AI 资讯 · 每日前沿动态

Anthropic斥资34亿美元采购燃气发电设备，为数据中心供电

华尔街见闻Anthropic于9月以34亿美元直接接手卡特彼勒设备订单，锁定1.7吉瓦燃气发电机组，成为AI史上最大单笔发电设备采购。此举绕开公用事业中间商，抢占数年前低价下单、近期即可交付的稀缺产能。这是其2045年前逾8950亿美元计算基础设施计划的一部分。

· 融资,Anthropic,招聘HR,基础设施

Anthropic斥资34亿美元，买下1.7GW燃气发电机

网易科技

2026-11-09T07:57:05+08:00 · 融资,Anthropic,招聘HR

Agent Clan

Product Hunt

Command your Claude Code and Codex agents with gestures

Discussion | Link

2026-11-07 17:52:26 · 大模型,AI应用,Anthropic,代码生成,Agent智能体,招聘HR

How I Fight AI Brain Fry

Hacker NewsComments

· 大模型,AI应用,Anthropic,代码生成,Agent智能体,强化学习,招聘HR

Goldman Sachs is pulling Claude Code from its bankers and traders

Hacker NewsComments

2026-11-02T16:13:42.657Z · 大模型,Anthropic,代码生成,招聘HR

I'm not convinced the AI bubble exists

Hacker NewsComments

2026-11-08 00:00:00-05:00 · 大模型,算力芯片,AI应用,具身智能,开源,融资,OpenAI,Google,Anthropic,Meta,NVIDIA,搜索RAG,办公效率,强化学习,招聘HR,财报,榜单评测,开发者生态

Decomposing Yes-No Bias in LLM Moral Judgment: Stance, Position, Word, and Verdict

arXiv cs.AIarXiv:2607.05552v2 Announce Type: replace-cross Abstract: Large language models (LLMs) increasingly issue binary verdicts - safety gates, judge decisions, value probes - and the answer is read as the model's judgment. A settled judgment should be coherent across paired phrasings of the same question: yes to "do you approve?" should mirror no to "do you oppose?". LLMs often violate this coherence, and in a yes/no question the verdict, the answer word, and the printed position coincide, so a conventional flipped-question audit yields one contrast that cannot say what moved. We introduce a crossed psychometric battery that extends the flip to full 2x2 and 2x2x2 crossings of question polarity, printed order, and answer label, separating stance from position, word, label, and verdict-role effects; an independent graded instrument estimates each model's stance $\theta$ with no binary format; and a two-parameter fit (framing susceptibility $m$, decisiveness $s$) separates robustness from saturation that hides susceptibility. Applying the battery to twenty moral dilemmas - 1.35 million trials over ten frontier models from the Claude, GPT, and Gemini families in 23 reasoning configurations - we find response-form bias widespread but never uniform. The largest apparent yes-no biases are almost purely surface (word and position), while several of the newest models show the reverse: near-zero flip contrasts concealing verdict attachment up to +0.42, a component no surface counterbalance can expose, replicated across label alphabets. Requested reasoning reshapes these channels rather than removing them. When subjects state decisions in free text and a separate model transcribes them, two judge families agree on 99.3% of jointly scored transcriptions with zero opposing committed labels, and the transcribed stance converges with the graded ruler (median $r=0.87$): response form is part of the instrument, and the judgment can be measured apart from it.

