News about data centers

The report values 3D integrated circuits at $16.8bn in 2024, with the memory segment projected to reach $22.2bn by 2030 as data centres need denser, more energy-efficient chips. Growth is driven by cloud computing, data centres, AI and machine learning, along with demand for miniaturised consumer devices and heterogeneous integration. The memory segment is forecast to grow at a CAGR of 19.2% and LEDs at 18.2%. The US market was about $4.6bn in 2024, while China is forecast to grow at a CAGR of 16.6% to roughly $6.7bn by 2030. The analysis also covers tariff risks, supply chain shifts and 26 companies, and names TSMC, NVIDIA, Micron and Amkor among those in the competitive race.

According to Reuters citing The Telegraph, Microsoft (MSFT) has withdrawn from an AI data center in Loughton, Essex that Nscale was building for it, though no reason was given. Grid officials reportedly told Nscale power would not be available until the 2030s, and Nscale says it remains committed to the site and is exploring other power sources. The site was planned to house about 23,000 Nvidia chips, roughly 11.5% of a program of about 200,000 chips. Meanwhile, Microsoft's cash capex in the June quarter reached $35.80 billion, more than double the year-earlier figure. The article argues the problem is grid timing rather than AI demand, and Microsoft reports results on October 28.

President Donald Trump struck a deal with Vladimir Putin for a fresh supply of Russian diesel to lower US fuel costs, but prices barely moved and the move drew a furious backlash from both parties. The announcement on October 9 came alongside a waiver relieving Russia from a new round of sanctions under a bill Trump had recently signed into law. Russian envoy Kirill Dmitriev said on October 16 that the US and Russia will soon announce an agreement on data center infrastructure, adding to political pressure on Republicans as data centers that have spread into suburban and rural areas draw anger over energy and water use ahead of the 2026 midterms. Time reminded Trump in a September 28 interview that 71% of Americans oppose building data centers and 57% say AI will do more harm than good. Trump defended data centers, saying they would lower taxes, while adding that he did not want them right next to his house.

The 5th GMIF Innovation Summit (GMIF2026) concluded in Shenzhen, co-hosted by the Shenzhen Memory Industry Association (SMIA) and the School of Integrated Circuits at Peking University, under the theme 'The Future Built on AI Memory and Storage.' As AI shifts from training to large-scale inference, tokens are becoming a unit of value, making memory and storage bandwidth, capacity and energy efficiency key to system performance and cost. Morgan Stanley's Daniel Yen projected that more than half of cloud data center capex will go to memory and storage (excluding HBM) by 2027, and identified six paths to overcome the 'memory wall.' Speakers from Peking University, Samsung, Sandisk, Solidigm, Arm and Silicon Motion said HBM remains central to high-end AI but is constrained by yield, cost and thermal limits, pointing to NAND-based high-bandwidth options, tiered storage and liquid cooling. The summit also highlighted that value-chain collaboration is becoming a deeper partnership built around workload requirements and joint scenario validation.

BlackRock is teaming up with Nvidia and other major financiers to build AI infrastructure financing platforms targeting over US$500 billion. The consortium plans to treat high-end AI compute capacity as a distinct asset class for large institutional investors, with BlackRock's credit desks expected to play a central role in structuring the vehicles. The article notes that the move fits BlackRock's existing focus on private markets and higher-fee infrastructure, which saw US$68 billion of private markets fundraising over the 12 months to Q3 2026 and operating margins of 47.8% in the same quarter. It says the key checkpoint for investors is whether AI-linked infrastructure and private credit commitments show up in that fundraising line and help sustain margins.

The 5th GMIF Innovation Summit (GMIF2026) was held in Shenzhen under the theme 'The Future Built on AI Memory and Storage', co-hosted by the Shenzhen Memory Industry Association and the School of Integrated Circuits at Peking University. Daniel Yen of Morgan Stanley projected that more than half of cloud data center capex will go to memory and storage (excluding HBM) by 2027, and outlined six innovation paths to overcome the 'memory wall'. Prof. Yimao Cai said HBM remains central to high-end AI compute but is limited by yield, cost and thermal constraints, pointing to NAND-based high-bandwidth technologies such as HBF as a major research focus. Samsung's Kevin Yoon said AI agents are generating massive KV cache demand, while Sandisk's Maya Zhang called this the first year of the 'inference boom', in which SSDs act as a 'token battery'.

