若台北盆地建9GW数据中心，每天排放到该地区的能量大约相当于23颗原子弹的威力，有可能2050年后没有冬天。(AI P-task)

aidj

犹他州拟建的人工智能数据中心可能会释放大量热量，一位物理学家将其比作每天引爆二十三颗原子弹。

该项目名为“平流层”（Stratos），计划建在犹他州博克斯埃尔德县。开发商表示，这座“超大规模”设施可能需要高达9吉瓦的电力——是整个犹他州用电量的两倍多。

但电力消耗只是问题的一部分。

据犹他州立大学物理学教授罗伯特·戴维斯称，该设施还会产生7至8吉瓦的废热。两者加起来，将向当地环境排放约16吉瓦的总热负荷。

这些热量不会分散到整个城市或地区，而是会集中在汉塞尔谷——一个位于大盐湖附近的高原沙漠盆地，那里空气稀薄，如同一个碗。

戴维斯计算得出，每天排放到该地区的能量大约相当于23颗原子弹的威力。 另一个对比：该数据中心的能耗相当于约 4 万家沃尔玛超级购物中心。

生态学家们担忧的是，周边环境已经不堪重负。大盐湖正在萎缩，裸露的湖床加剧了有毒粉尘问题，而该地区的干旱生态系统极易受到气温变化的影响。

戴维斯估计，该项目可能使当地气温在白天升高约 5 华氏度（2.8 摄氏度），夜间升高高达 28 华氏度（15.6 摄氏度）。

这种夜间升温可能会显著改变山谷中的植物、土壤、野生动物以及用水需求。

数据中心是人工智能的物理支柱。每一次聊天机器人回复、图像生成、云服务和模型训练都依赖于电力、冷却系统和土地。

现在的问题是，支撑人工智能的基础设施能否在不悄然改变其所在环境的前提下发展壮大。

资料来源：

《“比我想象的还要糟糕得多”：犹他州的“超大规模”数据中心可能在大盐湖附近造成巨大的热岛效应。——《盐湖城论坛报》

aidj

Here is a structured summary of the video **"The AI Data Center Boom Doesn’t Survive Basic Physics"** by Jovon (an optoelectronics engineer and researcher):

### 1. The Utah "Wonder Valley" Controversy

* **The Project**: A massive 9-gigawatt (GW) AI data center called "Wonder Valley" (or "Stratos") was unanimously approved in Box Elder County, Utah, championed by billionaire Kevin O'Leary [[00:00](https://www.youtube.com/watch?v=9lDc8NpU2c0&t=0)], [[00:44](https://www.youtube.com/watch?v=9lDc8NpU2c0&t=44)].
* **The Energy & Waste Heat Crisis**: Atmospheric physicist Dr. Robert Davies ran the thermodynamics numbers from the developer's filings [[03:13](https://www.youtube.com/watch?v=9lDc8NpU2c0&t=193)]. The facility will consume 9 GW of compute load powered by an on-site natural gas plant. At 55% thermal efficiency, the total continuous thermal load dumped into a closed desert basin equals **16 GW** [[04:28](https://www.youtube.com/watch?v=9lDc8NpU2c0&t=268)], [[04:44](https://www.youtube.com/watch?v=9lDc8NpU2c0&t=284)].
* **"23 Atomic Bombs"**: Davies calculated that this heat output is equivalent to releasing the energy of **23 Hiroshima-yield atomic bombs into the valley every single day** for its 30-year operational life [[05:09](https://www.youtube.com/watch?v=9lDc8NpU2c0&t=309)], [[05:31](https://www.youtube.com/watch?v=9lDc8NpU2c0&t=331)].
* **Environmental Impact**: This thermal load is projected to permanently alter the local microclimate, raising daytime temperatures by 2–5°F and nighttime temperatures by 8–12°F [[06:11](https://www.youtube.com/watch?v=9lDc8NpU2c0&t=371)], while demanding water infrastructure that defies physics to keep it cool [[07:12](https://www.youtube.com/watch?v=9lDc8NpU2c0&t=432)], [[07:31](https://www.youtube.com/watch?v=9lDc8NpU2c0&t=451)].

### 2. Hyper-Scalers and Financial "Delusions"

* **Unprecedented Capital Spending**: In 2026, the big five hyperscalers (Amazon, Alphabet, Microsoft, Meta, and Oracle) are projected to spend $660–$690 billion on capital expenditures, with roughly $450 billion going directly to AI infrastructure [[09:04](https://www.youtube.com/watch?v=9lDc8NpU2c0&t=544)].
* **Unsustainable Rebuild Cycles**: Unlike traditional infrastructure, AI data centers must be gutted and rebuilt every 3 to 5 years because each new generation of chips (e.g., Nvidia's Blackwell to Rubin) increases power density drastically, forcing expensive overhauls from air to liquid cooling [[13:14](https://www.youtube.com/watch?v=9lDc8NpU2c0&t=794)], [[13:35](https://www.youtube.com/watch?v=9lDc8NpU2c0&t=815)].
* **Circular Financing**: The video highlights massive circular loops in AI accounting, where tech giants invest billions into OpenAI, which then commits those billions back to the tech giants for compute reserve or chip purchases, artificially inflating revenue narratives [[28:55](https://www.youtube.com/watch?v=9lDc8NpU2c0&t=1735)], [[29:13](https://www.youtube.com/watch?v=9lDc8NpU2c0&t=1753)].

