Why do so many people love calling things a "bubble"?

Every AI headline is about the grid. The grid is full. The grid can't keep up. The grid is the bottleneck. But what is the grid, actually? Who runs it? And why does a data center wait years to plug into it? I took it apart, step by step.

"Create an image of what the world would look like if I were in charge, based on my tweets."

SemiAnalysis's Jordan Nanos says CoreWeave is so backed up a customer with $100 million to spend was told to wait until May, so Nebius has been capitalizing "They're serving this middle tier of the market. CoreWeave has set the standard. I mean, they are the best in terms of technology, this reliability stuff I'm talking about. They just have so much data for monitoring." "I mean, these guys deploy 10,000 GPUs a week at the peak when they're building data centers. And so that's incredible. But like their balance sheet is full. There's only so many people that you can marshal and resources to build out another gigawatt when they're going to 1.5 and beyond." "So Nebius has really capitalized on this middle tier of the market, which is not small." "Like I said, this is customers who are coming to us and saying, hey, I saw you write such good things about CoreWeave. I just have $100 million. I'm trying to give it to them. They're telling me they can't accept it until May of next year because they're backed up. And therefore, Nebius has been capitalizing." ______ For more commentary on CoreWeave and Nebius: firesidealpha.substack.com/p…

People assume a data center hums because thousands of computers are running inside it. Almost none of the sound reaching the neighborhood comes from the computers. A GB300 NVL72 rack draws about 140 kW. Essentially all of it leaves the building as heat. Amazon’s campus on Tanner Way runs more than 400 rooftop exhaust fans, about 600 feet from the houses in Great Oak. Prince William County allows 55 dBA at night at a residential property line, and campuses are designed to come in under it. But here’s the catch: Fans and transformers make low-frequency tones. Low frequencies travel farther and go through walls. Counties measure in dBA, a scale that barely counts the low end. A 73 Hz hum reads about 24 dB quieter than it is. So a site can pass its noise limit while the people next to it still can’t sleep at night. How to fix this? Distance. Sound drops about 6 dB every time you double it. Low-frequency rumble carries farther than that. Enclosures and silencers. At xAI’s power plant in Southaven, Mississippi, a $7M sound wall went up last December, and the company promised a $12.8M earth berm as well, to screen the gas turbines. Slower fans. The fans have to get rid of the heat, you can’t just turn them down. Amazon put new motors and better blades on all 424 exhaust units at Great Oak, made the stacks 3 feet taller and rewrote the controls to run the fans slower and in sync. That cut the noise by about 10 dB, and the fans use less power doing it. It was a four-month retrofit of every fan on the roof. Amazon has never said what it spent.

How noisy are America’s largest data centers? I had my agents crank on this question for over a week. Together we investigated the 25 largest operating US AI campuses (according to Epoch), looking at satellite imagery, equipment, nearby homes, and noise complaints. I expected to dismiss resident concerns. I came away more sympathetic to the neighbors, but also convinced that these problems are avoidable.

How noisy are America’s largest data centers? I had my agents crank on this question for over a week. Together we investigated the 25 largest operating US AI campuses (according to Epoch), looking at satellite imagery, equipment, nearby homes, and noise complaints. I expected to dismiss resident concerns. I came away more sympathetic to the neighbors, but also convinced that these problems are avoidable.

Instructive to see a post like this get so much traction. B200 rental rates up 30% YTD. Hyperscaler cloud revenues tracking along. Remember: GPU rental rates and token prices were supposed to fall over time. The demand metrics this year have beaten all informed expectations.

On the AI front, the trade has been dead money for more than 3 months now. The metrics I am following (token expenditures and GPU lease rates) are all flat to down. The price of memory (DRAM) seems to be the only thing that is still going up. 🧵(1/2)

The tech sector isn’t just consuming energy – it’s also shaping supply It made up around 40% of all corporate power purchase agreements for renewables signed in 2025 Data centre deals alone covered about half of the sector’s current electricity use 👉 iea.li/3Qf9zxV

🚀 #NVIDIA’s rack-scale AI servers are surging: GB300, VR200, and VR300 NVL72 systems are projected to top US$710B in combined output value in 2027, up 214% YoY.💡#TrendForce on how this could boost ODM, power, and cooling suppliers: buff.ly/TqR3mWx 🔗

NVIDIA 2Q Revenue Surges 106% YoY; FY2028 Revenue Expected to Grow 70% NVIDIA announced its fiscal 2Q27 results on August 26, reporting record revenue of $96.22 billion, up 106% YoY. Data Center revenue reached $89.02B, surging 117% YoY and serving as the company’s primary growth engine. Looking ahead, NVIDIA expects third-quarter revenue to reach approximately $108B. CFO Colette Kress said during the earnings call that FY2028 revenue is expected to grow by around 70%, with the outlook constrained primarily by supply chain capacity rather than a lack of market demand. Vera Rubin Platform Enters Full Production Commenting on current market developments, NVIDIA founder and CEO Jensen Huang said AI has reached an inflection point, with the technology now capable of performing useful work and generating productive, economically valuable tokens. He emphasized that the global AI infrastructure buildout is advancing at full speed, while NVIDIA’s latest Vera Rubin platform has entered full-scale production. Huang added that companies will eventually operate multiple AI agents on a continuous basis, potentially driving a 15- to 100-fold increase in overall computing requirements. Gross Margins Face Downward Pressure Despite maintaining a strong gross margin of 75% in the second quarter, NVIDIA is expected to face mounting cost pressures. Kress noted that surging AI demand has tightened memory supply, with extreme pricing conditions exceeding the company’s previous expectations and likely to persist through next year. NVIDIA has therefore lowered its gross margin outlook. The company expects gross margin to decline to 74% in the third quarter and bottom out at 71–72% in the fourth quarter before recovering to 72–73% in the first quarter of FY2028. Huang Addresses Circular Financing Concerns NVIDIA’s substantial investments in frontier AI labs have raised market concerns over so-called “circular financing.” Kress acknowledged these concerns but said NVIDIA views these labs as once-in-a-century companies, with limited investment risk and attractive potential returns. AI labs requiring balance-sheet support are expected to account for approximately one-quarter of NVIDIA’s business next year. Huang added that NVIDIA provides a complete AI factory platform that is fundamentally different from custom inference chips designed for individual functions. He expressed confidence in NVIDIA’s technology and expects these AI labs to remain significant long-term users of the company’s computing architecture. 📊More semiconductor insights: buff.ly/YckK6fp 📊More financial charts: buff.ly/3AHP9WB

Could AI data centres really move into space? Orbital data centres could eventually help address growing power and cooling demands, but the technology still faces major technical and economic hurdles. For now, terrestrial infrastructure remains the clear investment priority. @WoodMackenzie forecasts US$9 trillion in global data centre investment by 2040. Robert Liew, Research Director at Wood Mackenzie, shares his perspective in The New Stack’s latest look at the future of orbital AI infrastructure. Read the full article: okt.to/JURuLh

Big Tech’s AI spending spree is much bigger than it looks. Nine major tech companies have roughly $3 trillion in off-balance-sheet commitments, far above reported capex ($600 billion). The AI arms race is creating enormous obligations investors may be underestimating.

Orbital compute will be the only way to scale AI probably sometime in 2029 due to power availability& permitting problems on land