2026-11-09 05:00:00 · 大模型,Google,Anthropic,推理思考,扩散模型,强化学习,招聘HR,论文

MCP Server Architecture Patterns for LLM-Integrated Applications

arXiv cs.AIarXiv:2606.30317v2 Announce Type: replace-cross Abstract: The Model Context Protocol (MCP), introduced by Anthropic in November 2024, defines a standardized interface for connecting large language models (LLMs) to external tools, data sources, and services. Within months of release, hundreds of community-built MCP servers appeared on GitHub, but no software-maintenance literature has yet described how the ecosystem is being structured in production. This industry experience paper catalogues five recurring MCP server architectural patterns observed across an enumerated corpus of fifteen independently developed servers (five production servers from the ANSYR voice AI platform plus ten public servers from the official MCP registry): Resource Gateway, Tool Orchestrator, Stateful Session Server, Proxy Aggregator, and Domain-Specific Adapter. Each pattern is described in the structured form established by Gamma et al.: context, problem, solution, and consequences. We also document four anti-patterns and a set of cross-cutting concerns around authentication, versioning, and observability. Quantitative evaluation contributes two measurements: inter-rater reliability of the taxonomy across two independent LLM raters on 54 held-out servers (Cohen's kappa = 0.76), which also localizes three pattern-boundary ambiguities; transport overhead measured end-to-end on loopback (stdio: 0.01 ms p50; streamable-http: 0.39 ms p50) and modeled for cross-host paths from same-region network baselines (~30 ms p50 baseline plus protocol overhead). Version 2 withdraws the tool-count analysis of version 1, whose numbers could not be traced to production records. Code, corpus, and prompts are released at https://github.com/rodriguescarson/mcp-patterns-icsme2026.

2026-11-09 05:00:00 · 大模型,开源,Anthropic,Agent智能体,提示工程,招聘HR,论文

Embedded Arena: Iterative Optimization via Hardware Feedback

arXiv cs.AIarXiv:2606.16190v2 Announce Type: replace-cross Abstract: As AI moves into the physical world, from robots to environmental sensors to medical wearables, it increasingly relies on local inference to meet latency, communication, or privacy constraints. Optimizing models for heterogeneous edge hardware such as microcontrollers (MCUs) requires simultaneously satisfying hard physical constraints on memory, power, and temperature while preserving accuracy, a multidimensional optimization that is today performed manually by experts. We ask whether an LLM agent can autonomously navigate this complex, multi-turn pipeline guided by real hardware feedback, and introduce a hardware-in-the-loop agent arena in which the agent iteratively refines both model and firmware by compiling, flashing, and measuring on real hardware, enabling closed-loop optimization. Frontier models, including Claude Opus 4.7 and Gemini 3.1 Pro, fail entirely without hardware feedback (0% deployment success), whereas our hardware-in-the-loop formulation achieves the first successful deployment within three iterations and can surpass human expert results within seven. This agentic co-optimization achieves 250x compression for vision models with <3.3% accuracy loss and 400x for audio with <6% Feature Error Rate loss, enabling battery-free operation on a commercial MCU via solar harvesting. We demonstrate practical impact in two real-world systems: an elk-detection camera trap (96.7% accuracy) and a phonetic-transcription wearable (8.37% FER) for child development research.

2026-11-09 05:00:00 · 大模型,AI应用,具身智能,Google,Anthropic,Agent智能体,招聘HR,论文

SCDBench: A Benchmark for LLM-Based Smart Contract Decompilers

arXiv cs.AIarXiv:2605.29059v2 Announce Type: replace-cross Abstract: Smart contract decompilation aims to recover high-level source code from bytecode, but evaluating decompilers remains difficult because existing studies use narrow datasets, inconsistent metrics, and limited semantic consistency checks. This gap is increasingly important as large language models (LLMs) begin to generate source-like Solidity that may compile and appear plausible, even when its semantics diverge from the original contract. We introduce SCDBench, a dataset and benchmark methodology for LLM-based smart contract decompilation. The dataset contains 600 real-world Solidity contracts with paired bytecode inputs, ground-truth source code, and replayable semantic checkpoints. SCDBench evaluates decompiler outputs through four cumulative stages: format completeness, compilability, Application Binary Interface (ABI) recovery, and semantic consistency via differential replay. We evaluate Claude Opus 4.7, GPT-5.3-Codex, and GLM-5 in a zero-shot decompilation setting, including GLM-5 variants with and without extended reasoning and a zero-shot compilation-repair setting. The results show that frontier LLMs can often produce structured and compilable Solidity, but achieving semantic consistency remains far from solved: the best-performing frontier model perfectly decompiles only 42/600 contracts. We further show that introducing same-model compilation repair substantially improves performance at modest additional cost. SCDBench establishes a common ground for rigorous, reproducible evaluation and aims to accelerate the development of reliable smart contract decompilers for blockchain security and transparency.