### 3. Supply Chain and Infrastructure Bottlenecks

* **The Turbine Shortage**: Because grids cannot support these massive data centers, companies are trying to build their own on-site natural gas generation [[11:13](https://www.youtube.com/watch?v=9lDc8NpU2c0&t=673)]. Consequently, lead times for combined cycle gas turbines have skyrocketed to 5–7 years, with order books closed until 2028 [[10:47](https://www.youtube.com/watch?v=9lDc8NpU2c0&t=647)].
* **The Chip Bottleneck**: Advanced AI accelerators rely on TSMC's CoWoS (Chip-on-Wafer-on-Substrate) packaging and High Bandwidth Memory (HBM) [[11:45](https://www.youtube.com/watch?v=9lDc8NpU2c0&t=705)]. These components are completely sold out through 2026, and manufacturing new Fabs takes 18 to 24 months [[12:09](https://www.youtube.com/watch?v=9lDc8NpU2c0&t=729)], [[12:55](https://www.youtube.com/watch?v=9lDc8NpU2c0&t=775)].
* **National Water Strain**: A peer-reviewed study from UC Riverside and Caltech reveals that US data centers will need 697 million to 1.45 billion new gallons of water per day by 2030—roughly adding the water demand of another entire New York City [[22:06](https://www.youtube.com/watch?v=9lDc8NpU2c0&t=1326)].

### 4. The Enterprise Demand Deficit and "DeepSeek" Disruption

* **No ROI for Enterprise**: An MIT study ("Project NANDA") found that despite $30–$40 billion in enterprise spending on generative AI tools, **95% of organizations have seen no measurable return on investment** [[20:17](https://www.youtube.com/watch?v=9lDc8NpU2c0&t=1217)].
* **OpenAI Internal Tension**: Wall Street Journal reporting revealed that OpenAI's CFO Sarah Friar is privately at odds with CEO Sam Altman, expressing that current revenue cannot support their $1.4 trillion infrastructure commitments [[16:52](https://www.youtube.com/watch?v=9lDc8NpU2c0&t=1012)], [[17:07](https://www.youtube.com/watch?v=9lDc8NpU2c0&t=1027)].
* **The DeepSeek Factor**: The economic model of the trillion-dollar buildout assumes high-premium pricing for AI tokens [[20:48](https://www.youtube.com/watch?v=9lDc8NpU2c0&t=1248)]. However, Chinese lab DeepSeek released models (V3/V4) that offer comparable performance at an 85% to 95% cheaper rate [[26:50](https://www.youtube.com/watch?v=9lDc8NpU2c0&t=1610)], [[28:00](https://www.youtube.com/watch?v=9lDc8NpU2c0&t=1680)]. This completely shatters the "defensible moat" of premium US frontier AI models [[28:40](https://www.youtube.com/watch?v=9lDc8NpU2c0&t=1720)].

### 5. Industry Insiders Admitting the Bubble

Jovon points out that top leaders have already quietly begun warning the public [[15:42](https://www.youtube.com/watch?v=9lDc8NpU2c0&t=942)]:

* **Sam Altman (OpenAI)** & **Jeff Bezos (Amazon)** have both publicly categorized the market environment as an "industrial bubble" [[15:57](https://www.youtube.com/watch?v=9lDc8NpU2c0&t=957)], [[16:04](https://www.youtube.com/watch?v=9lDc8NpU2c0&t=964)].
* **David Solomon (Goldman Sachs)** warned that deployed capital will turn out to "not deliver returns" [[16:12](https://www.youtube.com/watch?v=9lDc8NpU2c0&t=972)].

### Conclusion

Jovon notes that AI is objectively a real, revenue-generating technology [[31:34](https://www.youtube.com/watch?v=9lDc8NpU2c0&t=1894)]. However, just like the Dot-Com crash where the internet survived but thousands of overhyped businesses failed, the physical constraints of today’s power, cooling, supply chains, and debt-laden financing models mean the *current scale* of the AI data center buildout is mathematically and physically impossible to sustain [[36:12](https://www.youtube.com/watch?v=9lDc8NpU2c0&t=2172)].

*For more details, you can view the video here: [https://youtu.be/9lDc8NpU2c0*](https://youtu.be/9lDc8NpU2c0)

Engineer

台北盆地的数据中心应该设立在海边

aidj

aidj 发表于 4-6-2026 10:07 PM
Here is a structured summary of the video **"The AI Data Center Boom Doesn’t Survive Basic Physic ...

https://youtu.be/9lDc8NpU2c0 Grow the Flow

‘So much worse than I even thought’: Utah’s ‘hyperscale’ data center could create massive heat island near Great Salt Lake
https://www.sltrib.com/news/envi ... enter-could-create/

Utah’s fragile desert could feel like the Sahara if America’s biggest data center gets built
https://grist.org/business/utah- ... perscale-box-elder/

aidj

https://www.brockovichdatacenter.com/index.html