Anthropic IPO融资与算力采购协议规模盘点
On the bright side, capital flowing from other sectors + semis to a successful $…
On the bright side, capital flowing from other sectors + semis to a successful $ANT IPO...
从积极的一面来看,来自其他行业以及半导体行业的资金流向成功的 $ANT IPO……
will eventually flow back into semi balance sheets over the long run. Circle of life. but for semis.
长期来看,这些资金最终会回流到半导体公司的资产负债表上。生命之环,但适用于半导体行业。
If we look at Anthropic's $900.9B of eventual purchase agreements:
如果我们审视 Anthropic 高达 9009 亿美元的最终采购协议:
- $AVGO + financing partners: $161.2B in Google TPU leases - Google Cloud: at least $121.3B - SpaceX: Up to $120.5B (ceiling, can be cut) - Amazon: at least $110B - Microsoft: $47.2B-$49.2B - Nscale: up to $44.6B - Lambda: ~$41.3BN ($34.9B can be cut)
- $AVGO + 融资合作伙伴:Google TPU 租赁总额 1612 亿美元 - Google Cloud:至少 1213 亿美元 - SpaceX:最高 1205 亿美元(上限,可削减) - Amazon:至少 1100 亿美元 - Microsoft:472 亿至 492 亿美元 - Nscale:最高 446 亿美元 - Lambda:约 413 亿美元(其中 349 亿美元可削减)
With other identifiable counterparty deals such as:
还有其他可识别的交易对手方协议,例如:
- $MU (undisclosed), but supply agreement - $RIOT: ~$28B+ $9.1B - $WULF: ~$19B - Akami: ~$11.6B (CPU cloud) - Volta: ~$10B - OnePWR: ~$3.4B - $CRWV (undisclosed) - Everwind (undisclosed) - Likely $IREN (undisclosed) - Zerra DC (undisclosed)
- $MU(未披露),但有供应协议 - $RIOT:约 280 亿美元 + 91 亿美元 - $WULF:约 190 亿美元 - Akami:约 116 亿美元(CPU 云服务) - Volta:约 100 亿美元 - OnePWR:约 34 亿美元 - $CRWV(未披露) - Everwind(未披露) - 可能包括 $IREN(未披露) - Zerra DC(未披露)
Outside the $900.9B, there's another:
在 9009 亿美元之外,还有另一笔:
- Additional Google TPU capacity: $125B-$150B estimated
- $AMD: over $20B if Anthropic takes the full initial tranche of Helios capacity
- 额外的 Google TPU 产能:估计价值 1250 亿至 1500 亿美元
- 如果 Anthropic 全额获取 Helios 产能的初始批次,$AMD 将超过 200 亿美元
I'm actually personally longer term bearish on frontier lab valuations (I like OpenAI more if I had to pick one due to consumer stickiness). But raising a lot from IPO does help raise their counterparty status.
实际上,我个人对前沿实验室的估值长期持看空态度(如果非要选一个,我更喜欢 OpenAI,因为其用户粘性更强)。但通过 IPO 筹集大量资金确实有助于提升其交易对手地位。
Regardless of how they utilize it, that's a lot of GPUs, TPUs, HBM, optical networking, and compute needed for the next 5+ years.
无论他们如何运用这些资金,未来 5 年以上都需要大量的 GPU、TPU、HBM、光网络和计算资源。
更进一步:量化金融体系
看懂新闻只是起点——沿量化金融路径,把它变成能交付的工程能力