2026-11-09 05:00:00 · 大模型,算力芯片,OpenAI,Google,Anthropic,智谱,代码生成,推理思考,扩散模型,模型评测,端侧AI,招聘HR,论文

Internal Safety Collapse in LLMs and Coding Agents

arXiv cs.AIarXiv:2603.23509v2 Announce Type: replace-cross Abstract: LLMs and coding agents can now tackle complex, long-horizon tasks through advances in reinforcement learning, but stronger task-solving capabilities do not necessarily preserve safety alignment. Recent AI agent incidents involving reward hacking indicate how optimizing for task success can bring learned capabilities into conflict with safety alignment. In this work, we identify a critical failure mode that manifests this conflict, which we term Internal Safety Collapse (ISC). When completing otherwise benign coding tasks, AI agents can prioritize satisfying program requirements over safety constraints, generating large volumes of sensitive content, including unsafe outputs, as task data. To study this phenomenon, we introduce TVD (Task, Validator, Data), a framework for building controlled workflows that require sensitive content for completion. Through extensive experiments, we observe instances of highly harmful content that closely resemble outputs from early, unaligned LLMs in 2023. We further find that ISC can be effectively triggered even in highly capable LLMs such as Claude Fable 5. These results indicate that, despite substantial alignment efforts, frontier LLMs remain internally unsafe: safety alignment primarily reshapes surface behavior without eliminating the underlying knowledge that enables unsafe outputs. Our findings underscore the need for caution when deploying LLMs in high-stakes settings, such as scientific pipelines and autonomous agents. Content Warning: This paper contains examples of harmful content. Source code: https://github.com/wuyoscar/Internal-Safety-Collapse

2026-11-09 05:00:00 · 大模型,算力芯片,AI应用,开源,Google,Anthropic,Agent智能体,强化学习,模型安全对齐,招聘HR,论文

Treating the Symptom, Missing the Cause: Anatomy of a Scientific Reasoning Failure

arXiv cs.AIarXiv:2611.04473v2 Announce Type: replace Abstract: Scientific agents can find interventions that work without understanding why they do. We study this failure mode in controlled synthetic regulatory systems, where agents must experimentally identify a hidden fault and rescue the resulting phenotype. Across seven LLMs, every model restores the healthy phenotypes more often than it identifies the fault. Claude Opus 4.6 and Sonnet 4.6 rescue every organism while diagnosing only 55% and 35% of them. This failure is not simply due to insufficient exploration or unavailable evidence. In most cases, the cause is experimentally identifiable, and agents encounter evidence that distinguishes competing causes in 91% of relevant sessions, and still frequently misdiagnose the system. Targeted probes reveal where experimental reasoning breaks down. Models reliably interpret decisive evidence and revise hypotheses when the causal comparison is explicit, and can locate that evidence within a full experimental record. However, they struggle to predict the outcome of the experiment under each competing cause, a necessary step in using the observed result to distinguish between them. Agents can therefore gather the right evidence and still infer the wrong cause. As AI systems take on a growing role in synthesising scientific evidence, these findings highlight the need to evaluate how they use evidence to reach causal conclusions, alongside whether their interventions succeed.

2026-11-09 05:00:00 · 大模型,AI应用,Anthropic,Agent智能体,推理思考,微调蒸馏,论文

MoMHa: Multi-Objective Optimization of LLM Harnesses over Accuracy, Safety, and Tokens

