Mirae Asset：下调海力士目标价至280万韩元，近期股价跌幅过大但基本面并未明显恶化
Related
Mirae Asset's Hong Kong branch will launch a tokenized asset trading platform within three years, initially targeting Hong Kong clients.
According to a report by Straight News, Mirae Asset's Hong Kong branch plans to launch a tokenized asset trading platform within the next three years. This platform aims to convert existing investment assets and traditional financial assets into tokens for clients to trade. The company has recently expanded its business based on its newly launched trading platform, MAPS (Mirae Asset Portfolio Service), striving to transform from institutional business to a client-centric platform. Initially targeting Hong Kong clients, the platform's long-term goal is to establish a cross-regional trading architecture connecting South Korea and overseas markets.
Google Cloud will provide specialized AI models for scientific research.
According to Mars Finance, Alphabet's Google will offer professional artificial intelligence models from software company Sandbox AQ through its cloud service platform. This move aims to expand the application of these technologies by businesses and research institutions, technologies specifically designed to accelerate drug discovery, materials science, and semiconductor manufacturing. (Cailian Press)
Mirae Asset Securities launches MAPS, an investment platform that integrates traditional asset and cryptocurrency trading.
According to Foresight News , citing News1, Mirae Asset Securities of South Korea has launched MAPS, a unified global investment platform that integrates traditional financial asset and digital asset trading into a single app. MAPS plans to gradually expand to major markets such as the United States, Japan, and Singapore.
Serenity: Google's computing power limitations may explain Meta's major signing of AI Neocloud, which is beneficial for AI data center capital expenditure.
According to Mars Finance, on June 28th, Serenity published an article stating that Google reportedly limited Meta's computing capacity in March 2026 due to scarce computing resources. This may explain why Meta signed a large-scale cooperation agreement with AI Neocloud service providers (such as NBIS) at that time, and also resulted in some limitations on the capabilities of the Gemini model. Serenity pointed out that Google's CEO previously stated in an earnings call that insufficient computing resources limited Google Cloud from handling more customer demand, with the backlog of related orders almost doubling compared to the previous quarter. Serenity believes that this situation further illustrates that the computing power supply of hyperscale cloud providers is still far below market demand, and they cannot even complement each other's resources. In the long run, this will continue to support the logic of AI data center construction and capital expenditure expansion.
Garrett Jin: The memory sector has peaked for the time being; hyperscale cloud providers are poised for a rebound.
According to BlockBeats, on July 2nd, Garrett Jin, an agent of "BTC OG Insider Whale," stated that the memory sector has reached a temporary peak. Micron encountered resistance at $1250 and fell back. Even after its earnings report significantly exceeded expectations, its stock price still fell sharply on high volume, indicating that the positive news has been fully priced in and there is no follow-up buying – a clear signal of a temporary top. Funds are rapidly withdrawing from memory stocks, with DRAM ETFs experiencing a significant drop on high volume. SK Hynix and Samsung are also under pressure. In the past two months, foreign investors have withdrawn over 100 trillion won (approximately $65 billion) from the South Korean market. However, funds have not left the AI sector but are rotating within it. While chip stocks were sold off last Friday, Google, Microsoft, and Amazon saw significant bottoming out on high volume, and Meta also rose sharply on high volume today. Garrett Jin believes the underlying logic behind the rotation is token optimization: as more and more workloads are handled by inexpensive models for simple tasks, value gradually accumulates in the cloud and orchestration layers billed by tokens, rather than the model layer itself. This is precisely the moat of hyperscale cloud providers. Based on this, he predicts that hyperscale cloud providers will experience a round of catch-up growth, suggesting reducing holdings in memory and joining the catch-up rebound of large cloud providers.
NVIDIA launches "AI Computing Partner Program": Offering computing power in exchange for revenue sharing and equity with cloud providers.
According to BlockBeats, on July 2nd, Nvidia is advancing a new project called the "AI Compute Partnership." Cloud service providers Firmus and Sharon AI, who are participating in the project, revealed that Nvidia has committed to providing these startup cloud service providers with a guarantee of computing power capacity: if the GPUs they purchase cannot be leased out, Nvidia will fund the leaseback of these idle computing power. In return, Nvidia will take a share of the startup cloud service providers' revenue (the share decreasing gradually with the contract term) or provide them with stock options. Because data center construction and GPU procurement are extremely costly, startup cloud providers with lower credit ratings often struggle to obtain financing. Nvidia's capacity swap guarantee essentially provides them with crucial credit backing, helping them secure loans. Currently, giants such as Amazon, Microsoft, SpaceX, Oracle, Meta, and Google purchase most of Nvidia's chips, but these giants are also developing their own chips. Nvidia hopes to reduce its dependence on large customers by supporting startup cloud providers. Back in September 2025, Nvidia pledged to underwrite $6.3 billion worth of unsold computing power for CoreWeave until 2032. Furthermore, Nvidia is currently in talks to provide similar financial guarantees for OpenAI's planned $500 billion mega-data center.