Foldable iPhone May Be Hard to Get / Doubao and Qwen Pull Agent Features / Memory Prices Change by the Day
TL;DR
This Daily News edition covers a delayed foldable iPhone launch, Doubao and Qwen shutting down agent features, further DRAM and SSD price hikes, Anthropic's in-house chip plans, Huawei's Tao's Law V2, and AI data-center water use.
📱 Foldable iPhone launch reportedly delayed
🛑 Doubao and Qwen will shut down agent features
💻 iOS 27 beta code includes a B790 clue, possibly pointing to new AirPods with cameras
💰 Phones and PCs may get more expensive again as Samsung plans a third major DRAM price hike this year
🧠 Memory and SSD prices are changing by the day; dealers advise waiting unless the purchase is urgent
🤖 Anthropic is reportedly talking with Samsung on an in-house AI chip to catch OpenAI
⚙️ Huawei's He Tingbo releases the V2 paper for Tao's Law
🎮 Xiaomi's Wuhan youth apartment opens for move-in
☁️ The Wall Street Journal says tech giants' AI data centers use far more water than disclosed
🚗 Tesla delivered more than 480,000 vehicles globally in Q2
💰 Guangxiang Technology completes an angel round worth hundreds of millions of yuan
💡 Unitree's Chen Li says embodied intelligence needs two 80% thresholds for its ChatGPT moment
🚙 Ferrari launches a limited V12 manual model starting at 590,000 euros
🧠 Meta AI chief says the Watermelon model is aiming at GPT-5.5
🌐 Mistral open sources Leanstral 1.5, a theorem-proving model
📋 Europe plans tighter rules on loot boxes, limiting minors' access
🛍️ 7-Eleven sues Nike over an Air Max 95 colorway it says copies the convenience-store brand colors
⌚️ Casio releases a Pokemon 30th anniversary G-Shock with Pikachu hands and a Poke Ball case
Daily News
Foldable iPhone launch reportedly delayed

TF International Securities analyst Ming-Chi Kuo said yesterday on X that Apple's first foldable iPhone, rumored to be named iPhone Ultra, may follow the same release rhythm as the 2017 iPhone X: announced alongside the iPhone 18 Pro lineup, but with preorders and official availability pushed back to the fourth quarter.
Kuo wrote in his latest industry checks that because of manufacturing difficulty, foldable iPhone assembly shipments in the second half of this year are expected to be about 7 million to 8 million units. Third-quarter shipments may be only 500,000 to 1 million units, around 10% of the annual total.
On demand, Kuo said feedback from carriers, sales channels and resellers suggests that even with a starting price around $2,299 to $2,499, demand for the foldable iPhone should remain strong at least through the end of the year.
He expects the model could sell out quickly after preorders open, with delivery times stretching to four to six weeks or longer, and secondary-market premiums potentially reaching 50% to 100%.
Doubao and Qwen will shut down agent features

Doubao and Qwen have both recently announced that they will discontinue their agent features. A notice inside the Qwen app says Qwen's agent functions and services will officially go offline on July 15, 2026. After that, users will no longer be able to access related agent configurations or historical conversations, so the company recommends copying, taking screenshots of, or exporting conversations in advance.
Doubao issued a Doubao Agent Feature Shutdown Notice on the same day, saying that because of product adjustments, its agent feature will also go offline on July 15, 2026.
Notably, that shutdown date is the same day China's Interim Measures for the Management of AI Anthropomorphic Interactive Services, issued by the Cyberspace Administration of China and four other departments, will take effect.
The rules were published in April and bring under regulation ongoing emotional interaction services that use AI technology to simulate natural-person personality traits, thinking patterns and communication styles for the public in China. They cover emotional care, companionship and support delivered through text, images, audio, video and other formats.
iOS 27 beta code includes a B790 clue, possibly pointing to new AirPods with cameras

