на аукционе по продаже активов в рамках банкротства Google сделала заявку в размере 10 миллионов долларов, чтобы приобрести все корпоративные данные Spirit Airlines. Другой участник, Mercor, предложил 7,5 миллиона долларов — не дотянув до требуемой суммы на 2,5 миллиона.

Может ли тот, кто однажды помог Соросу сломить Банк Англии, теперь использовать те же тактики, чтобы защитить рынок казначейских облигаций США?
С начала этого года министр финансов США Скотт Бессент делает последовательные шаги: применяет серию неожиданных операций на рынке, чтобы поставить на кон свою репутацию подавления стоимости заимствований в Америке. По данным Bloomberg, он стал «самым активным министром финансов за десятилетия, вмешивавшимся в финансовые рынки».
После совместной интервенции США и Японии по иене последний шаг Бессента — расширить выкуп американских облигаций. Министерство финансов объявило, что оно «как минимум удвоит» объем выкупа 10–30-летних казначейских облигаций — и этот план выкупа был озвучен лишь две недели назад. В день публикации новости доходность 10-летних казначейских облигаций снизилась примерно на 6 базисных пунктов, доходность 30-летних — почти на 9 базисных пунктов, а индекс доллара США также опустился до минимума за три месяца.

КРАТКО
· Казначейство США расширит выкуп (репо) номинальных купонных ценных бумаг в сегментах 10–20 лет и 20–30 лет, увеличив предельный лимит в рамках одной операции с 20 млрд долларов до как минимум 40 млрд долларов.
· Эта операция направлена на временное улучшение ликвидности для долгосрочных ценных бумаг, снижая предельные премии за срок, но не является частью количественного смягчения со стороны ФРС.
· Связанные инструменты: TLT, QQQ, золото, BTC и акции роста, чувствительные к долгосрочным доходностям.
19 августа Казначейство США объявило о расширении операций по поддержке ликвидности для долгосрочных облигаций, увеличив предельный лимит в рамках одной операции для номинальных купонных ценных бумаг в сегментах 10–20 лет и 20–30 лет с 20 млрд долларов до как минимум 40 млрд долларов.

Google привязывает опционы на акции к поставкам чипов, клиническим результатам и усиленному короткому сжатию: долгосрочные ожидания превращаются в цену. Последние события в Homsud указывают, что вопрос о том, сможет ли поставка стабильно достигать рынка, остается жестким ограничением.
1｜Google привязывает опционы на акции к Marvell: закупки с встроенным в тот же контракт стимулом к капиталу
Marvell сообщила, что выпустила варранты для Google на подписку до 58,97 млн акций обыкновенного капитала по цене исполнения $206,58 за акцию, которые могут быть реализованы не позднее августа 2033 года. Если все они будут исполнены, то при цене исполнения будет рассчитано примерно $12,2 млрд. Варранты привязаны к целевому показателю выручки Google для закупок кастомных чипов.