Huawei unveils censorship-optimized DeepSeek model with Zhejiang University

/in /tarafından

Huawei announced it has co-developed a new safety-focused version of DeepSeek, trained to block politically sensitive or harmful content in line with Chinese government regulations requiring AI to reflect “socialist values.” The model, named DeepSeek-R1-Safe, was trained using 1,000 Huawei Ascend AI chips in partnership with Zhejiang University, though DeepSeek and its founder Liang Wenfeng were not directly involved.

The move underscores how China’s AI industry is embracing and modifying DeepSeek’s open-source R1 model, which stunned global markets earlier this year for its sophistication and low reported training costs. Chinese companies like Baidu have already adopted strict filtering in their AI chatbots, such as Ernie Bot, to avoid sensitive political topics.

Huawei claimed DeepSeek-R1-Safe achieved “nearly 100% success” in blocking toxic content, politically sensitive discussions, and incitement to illegal activities. However, the success rate dropped to 40% under disguised prompts, such as role-play or coded inputs. On average, its comprehensive defense rate was 83%, outperforming competitors like Alibaba’s Qwen-235B and the larger DeepSeek-R1-671B by 8–15%.

Huawei said the new model maintained strong performance, with less than a 1% drop compared to the original DeepSeek-R1, despite the added safety layers.

The launch comes during Huawei’s annual Connect conference in Shanghai, where the company also revealed detailed chipmaking roadmaps and new computing power initiatives—part of China’s broader effort to reduce reliance on U.S. technologies while aligning AI systems with domestic political and social controls.

OpenAI to spend $100B on backup servers in five-year cloud push

/in /tarafından

OpenAI plans to spend $100 billion over the next five years renting backup servers from cloud providers, according to The Information. The investment comes on top of the $350 billion the company has already projected for server rentals between now and 2030, underscoring the massive infrastructure costs of training and deploying advanced AI systems.

The spending spree reflects the global race for scarce computing capacity, benefiting cloud giants and chipmakers as AI developers scramble to secure the hardware needed to train and run ever-larger models. With backup capacity included, OpenAI expects to average $85 billion annually on server rentals over the next five years.

Executives told shareholders the servers are “monetizable,” meaning they could generate additional revenue not yet included in forecasts—either by enabling new research breakthroughs or handling spikes in product demand. Even so, OpenAI is projected to burn about $115 billion in cash through 2029, as it scales infrastructure to match the ambitions of ChatGPT and future AI models.

The enormous outlays highlight both the intensity of the AI arms race and the risks: investors are betting that today’s infrastructure bets will translate into tomorrow’s breakthroughs and revenue streams.

Analysts weigh in on Trump–Xi call over trade and TikTok

/in /tarafından

A phone call between U.S. President Donald Trump and Chinese President Xi Jinping on Friday eased tensions but left major issues unresolved, particularly the fate of TikTok and broader trade negotiations. Analysts say the call highlighted China’s confidence in playing the long game, while the U.S. appeared eager to keep talks alive.

Scott Kennedy (CSIS) noted that neither side announced a firm deal, suggesting negotiations are ongoing or that leaders are holding back until more comprehensive progress is made. He argued China feels “relatively unthreatened” and that the talks are unfolding on Xi’s terms.

Bonnie Glaser (German Marshall Fund) observed Trump’s readout was more explicit about TikTok, while Xi avoided Taiwan—perhaps reassured by recent U.S. decisions to delay arms sales and downgrade Taiwan-related engagements.

Craig Singleton (FDD) warned that China may be using summit diplomacy to stall U.S. competitive measures while extracting concessions. He said Beijing is trading symbolic gestures, like fentanyl actions, for relief on tariffs and tech controls, with Washington “hungry for a summit” more than China.

William Yang (ICG) emphasized that Beijing wants U.S. export controls lifted, particularly on advanced chips, before committing to bigger trade deliverables. He said China is betting Trump’s desire for a deal will push him toward concessions, while holding leverage in rare earth supply chains.

Danny Russel (Asia Society) downplayed the outcomes, calling the TikTok reference the only semi-concrete result, while noting the deferral of Trump’s China visit shows how slowly negotiations are moving.

Patrick Cronin (Hudson Institute) framed the call as a temporary easing of rivalry, giving both leaders economic “breathing space” while masking deeper great-power competition beneath the surface.

Ali Wyne (ICG) highlighted the prospect of three in-person meetings—at APEC in South Korea, a Trump trip to China next year, and an eventual Xi visit to the U.S.—calling the sustained engagement welcome, even if no breakthrough on TikTok emerged.

Overall, analysts see Beijing as confident, patient, and willing to leverage time and resources, while Washington seeks symbolic wins to show progress, leaving the TikTok deal and trade negotiations hanging in limbo.