Developer Sam Henri Gold recently shared code he found in the second iOS 27 developer beta. It references an unreleased Apple product with the codename B790.
Gold found a dedicated configuration template for the B790 device inside a JSON file in the system path. The template describes a special image-input scenario.
The device is equipped with two cameras on both sides of the head. Each input sends left and right images at the same time; the images contain the same scene but have slight horizontal parallax.
The code also includes image-quality processing logic built specifically for that configuration, as well as instructions for calling visual-intelligence features.
Gold initially speculated that B790 might refer to Apple's rumored smart glasses, but he updated his post the next day and said the internal codename for AirPods Pro 3 is B788, so B790 is more likely to be a next-generation AirPods product rather than glasses.
Phones and PCs may get more expensive again as Samsung plans a third major DRAM price hike this year
According to ZDNet Korea, Samsung Electronics plans to sharply raise the average selling price of DRAM products in the third quarter, targeting an increase of up to 20%.
This would be Samsung's third price increase this year. Data shows Samsung's DRAM average selling price jumped about 90% quarter-on-quarter in the first quarter, then rose another 50% to 60% in the second quarter.
Although the third-quarter target of 20% is lower than the previous two quarters, it is still aggressive given the already high industry base. Low-power DRAM for servers and mobile devices is expected to rise by more than 20% because of supply bottlenecks.
Cost pressure across the supply chain is now moving downstream. Higher memory-chip prices continue to raise costs for smartphones, personal computers, automotive electronics and other sectors.
The report says that although the mobile divisions of Apple and Samsung have locked in part of their supply for third-quarter flagship products, models scheduled for the fourth quarter of 2026 and the first quarter of 2027 will face greater cost pressure.
Memory and SSD prices are changing by the day; dealers advise waiting unless the purchase is urgent
According to The Paper, PC dealers in several Shanghai electronics markets say component and complete-machine prices are already at extreme highs, and the upward trend is expected to continue for at least one year. Several merchants advised consumers not to buy unless they have an urgent need.
The immediate trigger for this round of increases was Apple's official price adjustment. On June 25, Apple raised prices for its iPad and Mac lines, citing the continued surge in memory and storage-chip costs. Starting prices for the two product lines increased by about 20% on average.
Lenovo also warned at ISC 2026 that DRAM and NAND flash prices have entered a structural upcycle. Even if major manufacturers continue expanding capacity, prices will be very hard to bring back to early-2025 levels. Lenovo expects this to become the new normal through 2030 and beyond.
Anthropic is reportedly talking with Samsung on an in-house AI chip to catch OpenAI
According to The Information, Anthropic has begun developing its own AI chip and has started preliminary talks with Samsung Electronics, seeking Samsung as a potential manufacturing partner.
Anthropic is still in the early stage of defining processor functions, performance targets and server-cluster integration plans. The report says Anthropic has recently started hiring chip engineers and brought in Clive Chan, an early member of OpenAI's in-house chip team.
Anthropic is considering Samsung's 2-nanometer manufacturing process and advanced packaging technology. Samsung also participated in Anthropic's $65 billion financing in May this year.
If the two sides reach a deal, it would be a major breakthrough for Samsung's foundry business and help it narrow the gap with TSMC in advanced manufacturing.
Huawei's He Tingbo releases the V2 paper for Tao's Law

ChinaXiv, the Chinese Academy of Sciences preprint platform, shows that Huawei semiconductor leader He Tingbo released the V2 version of Time Miniaturization Theory for Multi-Level Electronic Systems on July 3. Compared with the V1 version released on May 25, the new version adds engineering implementation details, measured quantitative data and a product-evolution roadmap.
The V2 paper organizes post-Moore scaling theory around the time constant tau and adds diagrams for a tau layered space-time model, LogicFolding architecture, bonding-interface cross sections, Unified Bus interconnect architecture and Hi-ONE optical engine.
The paper also explains the gear-ratio concept in LogicFolding: when hybrid-bonding pitch approaches top metal-routing dimensions, 3D design can move from discrete macro-block optimization to continuous cell-level optimization.
The new version also adds voltage, frequency, normalized power, area and power-density parameters for Kirin 2026 and the baseline Kirin9030 Pro, and further details mobile TSV movement, multi-active-layer stacking and Ascend accelerator iteration cadence.
Xiaomi's Wuhan youth apartment opens for move-in

Xiaomi announced that its Wuhan Xiaomi Youth Apartment has officially opened for move-in. It is the third youth-apartment project Xiaomi has launched after Beijing and Nanjing. The company says it has built 12 apartment buildings in total, with a construction area of 210,000 square meters and more than 4,200 units.
The Wuhan project has an average monthly rent of 1,099 yuan and is aimed at employee housing needs. Xiaomi says employees were invited to co-create the project from the early design stage, and that supporting services cover commuting, public spaces and daily living.
AI data-center water use far exceeds most tech giants' disclosed numbers

According to The Wall Street Journal, Microsoft, Google, Amazon and other technology giants are expected to spend $1 trillion on AI infrastructure in 2025 and this year, but the water consumption behind that buildout is far higher than official disclosures suggest.
Research shows that because the indirect water used in electricity generation is not included, the actual water consumption of U.S. data centers has historically been about 12 times their reported figures.
At present, among the major tech companies, only Meta includes indirect water use in its reporting. Its indirect water consumption in 2024 reached 19 billion gallons, more than 20 times its direct water use.
Google's direct water use last year was 10.9 billion gallons, up 34% year over year, but its indirect water use is estimated to be three times that figure. Although the giants buy renewable energy offsets, critics argue that this does not solve resource depletion in specific water-stressed regions.
Attracted by cheap land and electricity, roughly two-thirds of new U.S. data centers are being built in water-scarce areas such as Phoenix. By 2031, data centers' share of total water use in Phoenix could rise from 3% to more than 20%.
Although Nvidia and Microsoft are promoting closed-loop cooling technologies that do not require extra water replenishment, many existing water-intensive evaporative cooling systems are expensive to retrofit and will continue to face severe water-resource pressure.
Tesla delivered more than 480,000 vehicles globally in Q2

