News Reminder
- Samsung Q2 profit surged 18-fold to $56.3 billion, AI memory demand pushed DRAM prices up 44%
- Boston Dynamics humanoid robot delivered the ball at the World Cup, billions of viewers saw embodied AI in action for the first time
- OpenAI's smartphone mass production plan accelerated to first half of 2027, has poached 40+ hardware talents from Apple
- Threads monthly active users reach 500 million, catching up with X, Meta's social landscape expands
- Foxconn Q2 revenue of $79 billion exceeded expectations, AI server demand fills consumer electronics weakness
- Meta released Brain2Qwerty v2, non-invasive brain-computer interface decoding performance sets record
Headline News
Samsung Q2 Profit Surged 18-fold to $56.3 Billion, AI Memory Demand Pushed DRAM Prices Up 44%
The computing power demand of AI infrastructure is rapidly translating into profits for chip manufacturers.
Samsung Electronics is expected to release its Q2 preliminary financial report on July 8. Based on predictions from 30 analysts, financial institutions estimate its April-June operating profit to be approximately 86 trillion won (about $56.3 billion), a surge of about 18 times year-on-year, marking the third consecutive quarter of record highs. The same period last year was only 4.7 trillion won.
The driving force of this growth has expanded from HBM high-bandwidth memory to all categories.
Citi Research data shows that Q2 DRAM average selling price rose 44% quarter-on-quarter, and NAND rose 53%. This is due to the rise of agentic AI (AI systems capable of executing multi-step complex tasks), which has increased demand for server processor memory and inference data storage. Analysts expect the memory market to remain in short supply at least until next year. Samsung's stock price has surged 158% year-to-date, joining SK Hynix (up 273%) and Micron (up 242%) in the trillion-dollar market cap club.
But risks are also accumulating. JPMorgan pointed out that AI memory already accounts for 52% of cloud service providers' capital expenditure, and is expected to exceed 70% next year, raising questions about the sustainability of this ratio. In addition, Samsung's semiconductor division needs to set aside 10.5% of operating profit as special bonuses for employees, with cumulative provisions possibly exceeding 40 trillion won, which may affect Q2 final data. (Reuters, Citi Research, JPMorgan)
Silicon Valley News
Boston Dynamics Humanoid Robot Delivers Ball at World Cup, Billions of Viewers See Embodied AI in Action
A humanoid robot completed a historic live operation on the world's largest sports stage.
On July 5, at the Brazil vs. Norway Round of 16 match held at the New York/New Jersey Stadium, Atlas, a robot developed by Boston Dynamics, walked through the player tunnel, performed several goal celebration moves, including Brazilian forward Matheus Cunha's iconic surfing celebration and Korean star Son Heung-min's photo pose, before handing the match ball to the referee.
Boston Dynamics' humanoid robot appeared at a FIFA World Cup match to deliver the ball, marking the highest-profile public appearance of a humanoid robot to date. The World Cup's global broadcast covers billions of viewers, effectively a public demonstration of embodied AI for all humanity.
This appearance marks the first public live demonstration of this robot since its mass-production version was unveiled at the CES consumer electronics show in January. From its on-site performance, its motion control capabilities have approached practical levels.
The significance of this event lies more in market education than technological breakthrough. There is still a huge gap between "ball delivery" and "replacing humans with complex tasks," but letting billions of people see a humanoid robot's actual performance for the first time is far more convincing than lab videos.
Hyundai Motor, the parent company of Boston Dynamics, has also gained incalculable brand exposure. (Fortune)
OpenAI Smartphone Mass Production Plan Accelerated to First Half of 2027, Has Poached 40+ Hardware Talents from Apple
OpenAI's plan to challenge the iPhone is accelerating across the board.
According to the latest supply chain survey by renowned analyst Ming-Chi Kuo, OpenAI's AI Agent smartphone mass production timeline has been moved up from the original 2028 to the first half of 2027, with total shipments expected to reach 30 million units in 2027-2028. The device is positioned as an "AI-native phone," with the core logic of replacing traditional app operation modes with natural language interaction. The chip uses a customized version of MediaTek's Dimensity 9600, based on TSMC's N2P process.
OpenAI's hardware offensive is extremely aggressive. The company has assembled a 200-person hardware team, with more than 40 from Apple, including former iPhone design head Evans Hankey and Vision Pro head Paul Meade. To retain talent, Apple has had to offer iPhone product design teams up to $400,000 in restricted stock retention bonuses. In terms of manufacturing, Luxshare Precision is the exclusive foundry partner. The intensity of this talent war is extremely rare in Silicon Valley hardware circles.
From a strategic perspective, the acceleration of the AI Agent Phone reflects OpenAI's consideration of listing pace. The company plans an IPO, and a convincing hardware product will greatly boost investor confidence. Meanwhile, OpenAI's first hardware in collaboration with Jony Ive (an AI smart speaker priced at $200-300) has been delayed to early 2027, giving the phone project higher internal priority. (MacRumors, Ming-Chi Kuo Supply Chain Report)
Chip Legend Jim Keller Builds His Own Fab, Mass-Produces Small Chip Equipment to Lower AI Chip Barriers
The "tape-out threshold" for AI chip startups is being rewritten by a legendary chip designer.
On July 5, Tom's Hardware reported that Jim Keller's startup Fab2 (formerly Atomic Semi, a US-based chip manufacturing startup) is building a factory in Texas to mass-produce small chip manufacturing equipment, targeting the small-batch foundry market for custom AI chips.
Jim Keller is a legendary figure in chip design, having led the design of Apple's A-series, AMD's Zen architecture, and Tesla's autonomous driving chips.
