据知情人士消息，AI 推理基础设施提供商 Modal Labs 正接近完成由 Accel 领投的 7.5 亿美元融资，投后估值 157.5 亿美元，较四个月前 46.5 亿美元的估值增长逾两倍。

Mistral 在慕尼黑开设德国中心，组建专注 Physics AI 与工业 AI 的研究团队，并计划到 2030 年建成 1 吉瓦欧洲算力。该中心将携手 BMW 开展碰撞仿真与工程 AI 合作、与 Siemens Energy 推进工业 AI 应用，并与慕尼黑工业大学（TUM）合作利用风洞设施开发汽车空气动力学数字孪生。

Google 的 Project Suncatcher 首颗试验卫星 MVP 将于 10 月 1 日发射，用于验证其轨道 AI 数据中心构想。该卫星约冰箱大小，搭载 4 块 Google 自研 TPU AI 加速器，太阳能板供电约 1 千瓦。

XPRIZE Wildfire 公布 1100 万美元竞赛结果，两大赛道大奖均空缺。太空探测赛道要求 10 分钟内识别澳大利亚大范围火情，SIRIUS Wildfire Alliance 获 50 万美元一等奖；自主响应赛道三支决赛队均在 10 分钟内探测到阿拉斯加高风险火情，但无一完全扑灭。

Mistral 与 Cloudera 宣布合作，将 Mistral 模型集成进 Cloudera 混合数据平台，企业可在私有云、公有云、本地及完全气隙环境中部署推理并保持完全控制。企业还能在受控环境中用专有数据训练定制模型，数据与模型所有权均归企业，模型基于开放权重。Cloudera 平台上客户管理的数据规模达 30 exabytes。

Mistral 与 HUMAIN 宣布战略合作，覆盖 AI 基础设施、先进模型开发与 AI 解决方案部署，初期聚焦网络安全和语音，并计划开发阿拉伯语表现强劲的前沿模型。合作规模达数亿欧元，Mistral 将探索使用 HUMAIN 数据中心基础设施，双方还将在沙特面向受监管行业制定联合市场策略。

China plans to invest about 2 trillion yuan, or $295 billion, over five years to build nationwide data centers, with funding covering large-scale data center infrastructure for domestic AI development.

Nvidia and SK Hynix signed a multi-year pact to co-design future generations of memory chips for AI applications; the RSS snippet does not disclose product specifications, production timelines, or financial terms.

Google pays SpaceX $920 million per month for GPU rentals, and the post says the contract includes 11% GPU utilization, a 90-day cancellation clause, and methane gas turbines used to bypass environmental approval.

SpaceX disclosed a cloud services agreement with Google: Google will pay SpaceX $920 million per month for computing capacity tied to xAI data centers, while the post does not disclose contract duration, GPU scale, or delivery terms.

The title says Google will pay SpaceX $920 million per month for compute capacity at xAI data centers; the RSS snippet does not disclose contract duration, GPU scale, or the capacity delivery mechanism.

Google agreed to pay SpaceX $920 million per month for computing power under a cloud services deal running through mid-2029; the post does not disclose compute specifications, deployment regions, or service-level terms.

Elon Musk said at a JP Morgan fireside chat that SpaceX will pursue an IPO to fund more than 100,000 next-generation Starlink satellites and orbital AI data centers; the snippet also says Starship V4 targets over 200 tons of payload and a future launch cadence of once per hour.

Broadcom lost more than $300 billion in market value after its revenue forecast disappointed investors; its shares fell as much as 15% in after-hours trading, and the post does not disclose the specific revenue guidance.

NVIDIA and Microsoft announced a unified agentic AI deployment stack at Build across Windows, Azure, and local environments; RTX Spark provides 1 petaflop of AI performance, while DGX Station for Windows offers 20 petaflops of FP4 performance and up to 748GB of coherent memory.

Alphabet plans to raise about $80 billion for AI compute expansion through underwritten shares, mandatory convertible preferred stock, a $10 billion Berkshire private placement, and a $40 billion ATM program, with about $30 billion tied to employee equity taxes.

Motern AI completed a nearly 1 billion yuan Series C round and plans to release its SparsePrime inference card this year; the article says its S30 and S40 cards achieved three consecutive wins in MLPerf Inference.

Wirescreen reviewed 3,800 procurement records and found more than 500 cases where Chinese military units sought Nvidia chips by name or specification. The records cover 2019 to 2025 and include A100, A800, H100, and H800, but they do not confirm final delivery.