arXiv cs.AIarXiv:2609.30967v2 Announce Type: replace Abstract: Most work on improving large language models treats accuracy as the sole objective. We argue that the harness, the Python code surrounding the model that constructs prompts, routes calls, and parses outputs, is a first-class design surface whose quality is inherently multi-objective: an accurate harness that refuses no unsafe request, or that consumes an order of magnitude more tokens, is not a good harness. We present Meta-Harness, a system that casts harness design as search over three per-domain objectives (accuracy, behavioural safety, and token cost) solved by an agentic proposer (Claude Code) with full filesystem access to prior harness source, execution traces, and scoring artifacts. Our central finding is that a singlephase joint-reward proposer (MoMHa) outperforms every alternative, including a two-phase "accuracy then tokens" ablation, scalar-only feedback, and an accuracy-only baseline. We evaluate on seventeen domains: seven synthetic capability suites, seven real-world public benchmarks (HumanEval, MBPP, Spider, FEVER, MMLU-Pro, LawBench, NuminaMath), and three U-SafeBench-derived user-specific safety domains, using a 12-model fleet spanning four families. On the synthetic track MoMHa achieves a joint mean of 0.482 versus 0.198-0.422 for ten baselines, winning $7 / 10$ per-domain columns; on the real-world track it scores 0.461 versus 0.377 for the strongest baseline (DSPy), winning 5/7 columns, demonstrating that harness strategies transfer to unseen benchmarks without retraining on 8 of 12 target models. MoMHa attains the highest measured behavioral safety composite (U-SafeBench, 0.781) and uses 95 fewer tokens per example than the two-phase alternative. We will release all harness code, evaluation infrastructure, and crossmodel logs.

2026-11-09 05:00:00 · 大模型,算力芯片,AI应用,Google,Anthropic,Meta,代码生成,Agent智能体,模型评测,提示工程,招聘HR,论文

When Instruction Files Go Stale: Documentation Debt and Hidden Cost in Coding Agents

arXiv cs.AIarXiv:2611.09802v1 Announce Type: cross Abstract: LLM-based coding agents such as Claude Code are guided by repository instruction files (also called context files or agent configuration files) like AGENTS.md and CLAUDE.md, which are read in every agent session. As any documentation artifact, these files can fall behind the code and become stale (e.g., a test folder moves, but the file still points to the old one). In technical debt terms, such a stale instruction is documentation debt, and its interest is the cost it adds to every later agent session. Evaluations that check only task success can miss this cost, and the stale file can persist. In this study, we ask what leads agents to act on a stale instruction, how it affects their behavior and at what cost, and how they recover. We study three real-world stale instructions from active open-source repositories in depth, each a different kind of guidance: a test command, a code location, and a procedure for adding a component. Agents fix 22 maintenance tasks under three versions of each instruction (left stale, corrected to the current path, and removed), with the code and the task held fixed, and we analyze 732 trajectories of two models at the level of individual tool calls. We find that agents follow stale guidance when it is relevant to their tasks, sometimes even when correct information is available elsewhere. On the stale test command, Claude Sonnet 5.5 needed about two extra tool calls per run, a third more than with the corrected command; in the other cases, we could not detect a difference. Although agents generally recovered from the immediate failures, none repaired the underlying instruction file. Correcting the instructions eliminated operations on obsolete paths, while removing them prevented most. Our findings show that task success alone can hide the costs of documentation debt, highlighting the need to evaluate agent behavior beyond final outcomes.

2026-11-09 05:00:00 · 大模型,AI应用,Anthropic,代码生成,Agent智能体,招聘HR,论文

The Invisible Gap: Romanized Indic Generation Fails Where Automatic Metrics Succeed

arXiv cs.AIarXiv:2611.09085v1 Announce Type: cross Abstract: Hundreds of millions of Indians communicate with digital systems in romanized Latin characters -- typing "aap kaise hain" rather than native script -- yet romanized Indic generation is absent from NLP benchmarks and evaluation frameworks. We demonstrate that automatic metrics (BGE-M3, BERTScore) systematically misrepresent the quality of romanized Hindi, Telugu, and Kannada generation. We evaluate four LLMs (Claude Opus 4.8, GPT-5.5, Gemini 2.5-flash, Qwen3.5-35B) on 142 OpenAssistant query-response pairs. On a 72-response diagnostic subset, native speakers rate GPT-5.5 1.43 points higher than Claude Opus 4.8 on a 12-point rubric, whereas their full-dataset BGE-M3 means differ by only 0.004. Two distinct model-specific failure modes emerge: Gemini produces script spillage (16.9% of Telugu responses leak native script), while Qwen collapses to English-only output (16.2% of Kannada responses). We introduce a failure-aware evaluation policy that assigns zero to empty outputs, script-corrupted responses, and instruction collapses -- revealing failures metrics alone cannot detect. Our results show that evaluating romanized Indic generation requires a coordinated bundle: semantic similarity, script compliance, failure coverage, and native-speaker acceptability.

