SnapshotLaborer

Recently, news that Meta is selling computing power has triggered sharp fluctuations in the market, with notable volatility in the technology sector. The market generally worries about an oversupply of computing power, thereby questioning whether the investment logic behind AI hardware has already broken down. At this moment of market divergence, Wall Street Insights invited Kong Rong, Vice President and overseas technology chief analyst at Guolian Minsheng Securities, to share the latest outlook for technology AI in the second half of 2026, after conducting on-site research and in-depth discussions with the industry supply chain.
I. Meta selling computing power ≠ computing power oversupply; Meta continues to build new data centers: Finding opportunities amid excessive panic
------------------------------------------
The biggest shakeup in the global tech sector recently came from Meta’s announcement that it would rent out computing power to the outside. The two things the market is most concerned about—“Is computing power becoming oversupplied?” “Is Meta giving up on large models?”—in our survey…

If the main theme of the semiconductor industry over the past decade and a half was “Moore’s Law,” then the loudest keyword today is definitely advanced packaging.
With large-model parameters surging from the tens of billions to the trillions, the path of simply boosting computing power through process miniaturization is approaching the physical limit. An AI chip must fit both massive numbers of compute units and high-bandwidth memory, and traditional 2D packaging is no longer up to the task. As a result, the “golden combination” of HBM + CoWoS has almost become the go-to choice for essentially all high-end AI chip makers.
From NVIDIA’s Blackwell-architecture GPUs, to AMD’s MI-series accelerators, and to training chips developed in-house by cloud companies—whoever can secure sufficient CoWoS packaging capacity will be the one that truly stands firm in the AI compute race.
A “positioning battle” centered on TSMC’s CoWoS packaging capacity has quietly begun among major global chipmakers.
Why is CoWoS a must?
---

The central bank has clearly stated that it will continue to increase its allocation of foreign exchange reserves to assets in Hong Kong, boosting the rebound in Hong Kong stocks and enhancing the market’s appeal. While the proportion of outstanding short selling in Hong Kong stocks has declined, it remains high, and there are signs of hedging activity. In the short term, focus should be on sectors with clear fundamentals such as innovative drugs, aviation, robots, and strong industrial metals. The U.S. stock market’s artificial intelligence theme has also shown signs of recovery; top-tier technology capital expenditure still has resilience. It is recommended to pay attention to sectors including software, defense, energy infrastructure, and financials.

China Oriental Red Satellite is expected to achieve net profit attributable to shareholders of the parent company of $30.5 million to $36.5 million in the first half of 2026. After excluding non-recurring items, net profit is expected to be $26.7 million to $32.0 million. It will turn from loss to profit, showing significant improvement compared with the year-ago period losses of $30.4915 million and $35.8403 million, respectively. The reasons include faster satellite R&D and manufacturing progress, more contract performance milestones that have been verified and accepted, and steady satellite applications. In 2026, the company will focus on driving transformation through development, coordinate space-aviation manufacturing and satellite applications, promote system overhauls such as cost control and integrated applications for communication, navigation, and remote sensing, advance the rollout of key projects, and explore new tracks. The above figures are preliminary estimates and have not been audited; please refer to the interim report disclosure.

When the Model S/X production line officially makes way for Optimus, when Unitree’s $42B valuation goes public on the STAR Market, and when the global shipment target for 2026 points to 50k units—this is a super-sized reshaping of an entire industrial chain that’s already underway.
A new industrial storm is at the breaking point. How should we understand and position the value of the industrial chain?

Universal Fusion completes and merges with a SPAC, with a valuation of $724 million and raising up to $338 million, with Bezos involved. The company follows a steampunk-style mechanical piston route using liquid-metal-filled cavity compression, unlike tokamaks. A new paper has prompted doubts about plasma temperatures and the commercial viability of the approach, and experts are divided in their views. The listing highlights intensifying competition for nuclear fusion financing; the industry has yet to prove it can achieve commercial power generation, and the future still needs to be verified.

Karp directly points to the momentum built by data and high pricing cultivated by large AI labs. Businesses worry they could become value contributors rather than value owners in the AI wave, and they have released a white paper, 《Institutional Sovereignty in the AI Era》, to help enterprises and governments protect themselves. Microsoft CEO Nadella also warned that if companies are only consumers of foundational models, it is difficult to retain corporate value, and disputes over who should own AI value are intensifying.

June CPI is expected to grow 3.8% year over year and -0.1% month over month. The sharp drop in energy significantly lowered inflation: oil prices and gasoline fell sharply, pulling CPI down by about 0.4 percentage points, while food and housing inflation eased. Core CPI remains sticky, with month-over-month above 0.2% and year-over-year around 2.8%. Core PCE could be stronger, and the Fed’s assessment has not changed. If core CPI stays elevated, the outlook for rate cuts will be limited; if it comes in below expectations, rate-cut expectations may reverse.

On July 9, Vaush announced that five external working groups would be led by him, with a lineup that includes a venture-capital titan, a Nobel Prize laureate, and former heads of central banks from multiple countries—an unprecedented move. The market thought he would only discuss rate cuts; in reality, he is reshaping the Federal Reserve’s decision-making cognition system: looking at data definitions, understanding inflation, assessing the impact of AI, determining the size of the balance sheet, and improving market communication—five directions pointing to the same endgame.