At least seven Chinese universities supporting the country’s armed forces and defense industry are seeking access to Nvidia H200 chips; the post does not disclose procurement channels, volumes, or specific US licensing conditions.

Bloomberg says at least seven Chinese universities that support China’s armed forces and defense industry are seeking Nvidia H200 chips, based on a review of procurement records; the RSS snippet does not disclose order volumes, suppliers, or procurement status.

Axios says Groq is seeking $650 million in internal funding while shifting from hardware toward AI inference, after Nvidia’s reported $20 billion not-acqui-hire; the RSS snippet does not disclose Groq’s valuation, investor names, deal structure, or fundraising timeline.

Apollo and Blackstone are seeking additional investors for about $36 billion in debt financing for Anthropic’s AI infrastructure; the title says the deal would buy Google chips, while the post does not disclose chip models, purchase volume, or timeline.

Cognition raised a $1B Series D at a $26B valuation and projects more than $1B ARR by year-end; the post says its valuation rose 2.5× from the $10B Series C eight months earlier, while the rest of the issue summarizes agent, inference, benchmark, and multimodal AI updates from May 26–27, 2026.

Snowflake signed a five-year, $6 billion AWS deal to secure chips for AI use; the post does not disclose chip models, delivery timing, or whether the capacity targets training, inference, or both.

OpenRouter raised a $113 million Series B led by CapitalG, lifting its valuation to $1.3 billion; the platform processes 25 trillion tokens per week, about 100 trillion per month, and provides one API for more than 400 models.

Latent Space says Fireworks is in talks for a $15 billion valuation round, Baseten is raising at an $11 billion valuation, and OpenRouter closed a $113 million Series C after volume grew 5x in six months.

The title says Qualcomm and ByteDance reached an AI ASIC chip partnership with procurement in the millions of units; the post does not disclose chip specifications, unit pricing, delivery timing, or production conditions.

Fireworks AI is in talks to raise a new funding round at a $15 billion valuation, according to people familiar with the matter; the post does not disclose the round size, investor names, or timeline.

Qualcomm will supply chips to ByteDance for artificial intelligence data centers, according to people familiar with the matter; the post does not disclose chip models, order volume, pricing, or delivery timing.

OpenRouter raised a $113 million Series B led by CapitalG; its weekly volume rose from 5 trillion to 25 trillion tokens over the past 6 months.

Greg Brockman recounted OpenAI’s 72-hour crisis: on November 17, 2023, the board removed Sam Altman as CEO and took Brockman off the board, after which Brockman resigned the same day and said he initially put the chance of taking the company back at 10%.

Mistral AI said it has reached a definitive agreement to acquire Emmi AI, a physics AI pioneer, to strengthen its position as an AI transformation partner for industrial companies. Austria-based Emmi AI works on physics AI and large engineering models that speed up engineering workflows, replace multi-day computations with real-time simulation and build digital twins. Emmi's co-founders and more than 30 researchers and engineers will join Mistral's Science and Applied AI teams in May.

DeepSeek-V4-Pro API will keep its promotional pricing from June 1, with cached input at RMB 0.025 per million tokens, while Bloomberg says DeepSeek is pursuing a RMB 70 billion round at a USD 45 billion pre-money valuation.

Latent Space summarized AI News for May 20-21, 2026, confirming TurboPuffer reached $100 million ARR and profitability, Exa raised a $250 million Series C at a $2.2 billion valuation, and Modal raised a $355 million Series C at a $4.7 billion valuation.

Trump approved Nvidia H200 sales to China six months ago, but Beijing has not allowed any company to buy even one chip and is steering firms toward domestic alternatives from Huawei and Cambricon.

SpaceX said in its S-1 filing that Anthropic agreed to pay $1.25 billion per month through May 2029 for access to Colossus I and II AI training centers in Memphis, totaling $15 billion annually.

Anthropic agreed to pay Elon Musk’s SpaceX nearly $45 billion over the next three years for computing resources to support its Claude AI software, according to a securities filing.

Anthropic will pay xAI $1.25 billion per month for compute; the post discloses the deal value but does not disclose compute scale, contract length, or deployment conditions.

Meta plans to cut about 8,000 jobs and move about 7,000 employees into AI-related roles, concentrating resources on AI infrastructure, foundation model development, and commercialization from model training to product work and profit generation.

A Blackstone-backed AI cloud group is set to receive a $5 billion investment to bring 500MW of data center capacity online next year; the post does not disclose the Google chip terms or deployment structure.