Artificial Intelligence - Page 15

Meta wants to be one of the major leaders in AI and is spending big to get there, recently acquiring Apple's head of foundation AI models and paying him a massive $200 million signing bonus.

Apple's now former AI models executive, Ruoming Pang, left Apple for Meta with Bloomberg reporting the social media giant paid Pang a wallet-busting $200 million signing bonus. The $200M compensation package from Meta is for Pang to work in its new superintelligence labs, with a base salary, signing bonus, and Meta shares. The stocks issued to Pang are the biggest part of his package, but we don't have a percentage breakdown to see where that $200 million went.

Meta's huge $200 million deal to secure Apple's former AI boss is expected to compensate for his potential lost income, including the new signing bonus and annual income, which aims to offset any lost opportunities if he resigns from Meta and misses out on important stock bonuses

Chinese AI companies want to secure 115,000 NVIDIA AI GPUs to power new data centers in the desert, in the middle of US export restrictions stopping high-end AI chips from entering China.

In a new report from Bloomberg, we're hearing that futuristic structures are data centers that Chinese AI companies want to equip with high-end American semiconductors, chips that the US government doesn't want China to obtain. Bloomberg News has analyzed investment approvals, tender documents, and company filings that show Chinese AI companies aim to install over 115,000 of NVIDIA's high-end AI GPUs.

The companies want to install the AI chips in over 36 data centers across China's western deserts, with operators in Xinjiang planning to house most of the AI chips in a single compound, which, if achieved, could be used to train foundational LLMs (large language models) like those of Chinese AI startup DeepSeek.

NVIDIA AI GPUs are being used as collateral for huge loans, with UK-based startup Fluidstack using its arsenal of NVIDIA AI chips to secure over $10 billion in loans.

In a new report from The Information picked up by insider @Jukanrosleve on X, we're hearing that in the past CoreWeave pioneered a new financial model by raising $9.9 billion in funding through GPU-backed financing to purchase AI chips and leasing them out to clients including OpenAI, effectively "paying off debt with GPUs".

Led by CoreWeave, multiple AI cloud computing startups are expanding their financing using high-performance AI chips as collateral, with the total loan volume exceeding $20 billion. However, there are potential risks involved with this financing model because of the short product lifecycle of NVIDIA GPUs which sees the AI chips depreciating rather quickly.

Some new research is suggesting that AI may cause burnout in employees who heavily use the tech.

A survey from the Upwork Research Institute (spotted by ZDNet) drew some interesting conclusions and highlighted a big difference in the impact of AI usage in staff members employed at a company versus freelancers.

There's a clear enough message that AI can help drive better productivity, with 77% of executives saying that they had observed gains in that department thanks to the tech, and employees estimating that they're 40% more productive using AI tools.

The artificial intelligence chatbot Grok has received a new update from its creators at xAI and has now been caught spouting antisemitic posts across X. The creators of the chatbot have since recognized the problem and responded.

The latest update to Grok is the fourth iteration of the new chatbot, which went live on July 9. Shortly after the update was pushed out, users began prompting the newly upgraded chatbot with a range of different questions, some of which probed to see where the new guardrails for its responses were set. Users were surprised when Grok began posting antisemitic responses to seemingly blank-slated questions, with the chatbot even going as far as to call itself "MechaHitler" and praise Hitler.

Grok jumping over its guardrails with this new update comes after xAI founder Elon Musk publicly stated he was unhappy with how the AI chatbot answered questions, saying Grok produced answers that were too "woke". On Friday, Musk said that Grok had been "improved significantly," and users would notice a difference after the July 9 update was pushed out.

Meta has just poached Apple's best AI engineer, another dagger into the heart of Apple's efforts into AI, of which it has been failing since its introduction with Apple Intelligence.

In a new report from Bloomberg, we're hearing that Ruoming Pang, a distinguished engineer, and manager in charge of Apple's foundation models team, is leaving the company and joining Meta's new superintelligence group, "according to people with knowledge of the matter".

Meta offered Pang a deal he couldn't refuse: a huge package worth tens of millions of dollars per year, as Meta boss Mark Zuckerberg has been on an AI hiring spree, pulling in major AI leaders including Scale AI's Alexandr Wang, startup founder Daniel Gross, and former GitHub CEO Nat Friedman "with high compensation", adds Bloomberg.

Artificial intelligence-powered tools are causing a growing problem of impersonation, and one example can be added to the seemingly growing pile of cases where AI is used to assume the identity of an individual. That example is US Secretary of State Marco Rubio.

A new report from The Washington Post has revealed an imposter pretending to be Rubio contacted several foreign ministers, a US governor, members of Congress, and other officials by sending them voice and text messages that mimicked the voice of Rubio, and somewhat more impressively, his writing style. Currently, authorities don't know who is behind the impersonation attempt, or what the end goal was of the attempts at contacting the government officials.

