-- Context for this story is provided by "Spending on AI Is at Epic Levels. Will It Ever Pay Off?", Eliot Brown and Robbie Whelan, WSJ1, 9/25/25
*** 1. Spending on AI Is at Epic Levels. Will It Ever Pay Off?
This is a combined summary of two articles, one covering the broad U.S. data center build-out and the other focusing on OpenAI’s massive expansion plans.
Tech companies are pouring unprecedented sums into AI data centers, chips, and energy. Spending over just three years rivals the inflation-adjusted cost of the entire interstate highway system.
-
Microsoft, Meta, and others are committing hundreds of billions.
-
Facilities like Ellendale, N.D., now carry $15B+ price tags.
b) Speculative business model
Revenue lags far behind capital spending, raising doubts about returns. Analysts warn of parallels to the dot-com bubble, when telecom firms overbuilt fiber networks.
-
AI revenue in 2023 was ~$45B against trillions invested.
-
Many consumers still use free AI services, limiting monetization.
Firms use branding and scale to showcase ambition. Zuckerberg promotes “Hyperion,” while OpenAI’s “Stargate” envisions $1T+ in build-out. Oracle alone will receive $60B annually from OpenAI.
-
Names signal ambition but risk overhype.
-
Echoes of 1990s telecom optimism.
Middlemen like CoreWeave lease facilities filled with Nvidia chips, backed by debt-heavy financing. CoreWeave alone carries ~$15B in debt and $56B in lease obligations.
-
Loan rates often exceed 8%.
-
Short-term contracts vs. long-term debt creates major risk.
Experts cite “collective hallucinations” that overlook weak demand signals. MIT and Chicago studies show minimal business productivity gains from AI.
-
AI models like ChatGPT-5 seen as incremental, not revolutionary.
-
Rapid chip obsolescence magnifies risk compared to past bubbles.
Executives argue AI could add 10% to global GDP and replace large swaths of white-collar work. Backers liken the boom to electricity’s rollout or the internet in the 2000s.
-
700M+ weekly ChatGPT users support optimism.
-
Oracle’s Ellison calls training a “multitrillion-dollar market.”
Boomtown effects are reshaping small towns like Ellendale. Housing shortages, infrastructure loans, and new jobs show how high-level bets ripple into local communities.
-
Construction doubles town populations.
-
Long-term viability depends on AI demand holding.
The U.S. government and financiers like Blackstone and SoftBank are underwriting expansion, betting AI will pay off like past infrastructure revolutions.
-
$400B+ hyperscaler capex expected in 2025.
-
Many see this as wartime-scale mobilization.
-
6,000+ construction workers onsite daily.
-
Scale compared to Central Park, with room for expansion.
OpenAI is collaborating with Oracle and SoftBank to build additional sites across the U.S., targeting nearly 7GW of power capacity.
-
Oracle to build 5.5GW at three major sites.
-
SoftBank adds 1.5GW in Texas and Ohio.
Executives estimate eventual demand of 20–100GW, implying $1T–$5T in investment—exceeding the GDP of major economies. Altman admits financing models remain unclear.
-
1 GW costs ~$50B.
-
Demand could dwarf today’s global software market.
While Texas leaders welcome the investment, residents in Abilene express mixed views over water and power use. Once built, the site will support 1,700 permanent jobs—far fewer than during construction.
-
Promises of re-industrialization vs. limited long-term employment.
-
Mayor highlights both heritage and openness to progress.
- Text WSJ
-
Investment will be progressive, tied to each gigawatt deployed.
-
The first phase uses Nvidia’s unreleased Vera Rubin platform, expected to double the power of current Grace Blackwell chips.
b) Circular financing and infrastructure boom
OpenAI will use Nvidia’s cash to buy Nvidia’s own chips, a circular arrangement common in AI but controversial for blurring real demand vs. recycled capital. Dozens of new data center clusters will be needed, adding to one of the largest construction booms in modern history.
-
Builds on OpenAI’s $300B Oracle deal and Stargate venture with SoftBank.
-
Helps reduce debt costs that startups usually face in capital-intensive projects.
Shares of Nvidia rose nearly 4% to a $4.5T valuation, while Oracle’s surged on its own OpenAI deal. The investments underscore OpenAI’s central role in the AI boom. Still, skeptics note slowing model progress—such as backlash to GPT-5—and warn that expectations may outpace reality.
-
Some experts question whether “superintelligence” is achievable on current timelines.
-
Altman himself has warned that overexcited investors will “get burned.”
For Nvidia, the deal locks in massive future chip demand while extending its dominance in AI hardware. For OpenAI, it secures critical financing and strengthens ties with leading infrastructure partners. Both companies frame the project as a path to AI “superintelligence.”
-
Aligns with Nvidia’s broader spree: $5B into Intel and $2.5B into U.K. AI projects.
-
Reinforces OpenAI’s reliance on Nvidia, Microsoft, Oracle, and SoftBank for capacity.
- Text TechCrunch
Alibaba will integrate Nvidia’s Physical AI software into its Cloud Platform for AI, focusing exclusively on applications that control physical objects such as robots, self-driving cars, and smart industrial spaces. This partnership does not involve large language models.
-
Nvidia’s tools generate 3D replicas of real-world environments to create synthetic training data.
-
Use cases include robotics, autonomous vehicles, and connected spaces like factories and warehouses.
The deal strengthens Alibaba’s push into AI while complementing its e-commerce core. The company is ramping up AI investment beyond its $50B budget, expanding its global data center footprint, and launching new facilities in Brazil, France, and the Netherlands.
-
Alibaba now operates 91 data centers across 29 regions worldwide.
-
Marks a significant alignment with Nvidia as it scales AI infrastructure.
While this partnership covers only physical AI, Alibaba is simultaneously advancing in language AI. It unveiled Qwen 3-Max, its largest LLM yet, with 1 trillion parameters optimized for coding and agentic applications.
-
Shows Alibaba’s dual-track strategy: Physical AI via Nvidia and LLM development in-house.
-
Highlights the company’s ambition to compete globally across multiple AI domains.
No comments:
Post a Comment
Your comments will be greatly appreciated ... Or just click the "Like" button above the comments section if you enjoyed this blog note.