【帖子标题】：The Two Trillion Dollar Bottleneck: AI & Energy
两万亿美元的瓶颈：人工智能与能源困局
【帖子正文】：
For a few years now the hyperscalers have been spending agressively, but it seems to me like spending this money is actually the easy part of the equation.
多年来超大规模云服务商始终在激进投资，但在我看来，花钱反而是这个等式里最简单的部分。
If we add up the contracted, not yet delivered cloud demand (Remaining Performance Obligations - RPO) of the hyperscalers (Microsoft, Oracle, Google and Amazon), we get a sum of about $2.3 Trillion.
若将微软、甲骨文、谷歌和亚马逊等云巨头的未交付合同需求（剩余履约义务/RPO）相加，总额约达2.3万亿美元。
(We still don’t know if this will ever get paid but still RPO is a better estimate than a forecast & a big portion of it is from frontier AI labs like OpenAI and Anthropic)
（虽无法确定这些款项最终能否兑现，但RPO仍比预测更可靠——其中大部分来自OpenAI、Anthropic等尖端AI实验室）
In order to serve this massive RPO, MSFT, GOOGL, AMZN, and META are spending about $660 Billion a year, moving from asset-light software companies into owners of physical plants, real estate, and equipment.
为应对如此庞大的RPO，微软、谷歌、亚马逊和Meta每年投入约6600亿美元，正从轻资产软件公司转型为实体工厂、地产和设备的所有者。
The real battle for big tech isn’t buying servers, but rather securing 24/7 power.
科技巨头的真正战场并非采购服务器，而是确保7×24小时不间断供电。
That kind of energy commitment completely breaks traditional budgeting.
这种级别的能源需求彻底打破了传统预算模式。
As JPMorgan put it, “money can’t buy you electric power” when a project’s start date depends on a years-long grid-connection queue.
正如摩根大通所言，当项目启动取决于长达数年的电网接入排队时，“金钱买不来电力”。
Out of these hyperscalers, Oracle is the outlier, it carries net debt of about 3x operating cash flow (an order of magnitude above the others), and its free cash flow has gone negative. Amazon’s trailing free cash flow also crossed into the red.
其中甲骨文最为异常——净负债达运营现金流的3倍（比其他公司高出一个数量级），自由现金流已为负值。亚马逊的滚动自由现金流同样跌入赤字。
I have to say, these companies remain among the most cash-generative business ever built, and they’re funding roughly half the build from cash and the other half from debt markets.
必须承认，这些仍是史上现金流最强的企业，它们约半数建设资金来自现金储备，另一半则依赖债务市场。
T. Rowe’s Dom Rizzo argues the funding gap is “not that big” against balance sheets like these, and that unlike 1998 (the dot-com bubble) the fundamentals are still accelerating.
T. Rowe的Dom Rizzo认为，相较于如此强劲的资产负债表，资金缺口”并不算大”，且与1998年（互联网泡沫）不同，基本面仍在加速增长。
Now the interesting part is that if we follow the money trail down the supply chain, we can see exactly who is extracting the value.
有趣的是，若沿着资金流向追踪供应链，便能看清价值究竟被谁攫取。
It starts with the regulated utilities like Southern, Duke and Dominion who are accelerating capex to build grid capacity.
起点是南方电力、杜克能源等受监管的公用事业公司，它们正加速资本支出以扩建电网。
These guys face huge regulatory hurdles and political blowback from residential ratepayers over increasing electricity bills.
这些企业不仅面临巨大监管障碍，还因电费上涨遭遇居民用户的政治反弹。
To go around the regulations, NextEra for instance is using a hybrid strategy by restarting a nuclear plant (Duane Arnold) for Google.
为规避监管，NextEra等公司采取混合策略——例如为谷歌重启杜安阿诺德核电站。
Because the public grid is too slow, big tech are chasing independent producers like Constellation, Vistra, and Talen energy.
由于公共电网效率低下，科技巨头转而追逐星座能源、Vistra等独立发电商。
These companies have 24/7 power and they spend almost nothing to expand.
这些企业能提供全天候电力，且几乎无需投入扩建成本。
Amazon and Meta are locking up deals with Vistra, Microsoft contracted to restart the Three Mile Island in Pennsylvania, and Google signed for small modular reactors with Kairos.
亚马逊和Meta正与Vistra敲定协议，微软签约重启宾州三哩岛核电站，谷歌则与Kairos签约开发小型模块化反应堆。
One layer down, we arrive to the equipment makers like Vertiv, GE Vernova, Eaton and Caterpillar which are facing an increasing days inventory outstanding (DIO).
再往下游是Vertiv、GE Vernova、伊顿和卡特彼勒等设备制造商，其库存周转天数（DIO）持续攀升。
Usually this is looked at as constraint on these companies, but as the gross margins also keep on rising for some of these companies, even hitting a record for Vertiv and GE Vernova, this is pure pricing power for them.
虽然DIO增长通常被视为不利因素，但随着Vertiv和GE Vernova等公司毛利率持续创新高，这反而彰显了它们的定价权。
There are so many other layers that plays into this like, cooling systems, precision HVAC, advanced packaging & foundry equipment, copper mining & refining for power grids, etc etc. that won’t fit into this single post.
其他环节如冷却系统、精密暖通、先进封装/晶圆厂设备、电网用铜矿冶炼等难以尽述。
The takeaway is big tech is bleeding money and the wealth is pooling at the bottom with nuclear operators collecting and hardware manfucaturers hitting record margins while the the payers wait for delivery.
核心结论是：科技巨头正在失血，财富向产业链底层汇聚——核电运营商坐收渔利，硬件制造商利润率创新高，而付款方仍在苦等交付。
If you’re interested to read more, I have the full write up with plots & figures on my free substack:
若想了解更多，欢迎阅读我Substack上的完整分析（含图表）：
https://secaura.substack.com/p/the-two-trillion-dollar-bottleneck