Tesla recently released its global production and delivery report for the second quarter of 2026. The data shows:
Tesla produced more than 451,000 pure electric vehicles globally in Q2, up about 10% year over year and about 25% quarter over quarter;
Global deliveries exceeded 480,000 vehicles, up about 25% year over year and about 34% quarter over quarter;
Tesla's energy-storage deployments reached 13.5 GWh, up about 41% year over year and about 53% quarter over quarter.
Data from the China Passenger Car Association shows that Tesla's Shanghai Gigafactory delivered more than 89,000 electric vehicles in June, up 24.4% year over year and a new high for the year. First-half deliveries were nearly 468,000 vehicles, up 28.4% year over year.
The Shanghai Gigafactory exported 36,000 electric vehicles in June, up 257% year over year, and continues to supply the mainland China market as well as exports to Asia-Pacific and Europe.
Guangxiang Technology completes an angel round worth hundreds of millions of yuan

According to 36Kr, embodied-intelligence company Guangxiang Technology announced that it has completed angel financing totaling hundreds of millions of yuan. The latest round included Zhuhai Technology Industry Group, Xingzheng Capital, Songhe Capital, Shunxi Fund, Muhua Kechuang, See Fund, Yichen Capital and Xingyun Technology, while existing investors 01VC and L2F Light Source Entrepreneurs Fund continued to invest.
Guangxiang Technology was founded in April 2025 and jointly incubated by Tsinghua University's School of Vehicle and Mobility and Institute for AI Industry Research. The new funds will be used to iterate its physics-native foundation model and promote commercial delivery of embodied-intelligence robots in industrial scenarios.
Unitree's Chen Li says embodied intelligence needs two 80% thresholds for its ChatGPT moment

According to Cailian Press, Unitree co-founder Chen Li recently said at the 12th Innovation Annual Meeting of the 2026 Yabuli China Entrepreneurs Forum that embodied-intelligence robots have not yet achieved broad deployment because embodied-intelligence foundation models are still immature.
Chen defined the ChatGPT moment for embodied intelligence as two 80% targets: in 80% of unfamiliar scenarios, robots should be able to complete about 80% of tasks through voice or text instructions.
He expects the goal of autonomously solving 80% of tasks may be achievable in vertical To B scenarios. In To C scenarios, where generalization requirements are higher, maturity may still take 8 to 10 years or even longer.
For the next two to five years, Chen proposed three key factors for the industry:
A unified end-to-end robotics foundation model;
Lower-cost, longer-life hardware;
Low-cost large-scale computing power.
He noted that because on-device compute is constrained by battery life, heat dissipation and cost, while cloud compute has latency issues, future compute may take a distributed form.
Chen remains optimistic about the industry's long-term prospects. He predicted that starting this year, embodied-intelligence robots will lead a new wave of consumption and significantly improve social productivity and labor upgrading.
Ferrari launches a limited V12 manual model starting at 590,000 euros

Ferrari recently introduced the 12Cilindri Manuale, a limited-edition V12 model equipped with a manual transmission. The model is intended to attract fans who value traditional internal-combustion sound and a pure manual-driving experience, and it also serves as an important supplement after the brand's first electric model, Luce, received lukewarm market feedback.
The 12Cilindri Manuale is a special-series version of the 2024 12Cilindri and is limited to 1,499 units worldwide. It introduces a new Manuale by-wire system, designed to electronically simulate the mechanical feel and pedal feedback of a traditional manual transmission while retaining the efficiency advantages of a modern dual-clutch gearbox.
The car is powered by a 6.5-liter naturally aspirated V12 engine with a maximum output of 830 horsepower.
The 12Cilindri Manuale starts at 590,000 euros in Italy, or about $672,000. The first deliveries are expected to begin in the first quarter of next year.
Meta AI chief says the Watermelon model is aiming at GPT-5.5

According to Business Insider, Alexandr Wang, head of Meta Superintelligence Labs, said at an internal all-hands meeting on July 3 that Meta's next-generation AI model, codenamed Watermelon, has already matched OpenAI's flagship GPT-5.5 on mainstream AI model benchmarks.
Wang said Watermelon is the model after Avocado and is currently in training. The amount of compute used by Watermelon is an order of magnitude larger than Avocado.
Avocado is the internal codename for Muse Spark, the first model series released by Meta Superintelligence Labs in April, which began powering Meta AI on the same day.
At the same time, Wang posted publicly on X that an upgraded version of Muse Spark will be released soon, with significant improvements in coding and AI-agent capabilities to narrow the gap with other leading models.
Mistral open sources Leanstral 1.5, a theorem-proving model

