AMD拿下Anthropic数百亿美元芯片大单，另将投资50亿美元获股权
Related
Anthropic: The Claude model contains elements similar to "conscious human thought."
BlockBeats reported on July 7th that Anthropic released a new research report discovering a spontaneously generated internal "global working space" called J-space (Jacobian space) within the Claude model. This is a dedicated set of neural activation patterns for the model to engage in silent thinking, allowing it to process concepts without writing them down, similar to conscious thought that humans can report. The research used Jacobian technology to identify neural activation patterns in the J-space, enabling it to read unspoken concepts and modify these patterns. Experiments showed that disabling the J-space weakens multi-step reasoning abilities but does not affect basic tasks or factual recall. Click the original link below to join the Beating · Lark AI news channel for 24/7 monitoring of global AI hotspots and news.
SemiAnalysis: Meta will accelerate, rather than slow down, its computing power procurement; it is in talks with Anthropic to build its own AI model service platform.
According to BlockBeats, on July 3rd, SemiAnalysis stated in its latest report that after news broke that Meta might become a new Neocloud, the market's initial reaction was to sell off computing power cloud companies like CoreWeave and Nebius, and to renewed concerns about "AI computing power oversupply." However, the firm's assessment is the opposite: this concern may be wrong. Meta's data center and computing power procurement will not slow down, but rather continue to accelerate. The article mentions that in the first half of this year alone, Meta has already signed contracts for over 5GW of capacity in the cloud services and managed data center sectors, and this does not include its accelerating self-built projects. SemiAnalysis states that Meta is in final negotiations with Anthropic, hoping to obtain access to Claude private instances. If this comes to fruition, the significance goes beyond "Meta buying more computing power"; it suggests that Meta may be building its own AI model service platform. This model is somewhat similar to AWS's Bedrock, Microsoft's Foundry, and Google's Vertex. Meta can initially use Claude internally, or it can package the model capabilities as a token-as-a-service to provide services externally in the future. In the short term, it might use its own models externally and Anthropic models internally; in the long term, Meta might even incorporate Anthropic and OpenAI models into its external service system. SemiAnalysis states that the underlying logic is that Meta has computing power, advertisers, social network distribution capabilities, and consumer-end entry points. If it can combine cutting-edge models, intelligent agents, and sales and marketing SaaS, it will not just be a company that buys GPUs, but will be moving towards the upper layers of AI application and model distribution.
Palantir CEO: Enterprises are dissatisfied with "cutting-edge labs" like OpenAI and Anthropic, which only pursue token maximization.
According to BlockBeats, on July 2nd, Palantir CEO Alex Karp, in an interview with CNBC's "Squawk Box," strongly criticized leading AI model companies, calling the way AI is sold "completely wrong." Karp emphasized that companies are already dissatisfied with "cutting-edge labs" like OpenAI and Anthropic, believing they only pursue token maximization, wasting companies' time and money while handing over proprietary value and IP. Karp stated that companies are "angry" and will commit to owning their own AI production resources rather than relying on third parties. On June 29th, Palantir partnered with Nvidia to deploy Nvidia Nemotron open AI models in sovereign environments, primarily serving the US government and critical infrastructure customers. The collaborative system reportedly integrates Nvidia AI technology with Palantir's AIP, Foundry, Ontology, and Apollo platforms, helping organizations train, customize, and deploy AI locally while maintaining complete control over data, intellectual property, and models.
Amazon AWS establishes a $1 billion AI FDE team to emulate the enterprise AI deployment models of OpenAI and Anthropic.
BlockBeats reported on June 30th that Amazon Web Services (AWS) announced the creation of a new Forward-Deployed Engineer (FDE) organization and will invest $1 billion in internal resources to help enterprise customers deploy customized AI agents and related systems. AWS stated that FDE engineers will be directly stationed at customer companies, responsible for building AI applications, optimizing workflows, and helping customers build the ability to independently develop and operate AI systems, rather than simply delivering and maintaining them. The FDE model was first promoted by Palantir and has rapidly gained popularity in recent years due to the growing demand for enterprise AI deployment. Previously, OpenAI and Anthropic also launched related FDE joint ventures, with sizes of approximately $4 billion and $1.5 billion respectively. Unlike the two companies, which formed joint ventures with private equity firms, AWS's $1 billion investment this time primarily comes from internal resources and is not an independent investment project. --------------------------------- Click the original link below to join the Beating · Lark AI news channel and monitor global AI hot topics and news 24/7.
Due to high costs, Amazon is renegotiating its agreement with Anthropic while also considering other AI models.
According to BlockBeats, on June 30th, Anthropic renegotiated with Amazon, changing its transaction protocol from being based on computation time to being based on tokens. Meanwhile, Amazon is evaluating other AI models to mitigate Anthropic's rising costs. (The Information) --------------------------------- Click the original link below to join the Beating · Lark AI news channel and monitor global AI hotspots and news 24/7.
Nvidia Blackwell tops first intelligent agent hardware benchmark: 20 times more energy efficient than H200, outperforming AMD.
According to Beating's monitoring, Artificial Analysis, a benchmarking organization, released the industry's first hardware benchmark for intelligent agents, AA-AgentPerf. Traditional benchmarks are like a single question-and-answer "sprint," only looking at response speed; intelligent agent tasks are like a "relay race," where AI needs to autonomously break down the goal and repeatedly go through reading and writing files, rewriting code, and running tests. Frequent interactions pose extremely high challenges to server memory capacity and scheduling efficiency. The benchmark, by replaying the real programming trajectory, uses "the scale of concurrent agents supported per megawatt of power consumption" as the core energy efficiency indicator, directly addressing the power and cost bottlenecks of data centers. The first phase of testing ran the 1.6 trillion parameter open-source model DeepSeek V4 Pro. The results show that the NVIDIA Blackwell liquid-cooled cabinet system GB300 NVL72 can support 61,400 concurrent agents per megawatt of power consumption, while the previous generation Hopper HGX H200 could only support 2,600, representing an energy efficiency improvement of over 20 times. The single-GPU concurrency capacity has also increased by 41 times. This allows data centers to support 20 times more concurrent intelligent agents under the same power budget, significantly reducing the deployment cost of applications such as automated programming and customer service. In the initial results, the AMD Instinct MI355X is temporarily lagging behind. The review agency pointed out that both AMD and H200 configurations are built using the general open-source vLLM framework without deep optimization; as the service framework and kernel operators are adapted, there is still room for improvement in AMD's performance. Currently, inference providers such as Together AI have taken the lead in deploying DeepSeek V4 Pro on Blackwell, providing real-time inference support for the intelligent agent programming tool Cursor.