AI训练算力AI training compute

训练AI模型的计算资源

每条都摘自下方编号的来源原文;只写事实,不写预测或建议。

技术简介

  • Wayve的训练流程包括收集真实驾驶经验、训练和评估,并仅部署优于在用版本的模型。1
    "Our AI Driver gets better by running this data flywheel: gathering real-world driving experience, training on it, evaluating it, and deploying only what outperforms what is already on the road," said Silvius Rus, SVP of Engineering at Wayve.
  • OCI高性能GPU集群提供AI训练和推理、数字孪生及大规模并行高性能计算所需算力。3
    This integration enables developers to access OCI’s high-performance GPU clusters and the scalable compute needed for AI training and inference, digital twins, and massively parallel HPC applications.

已有应用

  • Wayve将使用配备英伟达Blackwell GPU的CoreWeave云训练基础模型,并扩展训练、评估及仿真基础设施。1
    Wayve will train its foundation models on CoreWeave Cloud with NVIDIA Blackwell GPUs, expanding its AI infrastructure for training, evaluation, and simulation.
  • Meta的Muse模型和智能体使用英伟达基础设施。2
    Meta’s Muse models and agents, which leverage NVIDIA’s infrastructure, have delivered frontier-level benchmark performance and surging consumer downloads since launch.
  • Devin的训练、强化学习及生产推理运行于CoreWeave。4
    Cognition, the applied AI lab behind Devin, runs training, reinforcement learning and production inference for Devin on CoreWeave.
  • Almawave使用OCI基础设施和英伟达Hopper GPU训练及运行Velvet多语言生成式AI模型。3
    Almawave, an Italian AI leader, is using OCI AI infrastructure and NVIDIA Hopper GPUs to run training and inferencing workloads for Velvet, a family of multilingual generative AI models.

为何此时

  • FT在2026年10月报道,美国收紧出口管制后,数据中心租约价格、期限和预付比例均上升。5
    This has led to increased prices, longer terms and higher upfront payments for these data centre lease agreements.
  • 路透转述FT称,该租约原可让腾讯使用约10万颗因华盛顿和北京限制而无法在中国获得的先进AI芯片。6
    The deal would have given Tencent access to about 100,000 advanced AI chips that were not available in China due to curbs from Washington and Beijing, according to the report

主要参与者

  • CoreWeave为Wayve提供搭载英伟达Blackwell GPU的基础模型训练云。1
    Wayve will train its foundation models on CoreWeave Cloud with NVIDIA Blackwell GPUs, expanding its AI infrastructure for training, evaluation, and simulation.
  • 英伟达与Meta宣布扩大Muse及Muse系列模型的训练和推理工作。2
    NVIDIA and Meta today announced they will partner to expand their work on Muse and the Muse family of models to improve inference and training performance, reflecting popular demand for useful AI agents.
  • 甲骨文的OCI提供可扩展至131072颗英伟达GPU的液冷GB200 NVL72系统。3
    For example, OCI now offers liquid-cooled NVIDIA GB200 NVL72 systems on OCI Supercluster that can scale to up to 131,072 NVIDIA GPUs.

相关卡片

来源

  1. CoreWeave · CoreWeave to Power Wayve's AI Driver Training in Multi-Year Agreement
  2. 英伟达 · NVIDIA and Meta to Expand Collaboration on Muse and Muse Family of Models Amid Robust Growth · 2026-10-26
  3. 甲骨文 · Oracle and NVIDIA Help Enterprises and Developers Accelerate AI Innovation · 2025-06-12
  4. CoreWeave · CoreWeave Delivers NVIDIA Vera Rubin NVL72 Performance at Production Scale, Starting With Cognition · 2026-09-30
  5. app003.ftchinese.com · China’s Tencent leases 100,000 chips from Oracle to accelerate AI push
  6. investing.com · Oracle abandons $7 billion Tencent AI deal under Trump pressure, FT reports By Reuters · 2026-11-04