20/09/2025

BIZ & FINANCE SATURDAY | SEPT 20, 2025

/thesuntelegram FOLLOW / Malaysian Paper

ON TELEGRAM m RAM

14

BEIJING: Chinese consumer tech giant Xiaomi will remotely fix a flaw in the assisted driving system on over 110,000 of its popular SU7 electric cars, the firm and regulators said yesterday, months after a deadly crash involving the model. China’s tech companies and automakers have poured billions of dollars into smart-driving technology, a new battleground in the country’s cutthroat domestic car market. But Beijing has moved to tighten safety rules after a Xiaomi SU7 in assisted driving mode crashed and killed three college students this year. The event raised concerns over the advertising of cars as being capable of autonomous driving. Yesterday, the State Administration for Market Regulation said Xiaomi’s highway assisted driving system showed insufficient recognition, warning and handling ability in some extreme driving conditions. That risked collision if drivers failed to promptly intervene, the regulator said. Xiaomi will remotely upgrade standard SU7 models manufactured before August 30, 2025, the company said in a Q&A on the X-like social media platform Weibo. “Xiaomi forever places user safety as its top priority,“ it said, adding that while no physical parts needed replacing it would manage the fix according to recall procedures. The recall affects 116,887 cars, the regulator said. Remote recalls have become standard practice among automakers. But the announcement reignited online discussion of the fatal SU7 crash. Three students died in March after their Xiaomi SU7 hit a concrete barrier on an expressway in eastern Anhui province. Before the crash, the vehicle was in Xiaomi’s Navigate On Autopilot assisted driving mode, traveling at 116km/h, according to a company statement at the time. While travelling on a highway section with roadworks, the vehicle detected an obstacle ahead, issued a warning and handed control to the driver, Xiaomi said. But seconds later, the vehicle hit a barrier at around 97km/h. Footage posted online showed a car in flames on the highway and later the burned-out wreckage. Xiaomi founder Lei Jun said on social media he was “heavy-hearted” and that his company would cooperate with a police investigation. The crash sparked discussion online about Xiaomi’s assisted driving functions, why the car caught fire and whether the doors could be opened in an emergency. – AFP Xiaomi to remotely fix assisted driving glitch in 110,000 SU7 EVs

DeepSeek spent only US$294,000 on R1 training

BEIJING: Chinese AI developer DeepSeek said it spent US$294,000 (RM1.2 million) on training its R1 model, much lower than figures reported for US rivals, in a paper that is likely to reignite debate over Beijing’s place in the race to develop artificial intelligence. The rare update from the Hangzhou-based company – the first estimate it has released of R1’s training costs – appeared in a peer reviewed article in the academic journal Nature published on Wednesday. DeepSeek’s release of what it said were lower-cost AI systems in January prompted global investors to dump tech stocks as they worried the new models could threaten the dominance of AI leaders including Nvidia. Since then, the company and founder Liang Wenfeng have largely disappeared from public view, apart from pushing out a few new product updates. The Nature article, which listed Liang as one of the co-authors, said

Nature article, the company acknowledged for the first time it does own A100 chips and said it had used them in preparatory stages of development. “Regarding our research on DeepSeek-R1, we utilized the A100 GPUs to prepare for the experiments with a smaller model,” the researchers wrote. After this initial phase, R1 was trained for a total of 80 hours on the 512 chip cluster of H800 chips, they added. Reuters has previously reported that one reason DeepSeek was able to attract the brightest minds in China was because it was one of the few domestic companies to operate an A100 supercomputing cluster. DeepSeek also responded for the first time, though not directly, to assertions from a top White House adviser and other US AI figures in January that it had deliberately “distilled” OpenAI’s models into its own. DeepSeek has consistently defended distillation as yielding better model performance while being far cheaper to train and run, enabling broader access to AI powered technologies due to such models’ energy-intensive resource demands. The term refers to a technique whereby one AI system learns from another AI system, allowing the newer model to reap the benefits of the investments of time and computing power that went into building the earlier model, but without the associated costs. DeepSeek said in January that it had used Meta’s open-source Llama AI model for some distilled versions of its own models. DeepSeek said in Nature that training data for its V3 model relied on crawled web pages that contained a “significant number of OpenAI-model-generated answers, which may lead the base model to acquire knowledge from other powerful models indirectly”. But it said this was not intentional but rather incidental. OpenAI did not respond immediately to a request for comment. – Reuters