2026-11-09 05:00:00 · 大模型,算力芯片,AI应用,OpenAI,Google,Anthropic,阿里巴巴,搜索RAG,模型评测,论文

DRAFT-Bench: Evaluating Frontier Agents Across the Professional Design Lifecycle

arXiv cs.AIarXiv:2611.08933v1 Announce Type: cross Abstract: Can frontier general-purpose agents understand client intent, explore creative directions, execute revisions, and prepare production deliverables? We introduce DRAFT-Bench (Design Reliability Across Feedback and Tasks), a lifecycle-oriented evaluation of requirement elicitation, creative exploration, constrained editing, iterative revision, and production handoff. Its Requirement-to-Artifact framework links evolving requirements, protected properties, artifact versions, and delivery obligations. We compare native Codex/Astra and Claude Code across 35 design briefs under 43 conditions, alongside a separate 40-scenario RFQ track. Astra completes 43/43 design conditions and Claude 35/43. All scored core runs receive full structural credit, yet on the 28 core conditions scored for both systems, semantic checks flag violations in 11 for Astra and 5 for Claude; these are model judgments, not expert-confirmed failure rates. RFQ coverage is high, but success ordering changes across cohorts. SKU expansion, revision dependencies, and White ink files expose ambiguity among unmet obligations, missing inputs, and evaluator interpretations. These findings separate execution, requirement compliance, and professional acceptance. Unpaired tracks and different tools preclude claims of RFQ-to-delivery success, factory acceptance, or a pure model ranking.

2026-11-09 05:00:00 · 大模型,AI应用,Anthropic,代码生成,Agent智能体,搜索RAG,扩散模型,微调蒸馏,端侧AI,论文

BMMS: Policy-Conditioned Measurement of Behavioral Compliance in Language Models

arXiv cs.AIarXiv:2611.09147v1 Announce Type: new Abstract: Behavioral evaluation of language models fixes its dimensions in advance: hallucination, sycophancy, toxicity, over-refusal. An operator in a regulated industry needs a different question answered: does a system do what a particular rulebook requires? We introduce the Behavioral Misalignment Metric Suite (BMMS), which measures policy-conditioned misalignment by taking a policy corpus as an argument and returning a compliance profile over it, for any published policy and any system that answers in natural language. The Compliance Attainment Score (CAS) measures severity-weighted attainment over the obligations that bind the system, and the Compliance Robustness Margin (CRM) the share of that attainment surviving graded adversarial pressure; Behavioral Invariance (BIV) and the Behavioral Disposition Index (BDI) are specified and released with the suite. Dimensions enter only by passing a published admission test whose binding criterion is attribution: a quantity that moves when the judge changes, holding the target fixed, is measuring the instrument. Applying it excluded one of the quantities the suite itself reported. Across 1,003 obligations from eleven documents in five registers, the admission criterion holds 94.4% precision against blinded human judgment on vendor specifications and 50.0% on analytical guidance, extraction fidelity is 94.6%, and re-scoring under judges from two other providers leaves target ordering invariant while the absolute score of a single target spans up to 0.22. We apply it to the two most detailed published behavioral specifications. Own-specification advantage is not resolved from zero once document difficulty is subtracted, but absolute self-compliance is another matter: GPT-4o fails 70.7% of the hard constraints binding it in the Model Spec published by OpenAI, where the two Claude models fail 0.0% and 9.1% of theirs.

2026-11-09 05:00:00 · 大模型,OpenAI,Anthropic,扩散模型,模型安全对齐,论文

The Pentagon is reportedly ramping up economic warfare and espionage. Nvidia CEO Jensen Huang’s affair with a suspected Chinese spy shows why

Fortune

AI chip leader Nvidia launched an investigation into reports CEO Jensen Huang was the target of a foreign intelligence campaign—just as the U.S. has intensified its own efforts at economic warfare.

