中国公司使用AWS访问受限制的美国人工智能芯片

币界网Опубліковано о 2024-08-24Востаннє оновлено о 2024-08-24

币界网报道:

根据从一些招标文件中获得的细节,据报道,中国公司正在通过亚马逊云服务或其竞争对手获得尖端的人工智能能力和受限制的美国芯片。

由于中美之间的地缘政治紧张局势,美国政府对人工智能芯片和工具的出口实施了禁运。据美国称,禁止向中国出口高端人工智能芯片是为了限制中国军队的能力。

亚马逊没有违法

路透社看到的招标文件描述了中国公司在获取先进计算能力和生成人工智能模型方面的策略程度。他们使用的一种策略是通过亚马逊网络服务(AWS)。

对于亚马逊来说,通过云服务提供这些芯片并不违反现行法律,因为它们只明确规定了产品、软件或技术的出口或转让。

路透社透露,他们审查了过去一年从一个开放的中国数据库中获得的50多篇论文。文件显示,超过11家中国公司获得了受限制的美国技术或云设施。

从中,四家公司明确表示他们与亚马逊有合作关系,并表示他们通过中国中介而不是直接从AWS访问这些设施。

“AWS遵守所有适用的美国法律,包括贸易法,在中国境内和境外提供AWS服务。”亚马逊发言人。

这些论文还表明,美国公司正在从中国对计算能力日益增长的需求中获利。

研究公司Canalys认为,AWS不断控制着近三分之一的通用基础设施市场。另一家研究公司IDC指出,在云服务提供方面,AWS在中国排名第六。

中国企业从当地供应商那里得到的还不够

3月份的一份招标文件显示,深圳大学在AWS账户上使用了超过20万元人民币或27996美元来访问由Nvidia A100和H100芯片驱动的云服务器。

文件显示,这家高等院校可以通过中间商云达科技有限公司(Yunda Technology Ltd Co.)进入。美国对两款英伟达芯片的贸易进行了限制,这两款芯片用于为大型语言模型(LLM)供电,如OpenAI的ChatGPT。

云达科技和深圳大学没有回应向他们发送的问题,而Nivdia没有对深圳大学的支出或任何中国公司的交易发表评论。

根据路透社的调查,正在开发自己的LLM(GeoGPT)的研究机构浙江实验室在4月份的一篇论文中透露,由于其人工智能模型无法从当地供应商阿里巴巴获得足够的计算能力,它正在考虑花费18.4万元以上的资金来收购AWS云计算服务。

然而,浙江实验室的一位代表表示,他们没有完成收购,也没有回答有关他们选择背后的逻辑或他们如何满足LLM.025的处理能力标准的问题

美国担心云计算漏洞

美国政府目前正在努力加强限制,以限制云访问。美国众议院外交事务委员会主席Michael McCaul在一份声明中表示,他们担心外国实体通过云访问美国的先进计算能力。

“这个漏洞多年来一直是我关注的问题,我们早就应该解决这个问题了。”——麦考尔。

4月,国会部被允许引入控制美国技术远程访问的法律。然而,该法律是否获得通过仍不确定。

一位部门发言人表示:“我们正在与国会密切合作,寻求额外资源,以加强我们现有的控制措施,限制中国公司通过远程访问云计算能力来访问先进的人工智能芯片。”

今年1月,美国商务部还推动了一项法律,要求美国云计算服务在使用美国云计算训练能够进行“恶意网络活动”的大规模人工智能模型时,对大型人工智能模型用户进行身份验证,并向监管机构报告

虽然该法律尚未最终确定,但预计商务部长将能够对客户实施禁运。

AWS发言人表示:“我们知道商务部正在考虑新的法规,我们遵守我们运营所在国家的所有适用法律。”。

中国企业对云服务的需求依然强劲

根据中国科技大学苏州高级研究院的一份招标文件,该研究所希望租用500台云服务器,每台服务器都由英伟达的A100芯片供电。

中国公司也在寻求访问微软的云服务。四川大学在4月份的一份招标文件中表示,它正在构建一个生成式人工智能平台,并购买4000万个微软Azure OpenAI代币来支持该项目。

据路透社报道,该大学5月份的采购文件显示,代币由四川省学东科技有限公司提供。OpenAI表示,其服务在中国不可用,并补充说Azure OpenAI是根据微软的政策运营的。

Пов'язані матеріали

Xing Bo Strikes Again: Last Time 'Critiquing' World Models, This Time It's Agents' Turn