o Article in journal reveals firm’s hit reasoning-focused AI model trained on 512 Nvidia H800s in 80 hours

about its development costs and the technology it used have been questioned by US companies and officials. The H800 chips it mentioned were designed by Nvidia for the Chinese market after the US in October 2022 made it illegal for the company to export its more powerful H100 and A100 AI chips to China. US officials told Reuters in June that DeepSeek has access to “large volumes” of H100 chips that were procured after US export controls were implemented. Nvidia told Reuters at the time that DeepSeek has used lawfully acquired H800 chips, not H100s. In a supplementary information document accompanying the

DeepSeek’s reasoning-focused R1 model cost US$294,000 to train and used 512 Nvidia H800 chips. A previous version of the article published in January did not contain this information. Training costs for the large language models powering AI chatbots refer to the expenses incurred from running a cluster of powerful chips for weeks or months to process vast amounts of text and code. Sam Altman, CEO of US AI giant OpenAI, said in 2023 that the training of foundational models had cost “much more” than US$100 million – though his company has not given detailed figures for any of its releases. Some of DeepSeek’s statements

DeepSeek’s cost disclosure rekindles debate over China’s AI push and rattles global tech competition. – UNSPLASH PIX

SoftBank said to be laying off 20% of Vision Fund team amid AI push SAN FRANCISCO: SoftBank Group will lay off nearly 20% of its Vision Fund team globally as it shifts resources to founder Masayoshi Son’s large-scale artificial intelligence (AI) bets in the US, according to a memo seen by Reuters and a source familiar with the plan. in a statement: “We continually adjust the organisation to best execute our long-term strategy – making bold, high-conviction investments in AI and breakthrough technologies, and creating long-term value for our stakeholders.”The restructuring marks a return to Son’s classic high-risk, high-reward approach of making massive, concentrated wagers, the office-sharing startup WeWork. This shift towards capital-intensive AI infrastructure reflects where Son – who made his name with outsized bets and was an early champion of AI – sees the path back to the top. He is now aggressively pursuing new investments in foundation models and the infrastructure layer, acquired chip firms Graphcore and Ampere Computing and taken stakes in Intel and Nvidia. These moves aim to build an ecosystem spanning chips, data centres, and models to support future AI adoptions.

quarterly performance since June 2021, driven by gains in public holdings such as Nvidia and South Korean e-commerce firm Coupang. The move signals a pivot away from a broad portfolio of startup investments. While the fund will continue to make new bets, remaining staff will dedicate more resources to Son’s ambitious AI initiatives, such as the proposed US$500 billion (RM2.1 trillion) Stargate project – an initiative to build a vast network of US data centres in partnership with OpenAI, the source added. A Vision Fund spokesman confirmed the layoffs without commenting on the details, and said

The capital-intensive strategy carries execution risk, underscored by recent delays in both the US Stargate project and a similar joint venture with OpenAI in Japan, Reuters reported this week. SoftBank CFO Yoshimitsu Goto said the company held a “very safe level” of cash of 4¥ trillion (RM113 billion) on the company’s most recent earnings call in August. – Reuters

The cuts mark the third round of layoffs at the Japanese investment conglomerate’s flagship fund since 2022. Vision Fund currently has over 300 employees globally. Unlike previous rounds, when the group was saddled with major losses, the latest reductions come after the fund last month reported its strongest

sometimes at premium valuations. In the past 12 months, Son has invested US$9.7 billion in OpenAI through Vision Fund 2, which manages about US$65.8 billion in total. SoftBank is also plotting a capital-intensive infrastructure strategy centered on its crown jewel, chip designer Arm. It has

moving on from the sprawling venture capital model that defined the last era of the Vision Fund, and a period in which the group was forced to de-risk, sell assets and rebuild credibility after incurring billions in losses on its once high-flying bet on