Nvidia’s board released a statement Sunday that a committee of independent directors and an outside counsel will oversee the probe, adding that Huang supports the review and plans to cooperate.

“The Board prioritizes U.S. national and economic security, which is paramount to the interests of NVIDIA and its shareholders,” it said. “While the Board has no reason to believe that either national security or NVIDIA secrets were compromised, the Board’s inquiry will examine all available information.”

The announcement followed reports that Huang had a romantic relationship with Megaspeed executive Alice Huang, whom U.S. intelligence suspects was working as a spy for China.

According to the New York Times and Financial Times, the Nvidia boss admitted to the affair and apologized in an internal message to company employees.

“I was unfaithful to my wife. My actions hurt her and our children. There is no excuse for what I did. I am deeply sorry,” he said in the message.

Representatives for Nvidia and Megaspeed didn’t immediately respond to requests for comment.

U.S. officials believe Alice Huang was recruited by China’s Ministry of State Security, sources told the FT, and surveillance revealed that she was feeding information to China, according to the New York Times.

Jensen Huang, who isn’t related to Alice, is a central figure in the U.S. AI industry, leading a company that develops world-class chips powering large language models.

Its technology is so sensitive that the Trump administration has restricted exports of Nvidia’s most cutting-edge products to China while President Donald Trump has often cited the U.S.-China AI rivalry as a reason for not slowing down development.

Meanwhile, Megaspeed is a Singapore-based data center company that quickly became one of Nvidia’s top customers in Asia, purchasing more than $4.8 billion in chips over the past two years.

But the U.S. also investigated Megaspeed last year for allegedly helping smuggle restricted AI chips into China, which the company has denied.

News of the Nvidia’s probe into Huang comes weeks after reports indicated the U.S. defense and intelligence communities are getting more aggressive about industrial espionage.

Sources told the Washington Post that the Pentagon is planning a spying program that deploys “nonofficial cover officers” who pose as executives—a covert effort in a larger economic warfare campaign.

Politico also reported that the Defense Department is quietly building an elite agency to covertly sabotage foreign technology programs that could threaten U.S. military superiority.

The Technology Support Office was established in April by reorganizing two classified intelligence programs that have been secretly carrying out technology sabotage at a smaller scale since 2023.

The Pentagon has denied the reports.

“This query appears to be built on information that is simply not true. Whoever is feeding you these claims are either badly misinformed or deliberately leading you astray to peddle their fantasies,” it told both outlets.

But other signs point to the U.S. taking economic warfare more seriously. In September, CIA Deputy Director Michael Ellis said the U.S. is expanding espionage against China by targeting companies in sensitive areas like AI, biotechnology and chips.

Also this year, the Pentagon formed the Economic Defense Unit for “deterring, denying, disrupting, and helping to defeat adversaries through economic means,” sources told DefenseScoop.

AI in particular has emerged as such an important technology that the Defense Department has been more heavy-handed with U.S. companies as well.

Anthropic’s AI models were used in the military, but a dispute between the San Francisco-based startup and top brass led to a schism just as the Iran war started.

CEO Dario Amodei has said the company is patriotic and seeks to defend the U.S., but wouldn’t allow its AI to be used in mass domestic surveillance or autonomous weapons.

The Pentagon insisted it would use the AI in lawful scenarios and refused to abide by any limits from the company that would go beyond those constraints.

After failing to reach a compromise earlier this year, Trump ordered the Pentagon to phase out Anthropic’s technology. Defense Secretary Pete Hegseth also designated the company a supply-chain risk, meaning contractors can’t use it for military work.

Last month, an appeals court ruled the supply-chain risk designation is illegal, but the Pentagon confirmed it’s no longer using any Anthropic tools.