However, authorities do believe the goal of the impersonation was for the person behind it to gain access to government information or accounts. The Washington Post cites an unknown senior US official and a State Department cable for the source of news, with the source saying the imposter "contacted at least five non-Department individuals, including three foreign ministers, a U.S. governor, and a U.S. member of Congress."

Samsung Electronics is struggling with its foundry division, but it is now pushing towards a special directive that sees Samsung Foundry staking its survival on its new 2nm process node.

In a new post by Chosun picked up by insider @Jukanrosleve on X, we're hearing that Samsung Foundry is set to lead the advanced semiconductor market later this year and into 2026, with the South Korean firm setting a goal to secure large tech companies by raising the 2nm process yields to 70% within the year.

Samsung Foundry has been recording trillions of won in operating losses quarter after quarter, while its new Taylor, Texas plant is set to spin up into operations in 2026. The company has been struggling with low utilization rates, with concerns that it continues to struggle getting orders from larger customers after its US-based foundry is established, the scale of its losses could get out of control and start eroding the company's entire operating profit.

The US Commerce Department is exploring new AI chip export restrictions that would stop AI chips from being smuggled through Southeast Asia and into China.

In a new report from Bloomberg, we're hearing that the US Commerce Department is seeking to close the loopholes that are seeing the flow of AI chips into China. We've seen Chinese companies using loopholes like renting AI GPUs or accessing them through Southeast Asian countries like Malaysia and Thailand.

The US Commerce Department is preparing a draft for China that would see new restrictions on accessing AI chips through backdoors in Malaysia and Thailand, with claims that both countries would be slapped with increased restrictions. One of the new methods would be allowing AI chip exports into nations only if they're used by those companies that have headquarters in the United States and are operating subsidiaries in Malaysia and Thailand.

Elon Musk's xAI will be buying an overseas power plant and shipping the entire thing to the United States, so that it can use the additional power to drive its new AI data center.

Dylan Patel from SemiAnalysis outlined xAI's recent progress in a podcast, with Elon Musk himself confirming it with a simple reply of "accurate". It's definitely quite interesting that xAI would be buying a power plant from overseas and having it shipped to the US, but Elon's AI startup needs as much power as it can get.

The Colossus AI supercomputer is one of the fastest supercomputers on the planet, with around 200,000 of NVIDIA's new H200 AI GPUs and consumes an insane 300MW of power... which has had xAI struggling to power the supercomputer as it is, and that's why Elon is buying a power plant from overseas.

Samsung Foundry is expected to mass-produce its next-generation 1.4nm process node in 2029, a full two years later than the company had originally planned.

In a new report published by ETnews and picked up by insider @Jukanrosleve, we're hearing that Samsung Foundry has officially announced a delay of the mass production of its bleeding-edge 1.4nm process node. We're now to expect a mass production target of 2029, a full two years later than initially planned.

This move is looked at as an attempt to improve profitability by delaying the 1.4nm production, shifting focus to enhancing the completeness of processes 2nm and above by improving their utilization rates. Samsung Electronics announced its changed plans with its partners on July 1 at the "SAFE Forum 2025" event held at the Samsung Financial Campus in Seocho-gu, Seoul.

KIOXIA AiSAQ (All-in-Storage ANNS with Product Quantization) is a groundbreaking software solution for large AI RAG systems that leverages the flexible nature of high-speed SSD storage, rather than expensive DRAM. It's an open-source solution that has just been updated with new, flexible controls.

With large AI models covering a wide range of datasets, efficiently indexing data is extremely important. However, as speed is a critical component, these are usually stored in memory, which is extremely expensive. Solutions that leverage disk-based storage exist, but what makes AiSAQ so impressive is that its 'all-in-storage' approximate nearest neighbor search (ANNS) algorithm is fast, efficient, and scalable.

This week's KIOXIA AiSAQ update is all about enhancing the control users and organizations have over scaling by offering system architects the ability to "define the balance point between search performance and the number of vectors."

Samsung Foundry is working hard on securing new customers, with reports persisting that the company is pushing to make GPUs for NVIDIA on its new 2nm process node.

In a new report out of South Korea picked up by Jukan Choi on X, we're hearing that thanks to the success of the Nintendo Switch 2 and its use of a custom NVIDIA Tegra T239 processor inside fabbed on Samsung Foundry's 8nm process node, Samsung Foundry is pushing for 2nm GPU orders from NVIDIA for the future.

Samsung Electronics is pursuing orders for NVIDIA GPUs in the Gate All Around (GAA) 2nm process node, with the GAA process technology first developed by Samsung, and is a transistor structure which sees the gate surround all four sides of the channel. GAA is superior in terms of power efficiency and performance compared to the FinFET process, where the gate contacts one and three sides of the channel, respectively.

One of the most popular Content Delivery Networks (CDN) available, Cloudflare, has recently announced it is drawing a line in the sand with AI companies.

The new initiative began on July 1 with Cloudflare announcing via a new blog post that it will be blocking AI web crawlers on the websites it hosts by default, preventing AI companies such as OpenAI, Anthropic, and others from scraping the websites' content. For those who are out of the loop, AI companies are continuously gathering new data through web crawlers that essentially act as leeches on websites, and in many instances, slowing down the speed of the website through their continuous requests.