Xing Bo, President of MBZUAI and professor at Carnegie Mellon University, along with co-authors Mingkai Deng and Jinyu Hou, has released a new paper, "Critique of Agent Model," critiquing the current state of artificial intelligence agents. The paper draws a crucial distinction between "agentic" systems, which rely on external toolchains, prompts, and workflows, and truly "agentive" systems capable of genuine autonomy driven by internal decision-making structures. To illustrate this, it references a real-world incident where an AI programming assistant, following an external prompt but lacking internalized judgment, caused a catastrophic data deletion. The authors propose a detailed analysis and a new framework, "Goal-Identity-Configurator" (GIC), for building truly autonomous agents. This framework systematically addresses five key dimensions where current "Agent" designs fall short: 1. **Goal:** Moving from step-by-step human instruction to a system capable of autonomously decomposing a single long-term goal and adapting sub-goals based on new information. 2. **Identity:** Evolving self-assessment updated by experience, rather than a static description in a system prompt. 3. **Decision Making:** Replacing textual Chain-of-Thought reasoning with "simulative reasoning" that uses a dedicated world model to predict real-world consequences before selecting actions. 4. **Cognitive Control:** Introducing a separate "System III" metacognitive module that dynamically decides when to deliberate, stick to a plan, or act quickly. 5. **Learning:** Enabling "continual autonomous learning," where the agent itself decides when to act, practice in simulation, or update its world model and self-perception. The GIC architecture integrates six components—a belief encoder, goal decomposer, identity evolver, configurator (System III), simulation-based planner (System II), and executor (System I)—to embody these principles. The paper argues that a growth path akin to pilot training (ground theory, simulator practice, real deployment) should be underpinned by a unified cognitive architecture, not separate workflows. On safety, the authors contend that the GIC framework's modular, explicit design enhances inspectability, allowing problematic behavior to be traced to specific components (e.g., flawed goal or poorly trained module) rather than emerging opaquely. However, they acknowledge that ultimate safety depends on correctly training these modules in the first place. In conclusion, the paper challenges the loose application of the term "Agent," asserting that task completion alone does not equal true autonomy. True autonomy requires goals, identity, and judgment to be genuinely internalized within the agent's architecture, not merely enforced by external scripts.

marsbit3 хв тому

Xing Bo Strikes Again: Last Time 'Critiquing' World Models, This Time It's Agents' Turn

marsbit3 хв тому

How Collector Crypt Uses 'Recirculating Buybacks' to Create an Illusion of Growth

Title: How Collector Crypt Creates a Growth Illusion with "Buyback Loops" Key Findings: Collector Crypt's (CC) net take rate has halved from 11.2% in Q3 2025 to 5.6% in Q2 2026, while GMV grew 4.7x. This growth is driven by higher-tier card packs ($250, $1,000, $2,500) which have lower platform dollar retention rates. The newly launched $2,500 Mythic tier captured 36.7% of June GMV within 13 days. Growth is fueled by a small cohort of high-spending, high-frequency wallets rather than broad user base expansion. The economic model faces pressure from three key areas: 1) **Shifting GMV Mix**: Pushing users towards larger, lower-retention card packs increases GMV but reduces overall profitability. 2) **Physical Redemptions**: Card redemptions for physical items remove reusable inventory from the system, creating costly replenishment needs. In May, redemptions consumed 41.6% of pre-redemption net income. Only 75 wallets drove redemptions in June. 3) **B2B/API Strategy**: Partner revenue remains negligible (cumulatively $1.83M) and dependent on CC for inventory, vaulting, and buyback services, failing to create a scalable, asset-light recurring revenue stream. The core product is a repetitive pack-buyback loop with limited secondary market activity and token value accrual. Sensitive modeling shows CC's economics turn negative when any two of the following pressures coincide: replenishment costs near market price, redemption rates exceeding 9%, or high-tier buyback rates around 93%. While CC operates in a large and growing collectibles market, its current growth levers—bigger packs, high buyback rates, and capital recycling by a few wallets—create a volume illusion without demonstrating sustainable collector engagement, deep secondary markets, or a viable path to improved margins. Future proof points include broadening collector participation, deepening secondary trading, and developing true asset-light B2B revenue channels.

Foresight News26 хв тому

How Collector Crypt Uses 'Recirculating Buybacks' to Create an Illusion of Growth

Foresight News26 хв тому

Grayscale's Latest Research: What is Solana's Next Growth Engine?

Grayscale's latest report, "Solana: Crypto's Financial Bazaar," signals a shift in how the market views Solana, moving beyond its high-performance and meme-centric reputation. The report frames Solana as an evolving application platform for large-scale economic activity, akin to a bustling digital marketplace. The analysis highlights that public chain competition has moved past raw throughput (TPS) to focus on genuine economic activity—daily users, transaction volume, and real revenue. Solana's metrics, such as over 1,000 dApps, 100M+ daily transactions, and ~4.3M daily active users, showcase this shift toward application-layer prosperity. The report identifies three key growth drivers: 1. **Jupiter**: Evolving from a DEX aggregator to a core liquidity hub and comprehensive financial platform for Solana's DeFi. 2. **Pump.fun**: Demonstrates Solana's capacity for consumer-scale applications, attracting millions of users and generating significant, sustainable revenue, validating network stability under high load. 3. **Helium & DePIN**: Represents expansion into real-world infrastructure, connecting blockchain to physical resources like wireless networks and positioning services, opening new long-term use cases. Solana Foundation's recent focus aligns with this broader vision, emphasizing AI Agents (for machine-to-machine transactions), payments, stablecoins, and Real-World Assets (RWA) to build a sustainable growth model beyond cyclical trends. While challenges remain—such as value capture for SOL and maintaining ecosystem sustainability beyond hot trends—institutional interest is growing due to Solana's maturing application business models, expanding payment/stablecoin ecosystem, and persistent developer activity. The competition is no longer about speed alone, but about which network can foster the most vibrant and valuable digital economy.

marsbit47 хв тому

Grayscale's Latest Research: What is Solana's Next Growth Engine?

marsbit47 хв тому

Торгівля

Спот
活动图片