Shares of neocloud operators CoreWeave (CRWV), Nebius (NBIS) and IREN edged up overnight after a week of selling driven by doubts over business models and higher financing costs. CoreWeave had fallen more than 20% in the early part of the week before gaining 6% on Thursday, supported by Nvidia CEO Jensen Huang. According to The Information, industry executives expect only seven to ten of hundreds of neoclouds to survive, with Nvidia supporting them through easier credit terms and $20 billion in leases. Bloomberg reported that Nscale faces opposition to its $30 billion IPO valuation, while CoreWeave's unsecured debt due in 2032 fell below 80 cents on the dollar, a level widely seen as a sign of financial strain.

American Electric Power (AEP) plans to invest about US$4 billion in a 702 MW Bloom Energy fuel cell plant near Cheyenne, Wyoming, under a 20-year power purchase agreement to supply Google's local power demand. The analysis argues the deal strengthens AEP's investment case, which relies on commercial and data center demand to justify its US$78b capital plan. Analysts expect earnings to rise from US$3.1b today to US$4.7b by 2029, with a P/E of 21.7x on those forecasts compared with an industry P/E of 19.9x. Key risks include reliance on large-load customers that often carry lower margins, rising interest costs, and a dividend that free cash flow does not fully cover.

AMD shares have sold off recently, even though the company has benefited from an AI chip shortage because Nvidia cannot meet all demand. The article argues that valuation is a bigger risk than the AI business, with AMD's P/E at 139 and forward P/E at 72, several times higher than Nvidia, Broadcom and Taiwan Semiconductor. If AMD's P/E fell to Nvidia's current level, the stock could lose close to 80% of its value. Still, the valuation is seen as more of a threat to the stock than to the company, since data center AI chips made up 58% of AMD's latest-quarter revenue versus almost 93% for Nvidia. The author reduced AMD holdings for diversification but kept a substantial long-term position.

Micron is benefiting from a memory chip shortage driven by AI data center demand, with revenue rising nearly 380% year over year in its latest quarter. The author argues the company is reducing the cyclicality of its business through multiyear take-or-pay contracts covering about 35% of revenue through 2030, along with new production facilities expected in mid-2027 and late 2028. The stock trades at 5.3 times fiscal 2027 earnings estimates, while Wall Street projects 16% growth for 2028; the author instead assumes 30% growth and a 20 times multiple, implying a price of $4,590 versus today's $920. According to the article, the market has not yet priced these changes into the stock.

In August, Nvidia brought together Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR to set up financing platforms that could mobilize over $500 billion for AI infrastructure. The article argues that access to cheap, large-scale capital will increasingly decide who can compete in AI, giving the United States an edge through its deep capital markets. The five largest US hyperscalers are on track to spend $730 billion on capital expenditure in 2026, with up to a third expected to be financed by debt. Nvidia's model aims to turn compute into an asset that pension funds and insurers can own, though the article flags risks such as short-lived GPUs and unpredictable AI demand. It also contrasts Europe's large household savings, which are not yet well channeled into startups, with China, where as much as 80% of newly built computing capacity reportedly sits unused.

The data center buildout totals $10.3 trillion, surpassing the railroad, electrification and dot-com booms, according to a report by Columbia Business School professor Stijn Van Nieuwerburgh. He estimates its economic impact at nearly 4% of GDP, almost twice that of 19th-century railroads. Data center construction value has more than doubled to $85 billion a year, while other construction fell 8%, including a $29 billion drop in new houses and apartments. States such as Nevada are trimming tax breaks, and economist Paul Krugman argues the spending will reduce labor demand in the long run.

Asia's AI trade is under pressure as a selloff in US tech stocks raises worries that higher financing costs could slow the data center spending that supports regional chipmakers and suppliers. Oil's surge to US$110 a barrel has added inflation concerns, while US chip shares had their longest rout since 2008. A Goldman Sachs basket of Asian AI hardware stocks fell 11 percent last week, with Samsung Electro-Mechanics down 23 percent and Unimicron off 14 percent. CoreWeave's shares dropped 25 percent in a week as its bonds signaled distress, and Gartner warned that more than US$3.5 trillion of AI infrastructure spending could be impaired by 2030. Still, there is little sign so far that orders have weakened.

Investment in data centers, put at $10.3 trillion, is now the largest financial bet in US history, surpassing the late 1800s railroad boom in both scale and economic impact. A Columbia Business School report estimates the buildout's economic effect at nearly 4% of GDP, compared with more than 2% for railroads, which cost about $173 billion in today's dollars. Texas, Virginia, Ohio and Georgia lead projected spending, and some states are rethinking tax breaks as revenues shrink and energy demand rises. Data center construction value has more than doubled to $85 billion a year while other construction fell 8%, and economist Paul Krugman warns this crowds out housing and jobs. The article also notes that AI profitability has yet to catch up with the spending.