Mistral AI announced Leanstral 1.5, an open-source model for Lean 4 formal proof. The model has 119 billion total parameters, about 6.5 billion active parameters, and supports a context length of up to 256,000 tokens.
Official data shows that Leanstral 1.5 set new records on multiple mathematical-logic benchmarks and demonstrated strong reasoning ability:
It achieved 100% full coverage on miniF2F;
It solved 587 of PutnamBench's 672 difficult problems;
It reached 87% and 34% accuracy on the abstract-algebra benchmarks FATE-H and FATE-X, respectively.
In cost control and efficiency, Leanstral 1.5 averaged about $4 per problem on PutnamBench, far below the hundreds of dollars in inference cost seen in some earlier systems.
Europe plans tighter rules on loot boxes, limiting minors' access

According to Bloomberg, a coming set of new European rules for video games may restrict minors' access to certain games and paid features, creating a major impact on in-game purchase revenue.
The core target of the regulation is the random-draw mechanism known as loot boxes. Regulators believe these paid boxes, which randomly grant digital items, have gambling-like characteristics and can encourage minors to become addicted.
In June this year, Pan European Game Information, or PEGI, began requiring games containing loot boxes to be rated as unsuitable for children under 16. EU regulators are also considering the Digital Fairness Act, which is planned for passage next year and may fully ban minors from accessing games that contain loot boxes.
7-Eleven sues Nike over an Air Max 95 colorway it says copies the convenience-store brand colors

According to Reuters, convenience-store chain 7-Eleven filed a lawsuit against Nike in federal court in Texas on July 2, accusing Nike's upcoming Air Max 95 sneaker of using an orange, green and red color scheme that is highly similar to 7-Eleven's trademarks and allegedly infringes its brand rights.
The complaint says the colorway is highly similar to 7-Eleven's iconic three-color stripe design and is likely to confuse consumers into associating the shoe with the 7-Eleven brand.
7-Eleven also pointed out that Nike scheduled the shoe's release for July 11, which happens to be 7-Eleven's annual brand promotion day. The complaint uses forceful language, accusing Nike of showing callous and malicious disregard for 7-Eleven's rights.
7-Eleven said it had tried several times to negotiate with Nike before filing the lawsuit, but Nike made clear that it would continue promoting the shoe and maintain the planned July 11 release date.
Casio releases a Pokemon 30th anniversary G-Shock with Pikachu hands and a Poke Ball case

To celebrate the 30th anniversary of the Pokemon series, Casio's G-Shock brand has officially launched the GA-110PKM-7A collaboration watch. The product draws inspiration from the iconic colors of the first Pokemon games released in 1996 and adds many tribute details.
The watch uses a translucent resin case with red, blue and green accents. Design details include a Poke Ball subdial at the 9 o'clock position, a Pikachu-silhouette hand and a special strap printed with 30 Pokemon.
It also comes with a Poke Ball-shaped storage case, has a 30th anniversary logo engraved on the case back, and retains core features such as shock resistance and 200-meter water resistance.
Fast & Furious finale Fast Forever starts filming

According to Variety, Vin Diesel posted behind-the-scenes photos from set on social media, confirming that the 11th Fast & Furious film and the final installment of the series, Fast Forever, has entered production. The film is scheduled to open in North America on March 17, 2028.
Since the first film in 2001, the Fast & Furious franchise has evolved from a street-racing movie into a global action blockbuster IP. Diesel described the new film as the series' final ride, and the title directly points to the closing chapter of the Fast brand. Next, attention will focus on returning original characters, the scale of action scenes and how the series resolves long-running character relationships.
Midjourney asks Disney and other studios to disclose their own AI use

According to TechCrunch, in the copyright lawsuit brought by Disney, Universal and Warner Bros. against Midjourney, Midjourney is asking the court to force the studios to disclose documents about their own use of generative AI.
The studios previously alleged that Midjourney's image model can generate images of copyrighted characters such as Bart Simpson and Darth Vader.
The dispute centers on the scope of discovery. The judge previously allowed the studios to submit only information about their use of generative AI for consumer-facing video and images. Midjourney's latest filing asks the court to overturn that limit, arguing that the studios should not be allowed to cherry-pick documents that support their market-harm theory while withholding materials that may support Midjourney's defenses.
Midjourney argues that if studios also use image-generation models internally for storyboarding, visual ideation or film and TV content development, that would show such training and use patterns have already become industry practice.

Editorial team
Product Team @ WebCal
The official product team behind WebCal. We build high-performance computing infrastructure and decentralized cloud solutions.