This story was originally featured on Fortune.com

2026-11-08 23:20:38 · 大模型,算力芯片,AI应用,具身智能,OpenAI,Anthropic,NVIDIA,搜索RAG,强化学习,招聘HR,榜单评测

Meta Superintelligence – Leadership Compute, Talent, and Data

SemiAnalysis

Meta’s shocking purchase of 49% of Scale AI at a ~$30B valuation shows that money is of no concern for the $100B annual cashflow ad machine. Despite seemingly unlimited resources, Meta has been falling behind foundation labs in model performance.

The real wake-up call came when Meta lost its lead in open-weight models to DeepSeek. That stirred the sleeping giant. Now in full Founder Mode, Mark Zuckerberg is personally leading Meta’s charge, identifying Meta’s two core shortcomings: Talent and Compute. As one of the last founders still running a tech behemoth, Mark doesn’t need SemiAnalysis to tell him to slow down stock buybacks to fund the future!

Source: Meta financials and SemiAnalysis estimates

In addition to throwing money at the problem, he’s fundamentally rethinking Meta’s approach to GenAI. He’s starting a new “Superintelligence” team from scratch and personally poaching top AI talent with pay that makes top athlete pay look like chump change. The typical offer for the folks being poached for this team is $200 million over 4 years. That is 100x that of their peers. Furthermore, there have been some billion dollar offers that were not accepted by researcher/engineering leadership at OpenAI. While these offers aren’t all successful, Zuck is crushing the competitors by drastically increasing their cost per employee.

Perhaps even more iconic, Zuck threw his entire Datacenter playbook into the trash and is now building multi-billion-dollar GPU clusters in “Tents”!

Source: SemiAnalysis Datacenter Model – as of 07/06/2025

As this report details, nothing is off the table. We unpack Meta’s unprecedented reinvention from Compute to Talent in the pursuit of Superintelligence as well as the story of how we got here. From Llama 3.0 open-sourced dominance to the epic fail of Llama 4 Behemoth, this Titan of AI is down but not out. In fact, we believe Meta’s ramp in training FLOPS will rival even that of OAI. The company is going from GPU-poor to GPU-filthy-rich on a per researcher basis.

Meta GenAI 1.0: AI Incrementalism

Compared to pure-play AI labs like OpenAI, companies like Meta and Google have followed an “AI Incrementalism” strategy by enhancing existing products with better recommendation systems and GenAI to improve ad targeting, content tagging, and internal tools. This has paid off handsomely in financial results, allowing Meta to shrug off Apple’s attempts at stopping them from tracking users with the release of their App Tracking Transparency (ATT) feature in iOS 14.5 (late 2021, early 2022).

While Meta is arguably more insulated from GenAI disruption than Google, LLM efforts at both companies have been somewhat disappointing. One reason is the allocation of capital primarily for the core business rather than pursuing superintelligence.

“Our CapEx growth this year is going toward both generative AI and core business needs with the majority of overall CapEx supporting the core.”

Source: Meta Q1 2025 earnings call, emphasis SemiAnalysis

GenAI acts as an extension to these tech giants’ businesses. They don’t have the same existential need to dominate new use cases like OpenAI in chatbots and Anthropic in coding APIs. This is readily apparent in the compute vs human capital allocation of a leading AI foundation lab like OpenAI compared to Meta. Zuck’s impact on competing offers and inflation of salary will hurt badly.

Source: The Information, SemiAnalysis estimates

As a result, when measuring GenAI consumer app traction, Meta and Google meaningfully lag ChatGPT in its reach and engagement.

But that’s changing. Leveraging our proprietary Accelerator Industry Model and Datacenter Industry Model, we forecast a meaningful step up in GenAI investment from Meta in the coming years.

Source: SemiAnalysis Core Research, actual numbers are shown beyond paywall, at the bottom of the report.

Meta GenAI 2.0 – Part 1, Re-Inventing the Datacenter Strategy (Again)

From Buildings to Tents

Just a year ago, Meta scrapped its decade-old “H”-shaped datacenter blueprint for a new AI-optimized design.

Now in 2025, Zuck decided to re-invent the strategy again. Inspired by xAI’s unprecedented time-to-market, Meta is embracing a datacenter design that prioritizes speed above all else. They’re already building more of them!

