US utilities are making AI data center developers prove projects are real, using higher rates, collateral, longer contracts and study fees to weed out “phantom” demand, since grid costs are shared. Developers have requested 1,066GW of connections, but Wood Mackenzie expects utilities to serve only about 28%. Exelon’s high-probability data center load fell from 18GW to about 11GW after it required binding agreements, yet its $41 billion capital plan is unchanged. PJM’s market monitor said 4,017MW of forecast data center load had been canceled or delayed, and households may bear the costs. Average minimum contract lengths for large customers have risen to about 10.8 years from 4.6 before 2024.
Summary written by Upside AI from the original article. Read the source for full details.
Original headline: How do you know what real ?: US utilities are making AI data centers prove they exist before handing over power connections as 50 % of planned buildouts are currently being delayed , stalled , or canceled
Read the original on techradar.com ↗
Keywords
data centerselectricityartificial intelligenceNVDA
Open a keyword to see other news on the same topic.
Related tickers and likely impact
NVDA
▼ Likely negativeNvidia
Relevance: medium
| Market | US stocks |
|---|---|
| Sector | Information Technology |
| Industry | Semiconductors |
| Market cap | $5.2T |
| Close on 10/20/2026 | 216.48 |
| Change that session | -0.28% |
AMZN
• NeutralAmazon
Relevance: low
| Market | US stocks |
|---|---|
| Sector | Consumer Discretionary |
| Industry | Broadline Retail |
| Market cap | $2.7T |
| Close on 10/20/2026 | 262.76 |
| Change that session | +0.12% |
GOOGL
• NeutralAlphabet Inc. (Class A)
Relevance: low
| Market | US stocks |
|---|---|
| Sector | Communication Services |
| Industry | Interactive Media & Services |
| Market cap | $4.3T |
| Close on 10/20/2026 | 340.24 |
| Change that session | -1.14% |
META
• NeutralMeta Platforms
Relevance: low
| Market | US stocks |
|---|---|
| Sector | Communication Services |
| Industry | Interactive Media & Services |
| Market cap | $1.9T |
| Close on 10/20/2026 | 694.44 |
| Change that session | -1.56% |
MSFT
• NeutralMicrosoft
Relevance: low
| Market | US stocks |
|---|---|
| Sector | Information Technology |
| Industry | Systems Software |
| Market cap | $4.0T |
| Close on 10/20/2026 | 532.13 |
| Change that session | -0.26% |
Impact is scored by AI from the headline and may be wrong. It is not investment advice.
Related news
- Tech companies need more data centers. So why are nearly half of all planned projects delayed or canceled?
Global hyperscale spending is projected at about $700 billion, yet many U.S. data center projects are running late: according to the Sandstone Group, nearly half of planned projects are behind schedule, while industry research puts 30% to 50% of the 16 GW of U.S. capacity due online at risk of severe delays or cancellation. Allianz Research cites lengthy permitting, a shortage of roughly 439,000 skilled construction workers, and electrical equipment deficits as key reasons. In markets such as Northern Virginia, Phoenix and Dallas, waits to clear grid interconnection queues routinely run four to seven years. Oracle and OpenAI also scrapped plans to expand the 1,000-acre Crusoe Energy-developed AI campus in Abilene, Texas, from 1.2 GW to 2.0 GW, after negotiations stalled over financing and utility timelines.
- Tech companies need more data centers . So why are nearly half of all planned projects delayed or canceled ?
- Tech companies need more data centers . So why are nearly half of all planned projects delayed or canceled ?
- Tech companies need more data centers . So why are nearly half of all planned projects delayed or canceled ?
- Jensen Huang: “Two Gigawatts to Us Is Like $80 Billion.” 43 New Gigawatts Coming in 2027
Nvidia CEO Jensen Huang said two gigawatts of AI capacity, like the ones Nvidia announced in Australia, are worth about $80 billion to the company, up from about $18 billion per gigawatt previously. He attributed the increase to Nvidia now selling far more of each data center, including networking, CPUs, storage and security, not just GPUs. Citing SemiAnalysis's Dylan Patel on the BG2 podcast, Altimeter Capital's Brad Gerstner said 43 gigawatts of new AI compute will come online next year. At about $40 billion per gigawatt, Nvidia would need around 17 gigawatts to reach the $692 billion revenue estimate for fiscal 2028, though Huang said land, power and data center buildings are the main obstacles to the build-out.
- Amazon stockpiles Nvidia AI hardware ahead of power availability -Business Insider
Amazon has agreed to buy billions of dollars of Nvidia GPUs before its data centers have enough power to run them, according to internal August emails obtained by Business Insider. Under a new Early Rack Receipt program, the company is taking delivery of about $5 billion in Nvidia GB300 server racks at U.S. facilities, with plans to accept another $3 billion each month. Because several AWS sites lack the utility capacity to energize servers, some hardware sits idle in warehouses as dark GPUs, tying up capital before it generates cloud revenue. The move comes as Amazon projects roughly $220 billion in capex this year and posted negative free cash flow last quarter for the first time since 2023. An Amazon spokesperson called the staging a best business practice for rapid installation once power arrives and disputed reports that the company lacked a clear path to bring the sites online.