Fab2's business logic is completely different from the large-scale foundries of TSMC and Samsung. TSMC excels in advanced processes with million-level shipments, but the demand for custom AI chips and small-model inference chips is characterized by "small volume, many types, and fast iteration." Fab2's small chip manufacturing equipment is filling this gap, allowing AI startups to verify and mass-produce custom chips without hundreds of millions of dollars in tape-out costs. (Tom's Hardware)
Threads Monthly Active Users Reach 500 Million, Catching Up with X, Meta's Social Landscape Expands
Meta's social platform Threads has reached 500 million monthly active users, catching up with Elon Musk's X platform in user scale.
According to NYT on July 5, Threads' growth curve far exceeds industry expectations, mainly driven by the significant user exodus from X due to changes in content moderation policies and product direction after Musk took over.
Threads and X have formed a distinct positioning difference. Musk has reshaped X as a "free speech town square," but advertisers continue to leave, while Threads positions itself as a "more civil public discussion space" to absorb spillover users. Notably, Threads' user growth mainly comes from content consumption rather than original posting, and whether its activity and monetization capabilities can truly compete with X remains to be seen.
From an AI industry perspective, Threads' 500 million user base makes it an important channel for AI content distribution. Meta can embed Llama model-generated content and AI Agent public interaction scenarios directly into the Threads ecosystem, forming an AI application loop of "model to content to social distribution." Once this model works, it may be more imaginative than pure ad monetization. (NYT)
Foxconn Q2 Revenue of $79 Billion Exceeds Expectations, AI Server Demand Fills Consumer Electronics Weakness
NVIDIA's largest AI server manufacturer Foxconn (Hon Hai Precision) delivered better-than-expected results. For the second quarter ending in June, Foxconn's revenue reached NT$2.51 trillion (about $79 billion), exceeding the average analyst estimate of NT$2.37 trillion. Strong demand for AI-related products successfully compensated for slight declines in consumer electronics and computing product demand. (Bloomberg)
Meta Releases Brain2Qwerty v2, Non-Invasive Brain-Computer Interface Decoding Performance Sets Record
Meta has launched the most powerful non-invasive solution in the brain-computer interface field. On July 5, Meta AI team released Brain2Qwerty v2, claiming it is "the highest-performing non-invasive EEG decoding solution ever." The technology reads brain signals through external sensors and decodes them into text using AI models, without any surgical implants. Version 2 shows significant improvements in decoding accuracy and speed compared to v1.
In stark contrast to Neuralink's invasive approach requiring open-brain surgery, Brain2Qwerty takes a wearable device direction. If decoding accuracy continues to improve, the non-invasive path may become the mainstream route for BCI commercialization, as the vast majority of users would not undergo brain surgery just for typing speed. Meta's timing of this release suggests it sees BCI as a key entry point for "next-generation human-computer interaction."
From an application perspective, Brain2Qwerty can complement Meta's Muse Spark tool invocation capabilities, offering tremendous potential in AR/VR interaction and accessibility assistance. However, Neuralink has already achieved phased progress with its invasive approach in human clinical trials, and the competition between the two technical routes is still in early stages. (Meta AI Blog)
Amazon Stops Accepting New Customers for Mechanical Turk, AI Data Annotation Infrastructure Changes
The infrastructure for AI training data is undergoing structural changes.
On July 5, TechCrunch reported that Amazon will no longer accept new customers for the Mechanical Turk (MTurk) platform. MTurk is the core crowdsourcing platform for global AI training data annotation, with many AI companies relying on its crowd labor for data cleaning and annotation. Amazon did not give specific reasons for shutting down new customers.
The timing of this decision is noteworthy. The AI industry is in a period of explosive data demand, with the supply-demand gap for high-quality annotated data continuously widening. MTurk ceasing to accept new customers, as a cornerstone of data annotation infrastructure, means AI companies must find alternatives, either by turning to more expensive professional annotation services or investing in automated annotation technology.
The latter may be exactly what Amazon wants to promote: AWS has been promoting automated annotation tools, and limiting MTurk may be paving the way for its own AI data services. (TechCrunch)
Tesla Sets Employee AI Spending Cap at $200/Week, xAI Products Exempt
Tesla's attitude towards AI tools is undergoing a dramatic reversal.
According to an internal memo obtained by The Information, from July 6, Tesla will cap employee spending on third-party AI tools at $200 per week. Just a few months ago, the company was vigorously promoting AI usage, with some teams even establishing internal leaderboards ranked by token consumption to incentivize usage, resulting in software engineers often consuming "thousands of dollars in tokens" per week.
Beta versions of xAI products are not counted toward the $200 cap. (The Information, Electrek)
Blackstone's QTS Terminates 2,100-Acre Data Center Project in Virginia, AI Infrastructure Faces Community Resistance
The unchecked expansion of AI data centers is encountering increasing local resistance.
On July 2, Reuters reported that QTS, a data center operator under the world's largest alternative asset manager Blackstone, officially terminated the Digital Gateway project in Prince William County, Virginia, and withdrew all appeals to the Virginia Supreme Court.
The project planned an area of 2,100 acres (about 8.5 square kilometers) and went through years of planning and intense legal disputes. The project's demise stemmed from strong opposition from the local community. Prince William County residents have long protested the noise, power consumption, and environmental impact of data centers. QTS ultimately chose to abandon the project after years of litigation.
Although Northern Virginia remains the global data center capital, the triple constraints of land, electricity, and community tolerance are narrowing its expansion space. (Reuters)
Today's Hot Video
