微软第四财季因投资 Anthropic 录得 32 亿美元收益
Related
Foreign media: Anthropic Mythos's threat prompts DeepSeek to launch its first major funding round.
According to Beating's monitoring, The Information, citing sources familiar with the matter, reported that DeepSeek has decided to abandon its founder-funded model and launch its first round of financing exceeding 50 billion yuan (approximately 7.4 billion US dollars). The immediate trigger was Anthropic's demonstration of the Mythos security model in April. Mythos possesses the ability to find and exploit software vulnerabilities. DeepSeek's R&D team assessed that, facing models requiring massive computing power and data stacking, and even those with national security-level destructive capabilities, relying solely on Liang Wenfeng's personal wealth would not be enough to compete in the computing power arms race; external funding was necessary to expand the computing power pool. Last year, in response to US export controls, DeepSeek attempted to migrate its training and deployment system from Nvidia's CUDA ecosystem to Huawei chips, rewriting the underlying software and causing a 15-month delay in the release of its next-generation model. DeepSeek has historically maintained an extremely flat organizational structure, without human resources or public relations departments; everyone reports directly to Liang Wenfeng. With the completion of the massive financing, the team is now recruiting for legal, finance, and human resources positions.
By abandoning Anthropic and switching to DeepSeek v4, startups saved millions of dollars while experiencing performance improvements.
According to Beating, Flo Crivello, founder of Lindy, iMessage's native AI assistant, announced that all user-side business traffic has been switched from Anthropic to the open-source model DeepSeek v4. This switch not only saved the company millions of dollars but also improved performance in many core business scenarios. To run DeepSeek v4, the team developed a large amount of infrastructure and internal tools at the underlying level, with the actual workload reaching 100 times the expected amount. Regarding server vendor selection, Lindy evaluated all major players but surprisingly chose the relatively less well-known cloud computing provider Atlas Cloud. Although the user-side has been completely switched out, Lindy's internal employees still use the Anthropic model for programming and office work, mainly due to the extremely high discounts offered by large customer contracts. The system also retains Claude Opus as a backup plan, automatically invoked when DeepSeek v4 fails in complex tasks. Crivello believes that the low-price strategy of large Chinese models will put long-term profit pressure on American counterparts, but Anthropic's corporate relationships and brand advantages remain a competitive moat.
TeraWulf secures a massive 20-year lease on Anthropic, potentially generating $19 billion in revenue.
ChainCatcher reports that Anthropic has announced a 20-year data center lease with TeraWulf to build AI computing infrastructure in Kentucky, USA. Located approximately an hour's drive southwest of Louisville, the data center has a planned capacity of about 400 megawatts, with initial power supply expected in the second half of 2027. Under the agreement, the initial lease is expected to generate approximately $19 billion in revenue for TeraWulf. TeraWulf is a Bitcoin mining company that has been continuously transforming into an AI data center infrastructure provider in recent years, with its stock price rising by more than 80% this year, reflecting the continued pricing power for AI computing infrastructure in the market.
Samsung's foundry business is expected to secure orders from Meta and Anthropic, and is projected to return to profitability in the fourth quarter.
According to ChainCatcher, South Korean media reports that Samsung Electronics' foundry business is accelerating its expansion in the global AI chip market. Following its AI chip order from Tesla last year, Samsung is currently advancing its custom ASIC production collaboration with Meta and Anthropic. Meta is reportedly in talks with Samsung for a next-generation ASIC design and manufacturing contract worth over 10 trillion won. Meta's third-generation AI accelerator, "MTIA," plans to switch from TSMC to Samsung's most advanced 2nm process for mass production, with early chip architecture design handled jointly by Samsung's System LSI division to match its rapid development cycle of one generation every six months. Furthermore, US AI company Anthropic is also evaluating the use of Samsung's 2nm process to develop custom ASICs, aiming to internalize its AI infrastructure. Industry insiders point out that with the influx of AI chip orders from global tech giants, coupled with the ongoing negotiations with BYD for an automotive chip manufacturing agreement, Samsung's medium- to long-term order backlog for its wafer foundry business is expected to approach 50 trillion won, and it is expected to achieve a turnaround in operating profit in the fourth quarter of this year.
Anthropic's main model, Claude Sonnet 5, officially launched on the API platform | _2024111120230_
ChainCatcher reports that Anthropic's latest Sonnet series masterpiece, Claude Sonnet 5, is now officially available to developers via its API platform. This model achieves groundbreaking advancements in code generation and complex agent tasks. Leveraging a massive 1MB context window and innovative adaptive thinking capabilities, Claude Sonnet 5 can handle large and complex projects more stably and efficiently. Furthermore, to meet the needs of different business scenarios, Claude Sonnet 5 is the first in the industry to launch a dual-channel access model: "Official Direct Connection" and "Self-Selected Service Provider," comprehensively helping developers flexibly balance ultimate performance and operational costs. Starting today, users can log in to the console to apply for an API key for quick access and experience. For detailed technical guides and documentation, please refer to the official website.
A Ramp AI report shows that Anthropic's enterprise adoption rate has surpassed OpenAI's, with top companies' employees spending an average of $7,449 per month on AI.
According to Mars Finance, Ramp's Economic Lab released a new version of the Ramp AI Adoption Index report. Based on spending data analysis from over 70,000 US enterprise clients, Anthropic's enterprise AI adoption rate increased by 2.5 percentage points to 41%, officially surpassing OpenAI's slightly lower 39.5%, establishing its leading position in the commercial application field. The report focuses on analyzing the spending trends of leading companies that "deeply adopt AI." Data shows that the top 1% of companies spend as much as $7,449 per employee per month on AI, and this figure still achieved a 14.1% increase last month; in contrast, the top 10% of companies spend an average of $611 per employee per month, while the median company spends only $11.38 (approximately equal to the cost of a single basic subscription). In addition, the research points out that companies that deeply apply AI do not exhibit "vendor lock-in" and generally use multiple cutting-edge large models, open-source platforms, and vertical AI solutions simultaneously. Although companies are starting to try more cost-effective models (such as DeepSeek) when faced with cost pressures, overall AI spending is still on the rise.