Related markets
Latest news
- Bitcoin and ethereum prices today, Wednesday, October 21, 2026: Crypto to outperform stocks and gold?
Bitcoin opened at $82,940.14 on October 21, 2026, up 1.3% from Tuesday's open, and rose to $83,292.55 by 7:21 a.m. ET. Ethereum opened at $2,446.67, up 1.7%, and moved to $2,459.92 that morning. Prices have moved little over the past week and a half, though JPMorgan Asset Management expects Bitcoin to outperform stocks and gold over the long term. BlackRock has separately filed for another Bitcoin ETF aimed at investors worried about sharp losses.
- Nvidia's Jensen Huang wants to build AI factories in weeks, not months
Nvidia CEO Jensen Huang, speaking at Nvidia GTC Berlin, said the new DGX Pod is designed to build AI factories faster and more cheaply, potentially cutting deployment from months to weeks. He said a gigawatt of data center costs about $50 billion, and the 4 to 6 months from shipping components to operation could be shortened by 3 to 4 months if more manufacturing happens in the supply chain, reducing idle cash. Huang argued this would make financing easier because lenders would have more certainty about costs and when revenue starts, which matters amid market volatility over neocloud companies. He said demand is very strong, citing SpaceX, CoreWeave, Nebius and Nscale, and that the large capital tied up before revenue is the key challenge to solve.
- Supermicro Brings NVIDIA Aerial RAN Computer Solutions to the Network Edge with Nokia to Enable Low-latency AI Services
Super Micro Computer (SMCI) is working with Nokia to deploy AI-RAN from distributed cell sites to centralized network hubs, combining Nokia's RAN technology, the NVIDIA Aerial RAN Computer platform, and Supermicro infrastructure. The solution lets radio and AI workloads share the same infrastructure, supporting low-latency applications such as voice AI, vision AI, and physical AI. The companies say it can improve spectral efficiency, automate network operations, and optimize energy use, with a software-driven path toward AI-native 6G. Supermicro's dual-node ARC-Pro, with an NVIDIA Grace C1 CPU and RTX PRO 4500 GPU per node, delivers up to 1.6 PFLOPS of FP4 AI performance.
- SUSE Brings Sovereign Control to Enterprise Agentic AI with NVIDIA
SUSE introduced new capabilities that let enterprises adopt private AI while keeping full operational control, security, and governance over their systems, through SUSE AI Factory with NVIDIA. The release adds zero-trust agent security via the NVIDIA OpenShell Reference Architecture, with supervisor proxies inspecting agent traffic and blocking unauthorized requests. SLES for NVIDIA gives immediate access to NVIDIA's latest hardware on a SLES 16 base. SUSE is a launch partner for the OpenShell architecture, and a government Secretariat is testing the platform for AI-driven public services.
- Tesla and SpaceX Just Formed Special Committees to Weigh a Merger
Elon Musk said on SpaceX's October 20, 2026 earnings call that both companies have formed special committees of independent directors to study a possible merger, and he cannot join those talks because he runs both firms. He said a deal is not guaranteed, the committees have not agreed on terms, and Tesla would need approval from 50% of all shares, which he called a high hurdle. Tesla closed at $373.71 and SpaceX at $147.79, but no price or exchange ratio has been disclosed. The article also points to Tesla's weak standalone results, including Q2 non-GAAP EPS of $0.33 against a $0.5367 estimate, and says the next concrete signal would be an 8-K filing naming committee members, advisors or terms.
- Musk backs SpaceX-Tesla merger idea as both boards form special committees
Musk said on SpaceX's third-quarter earnings call that he thinks a SpaceX-Tesla merger would be a good idea, and that both companies have formed special committees of independent directors at shareholders' request. He cannot join those talks as a conflicted party, and he stressed a merger is not a certainty. Barclays analyst Dan Levy said a tie-up has become the consensus expectation, and that Tesla's annual meeting, which must be held by Dec. 6, is the appropriate venue for a shareholder vote, though a delayed announcement could push a deal out 6 to 12 months. At values of about $2.1 trillion for SpaceX and $1.5 trillion for Tesla, Levy estimated Musk would own about 37% of the combined company and 56% of its vote, and each Tesla share would be worth about 0.7 SpaceX shares before any premium.
Get the briefing on Telegram or Zalo
Join the Upside Telegram channel or Zalo group for the Vietnam stocks and gold briefing at 8:00 and 19:00 (Vietnam time) on trading days, plus a weekly wrap on Saturday morning.