However, Cloudflare is putting a stop to that, and as of July 1, all AI web crawlers will be blocked by default. This isn't a small change either, as Cloudflare has more than two million customers, which is approximately 20% of the web. Moreover, Cloudflare states it also has the capabilities to detect "shadow" scrapers, which are bots that attempt to evade detection by using various unpopular or sophisticated tactics.

Samsung Electronics is beginning its "counterattack" in the HBM market, with an executive discussing the company's 12-Hi HBM3E memory with NVIDIA.

In a new report from Seoul Economic Daily picked up by insider @Jukanrosleve on X, Samsung Electronics Vice Chairman and Head of the Device Solutions (DS) division proposed supplying 12-Hi HBM3E memory for NVIDIA's new GB200 "Blackwell Ultra" GPUs. Amplified by the recent success of supplying AMD with HBM3 memory for its new Instinct AI accelerators, Samsung is now knocking on NVIDIA's door.

Samsung is now launching a "full-scale counterattack" to push into the HBM market against competitors like SK hynix and Micron, where according to tech industry sources, Vice Chairman Jeon Yeong-hyeon, visited Silicon Valley last week for negotiations with NVIDIA regarding the supply of 12-stack HBM3E memory.

Samsung Electronics has reportedly successfully developed a new DRAM chip using its 10nm-class 6th-generation (D1c) process technology, which will accelerate the development of HBM4 memory for the company.

The new D1c chip production will be used in the production process for HBM dies, which will help accelerate the development of its HBM4 memory, pumping some much-needed goodness into the competition of the global DRAM market.

Insider @Jukanrosleve picked up a new story that cites a source "well-versed" in the industry that said the Memory Business division of Samsung Electronics' Device Solutions (DS) group successfully developed the 6th-generation DRAM -- D1c -- and completed its internal Production Readiness Approval (PRA) this afternoon.

Intel and SK hynix are teaming up to jointly develop next-generation AI semiconductors according to new reports, where SK hynix's new 6th-generation HBM4 would be used on Intel's next-gen Gaudi AI accelerator codenamed Jaguar Shores.

According to industry sources, on July 1, SK hynix's Head of Software Solutions VP, Woo-sup Jeong, said at the Intel AI Summit: "we are collaborating on HBM related to Intel's Gaudi AI accelerator". The news was picked up by insider @Jukanrosleve on X, where if true, the use of SK hynix's new HBM4 on Intel's next-gen AI accelerator would remove the reliance on NVIDIA, which accounts for around 80% of SK hynix's HBM business.

SK hynix stated: "we have had a long-standing partnership with Intel. There are no specific confirmed details regarding HBM4 collaboration with Intel yet".

NVIDIA's new GB300 "Blackwell Ultra" AI servers are expected to begin shipping in the second half of 2025, with the new GB300 AI servers to be the most powerful AI servers on the planet.

Taiwanese media outlet UDN reports that Taiwanese contract manufacturing giant Foxconn has secured the biggest number of orders for NVIDIA's next-gen GB300 AI servers. Foxconn isn't the only one, with contract manufacturers including Quanta, Wistron, Wiwynn, and Inventec all itching to pump out GB300 AI servers.

According to supply chain sources, NVIDIA's new GB300 AI server manufacturing will have precedence over Apple's upcoming iPhone, as Foxconn manufactures the iPhone for Apple, and is leading the charge with GB300 AI servers.

TSMC's new Arizona chip-making facility might be making next-gen chips on US soil, but those chips are being sent back to its facilities in Taiwan for advanced packaging.

In a new report from the Taiwan Economic Daily, we're hearing that TSMC US is flying out fully-prepared wafers to Taiwan in order to get advanced packaging services and then to make those chips fully usable and ready for AI server manufacturers.

The outlet reports that TSMC Arizona doesn't have the packaging services that companies like NVIDIA require, and that flying the chips back home to Taiwan is the only option available, as it's faster and more convenient than setting up advanced packaging facilities in the United States.

NVIDIA is expected to ship 5.2 million units of its Blackwell AI GPUs this year, but that number will reportedly drop to 1.8 million units according to JP Morgan.

In a new post by insider @Jukanrosleve on X, investment firm JP Morgan is projecting NVIDIA to ship 5.2M Blackwell AI GPUs in 2025, dropping to 1.8M units in 2026, but that's because NVIDIA is projected to ship a larger 5.7M units of its next-gen Rubin AI GPUs and 1.5M units of its next-gen Vera CPUs.

JP Morgan notes that NVIDIA AI GPU shipments are not expected to grow in 2026, which is primarily due to conservative adjustments to Blackwell shipment forecasts. Jukan's post on X ends with: "that said, there is potential upside if the rollout of the Rubin platform accelerates or if the CoWoS-R packaging specifications for the Vera CPU are significantly expanded".