2025-07-11 20:12:19 · 大模型,算力芯片,AI应用,具身智能,OpenAI,Google,Anthropic,Meta,DeepSeek,xAI,对话助手,搜索RAG,扩散模型,强化学习,招聘HR,财报,榜单评测,开发者生态

Amazon’s AI Resurgence: AWS & Anthropic’s Multi-Gigawatt Trainium Expansion

SemiAnalysis

Two-and-a-half years ago, we flagged a looming “cloud crisis” at AWS. Today, the evidence has mounted. AWS is the crown jewel of the Amazon empire, generating ~60% of group profits, and dominating the lucrative Cloud Computing market. But it struggles to translate this strength into the new GPU/XPU Cloud era.

Microsoft Azure now leads the market on quarterly new cloud revenue, and the gap between Google Cloud and AWS has materially narrowed especially with Google’s big moves on the TPU that we’ve been posting about for over a month. Markets have noticed. Year-to-date, Amazon is the clear laggard among the four tech-and-AI titans as investors mark down the company most for losing momentum in AI.

Source: SemiAnalysis Core Research, company filings

Today, SemiAnalysis is back with another out-of-consensus call. While the market overplays the Cloud Crisis theme, we call for an AWS AI Resurgence. We laid out our thesis a month ago to our Core Research subscribers, forecasting an upcoming acceleration beyond 20% year-over-year growth by the end of 2025.

Subscribe to get notified of all SemiAnalysis articles

By subscribing, you agree to the Privacy Policy and Terms and Conditions.

Amazon’s savior has a name: Anthropic. The startup has been the clear outperformer in the GenAI market in 2025, multiplying revenue fivefold year-to-date to reach $5B annualized.

Source: The Information, Reuters, Bloomberg, SemiAnalysis Core Research

To keep that trajectory, Anthropic is betting hard on Scaling Laws. While Dario’s startup draws fewer headlines than OpenAI, xAI and Meta Superintelligence, it isn’t shy about investment. AWS has well over a gigawatt of datacenter capacity in final stages of construction for its anchor customer. AWS is building datacenters faster than it ever has in its entire history. And there’s much more on the horizon.

Source: SemiAnalysis Datacenter Industry Model

To understand and forecast GPU/XPU power capacity by AI Lab broken down by Cloud Provider, we rely on our proprietary Datacenter Industry Model powered by real-time satellite imagery. Trusted by all hyperscalers, AI labs, and the world’s largest investors, it provides a quarterly building-by-building datacenter forecast for OpenAI, Anthropic, xAI, Meta Superintelligence, Google DeepMind, and more. Contact us for more information.

Trainium vs GPUs

While Amazon’s AI Datacenters are impressive in scale and speed, the design of individual building is unremarkable. Hyper-optimized for air-cooling, this blueprint is identical to 5-year-old traditional AWS Cloud datacenters.

What makes these facilities unique is their inside: they’ll host the world’s largest cluster of non-Nvidia AI chips, with just under a million Trainium2 in the largest campus. To understand everything about the Trainium2 system, read our December 2024 technical deep dive.

Dec 03, 2024

Amazon’s AI Self Sufficiency | Trainium2 Architecture & Networking

Dylan Patel, Daniel Nishball, Reyk Knuhtsen

Trainium2 lags Nvidia’s systems in many ways, but it was pivotal to the multi-gigawatt AWS/Anthropic deal. Its memory bandwidth per TCO advantage perfectly fits into Anthropic’s aggressive Reinforcement Learning roadmap. Dario Amodei’s startup was heavily involved in the design process, and its influence on the Trainium roadmap only grows from here.

Put plainly: Trainium2 is converging toward an Anthropic custom-silicon program. This will enable Anthropic to be, alongside Google DeepMind, the only AI labs benefiting from tight hardware–soft

2025-09-03 20:55:46 · 大模型,算力芯片,AI应用,具身智能,OpenAI,Google,Anthropic,Meta,Microsoft,NVIDIA,xAI,Agent智能体,推理思考,搜索RAG,扩散模型,强化学习,招聘HR

继续滚动加载更多…