The first country to be shorted by AI has appeared.
Once-celebrated and thriving Indian stocks are now facing an unexpected collapse. On July 8, India’s Nifty IT index closed at 23,568—down 49% from 18 months earlier. The combined market value lost by 10 major IT companies exceeded 190 trillion rupees, accounting for 60% of India’s 2024 fiscal year budget.
Even stranger is that every time India’s stock market suffers a sharp plunge, it coincides with a major move by Silicon Valley’s AI:
On February 4, Anthropic rolled out enterprise-grade AI tools, and India’s Nifty IT index plunged nearly 6% that day, recording the largest single-day drop since March 2020;
On May 12, OpenAI announced it would invest more than $4 billion to set up a “front-deployment engineer” team, and the index fell another 3.7%, dropping to a three-year low;
Worse than the stock market are the shattered hopes of India’s “IT golden age”:

According to CCTV News, South Carolina Senator Lindsey Graham died late on July 11 due to a sudden illness at the age of 71. Two days before his death, he returned from Kyiv to Washington and was scheduled to appear on a program. He has served as a senator since 2003 and is seeking a fifth six-year term; the governor will appoint a successor to fill the vacancy. Graham previously served as chair of the Budget Committee, advocated a hard line toward Russia and Iran, and has visited Ukraine multiple times, meeting with Zelensky. His relationship with Trump shifted from criticism to close alliance; Trump mourned him, calling him a great person and a great senator.

Elon Musk and Altman traded blows on public platforms, and on the same day, two AI companies under their respective umbrellas released flagship models one after another. Musk accused Altman of taking fraud to new heights; Altman hit back, saying he was hawking short-term space data centers to the public markets. Musk also claimed Altman stole open-source AI from a charitable organization and Apple phone technology, and named the relevant lawsuits. Apple sued OpenAI, alleging that employees leaked unpublished information, among other things, to help it build its own in-house hardware; OpenAI responded that it is focused on innovation. The two new models, GPT-5.6 and Grok 4.5, are going head-to-head in the AI agent field: GPT-5.6 is strong at reasoning, business processes, and network security, while Grok 4.5 is more efficient in autonomous programming and developer workflows, with lower costs; in abstract reasoning, OpenAI still has the edge. Investors and enterprise users’ choices will depend on the specific scenarios.

Two major domestic storage leaders are moving forward with their IPOs. Changxin Technology will open for subscription on July 16 and will list on the STAR Market. Changcun Holdings, which was founded in the same year, has also disclosed its instructor-led training progress. The training institutions are CITIC Securities and CITIC Construction Investment, with 31 people assigned to review governance matters, advance shareholder-through verification, and continue pushing for rectifications and improving internal controls. Judging by the time taken by peers, Changcun Holdings is expected to complete its training in the middle to late part of August at the latest. Changcun Holdings was established in Wuhan in 2016. Its subsidiary Yangtze Memory provides products such as 3D NAND. Starting in June, there was an equity change: parties including Optics Valley acquired shares of Wuhan Xinxin. After the controlling structure was adjusted, Changcun Holdings’ shareholding reached 100% following the related changes.

US stock ETFs are undergoing an unprecedented, serious expansion in history: assets under management are about $16 trillion, possibly reaching $17 trillion by year-end. Net inflows this year exceed $1 trillion, and issuance size could surpass $2 trillion—setting a new record again. Trading volume in the first half increased year over year by about 50%, with an average daily value of about $325 billion; in June alone, it was $160k. Actively managed ETFs attracted about $400 billion, accounting for about 40%, but total assets are only 13%. Leveraged ETFs have boosted trading volume: in June, nominal trading was $1.1 trillion, with exposure nearing $3 trillion. About 5,400 new products were launched; DRAM-themed ETFs have already surpassed EWY, reshaping the storage theme landscape.

According to disclosures by the CSRC, the first phase of IPO tutoring for Yangtze Memory Storage was conducted from May 19 to June 30 by a 31-member team composed of China CITIC Securities and China CICC. The focus is on due diligence, driving corporate governance and internal control construction, studying securities regulations and conveying regulatory requirements, while also initiating shareholder information disclosure and shareholder look-through verification. The main issues are that governance and internal controls still need to be improved, and look-through verification involves a large workload. In the next stage, efforts will continue to advance due diligence and standardization work, strengthen regulatory training, and ensure compliance with information disclosure.

This article analyzes CPO competition: TSMC leads in the switch CPO space with COUPE+CoWoS, with its optical engine getting close to the switching ASIC; Samsung is advancing 2.xD packaging and plans to integrate HBM, logic chips, and silicon photonics chips into a single package to enable optical I/O for AI computing. The two operate in different domains—switch CPO versus XPU‑HBM optical I/O. The former emphasizes coordination between a single process and packaging, while the latter requires cross-domain optimization from the initial design stage. If, within the next 12 months, there are named orders featuring HBM/logic/optical I/O in the same package, Samsung may benefit; otherwise, TSMC still has the advantage thanks to its mature ecosystem.

Bloomberg’s charts show that the ratio of MSCI China relative to MSCI Global has fallen to its lowest level since 2001, highlighting that global equities are trading at elevated levels as they are boosted by AI and other factors, while China’s core assets are in a phase of valuation digestion and earnings recovery. MSCI China is a composite index covering A-shares, H-shares, red chips, P-shares, and overseas China companies, with heavy offshore weight. The top ten constituents account for about 40% of the total weight, mainly in consumer, financials, and communications, while the information technology weight is lower than MSCI Global’s exposure to the U.S. tech and semiconductor sectors. This helps explain the long-term performance differences between the two.