Data centre projects in America's $10 trillion AI infrastructure boom are facing delays, cancellations, blocks and weaker investor confidence. RegLayer counted at least 47 large projects delayed, withdrawn or blocked in 2026, while Bernstein estimated only about 35% of the planned US pipeline is credible. More than 1,000 communities have tried to introduce temporary bans or restrictions on AI infrastructure in 2026, and rising borrowing costs threaten financing for neocloud firms. CoreWeave's stock lost almost 25% in a week, and several IPOs, including SoftBank's SB Energy and Nvidia-backed Firmus, were delayed or shelved. Construction delays could let customers cancel: Cipher agreed to cover the first $359 million of cost overruns on a Texas project tied to Anthropic, and Anthropic could cancel its $45 billion Nscale agreement if targets are missed.

Nvidia shares have gained 899% over the past five years (as of Oct. 15), and November has been mostly positive this decade, with the only decline coming in November 2025, when the stock fell 13%. Data center revenue rose 117% year over year to $89 billion in fiscal Q2 2027, with hyperscalers contributing about $49 billion. The article argues the key risk to this pattern is hyperscaler capital spending, which depends on the momentum of their cloud businesses. Hyperscaler backlogs have exceeded $2 trillion, and four of the top five will report third-quarter results within two weeks. If cloud growth slows, hyperscalers may cut capex, which would weigh on Nvidia's data center business.

The column examines several claims in the 2026 midterm campaign. It notes Sen. Tom Cotton's argument that inflation has fallen from a record 9.1% in June 2022 to 3.4% in August, but stresses that the rate was 3.0% in January 2025 when Trump returned to office and is now moving in the wrong direction, with September's rate rising to 3.7%. It argues that Republican claims about Iran's hostility toward America dating from 1979 overlook the 1953 coup that overthrew Prime Minister Mohammad Mosaddegh, which the U.S. government acknowledged in 2013. The author also quotes Rep. French Hill saying a data center energy bill that never became law would only have required states to “consider” making utilities cover data center energy costs. The author adds that emails to Little Rock city directors, several of whom share one email address, have gone largely unanswered.

Over the six months to October 15, NextEra Energy and Duke Energy returned -13.29% and -6.77%, respectively. Both utilities stand to benefit from rising electricity demand, including from data centers, but the article says NextEra has more growth avenues through renewables, battery storage and Florida Power & Light (FPL), where large-load demand expectations rose from 6 GW to 8 GW by 2032 and its renewable and storage backlog reached about 35.1 GW. Duke has secured 7.8 GW of electric service agreements with data center customers and reaffirmed annual EPS growth of 5% to 7% through 2030, with second-quarter adjusted EPS rising to $1.43 from $1.25. Both companies face heavy capital spending and risks from interest rates, project delays and regulatory decisions. NextEra trades at 17.31x trailing and 17.73x forward earnings, while Duke trades at 17.19x and 15.95x.

Seagate (STX) closed at $690.81 on October 15 and its shares are up 206.48% over the past year, even though hard disk drives have been largely replaced by flash memory in laptops and phones. Hard drives remain popular with data centers because they are considerably cheaper per unit of capacity, which suits the large volumes of rarely read data that AI produced. Revenue rose 48.5% in the latest quarter and earnings grew 165.2%, driven by scarce supply, as only three companies make hard drives and Seagate told investors its data center capacity is fully allocated through this year. The article warns that prices could fall once supply catches up, and notes debt is 179.88% of equity. The stock trades at 25.71 times next year's expected earnings, against 66.17 times over the past twelve months.

Oracle shares have fallen 53.84% over the past year, yet 80% of covering analysts still rate the stock a buy, with a median target 78.48% above the current price, according to CNN. The article notes revenue grew almost a third in the latest quarter and earnings grew faster, though that does not explain the drop. Analysts are weighing contracts to supply computing capacity worth $638 billion, far above annual revenue, while the market is focused on cash flow: capital spending of $75.66 billion over the past year exceeded $46.94 billion of operating cash, a gap funded by borrowing. The article frames the core question as whether returns on this spending arrive before the debt becomes a problem. The shares trade at 17.76 times expected next-year earnings.

Chestermere-Strathmore MLA Chantelle de Jonge, ahead of an Alberta referendum and amid AI backlash and US trade tensions, says the provincial government remains focused on responsible resource development, job creation and strengthening Alberta's economy within Canada. She said she ran on an Alberta First platform because of frustration with federal policies, but was not elected on a separatist mandate, and that Albertans should be able to voice their views through a democratic process. She cautioned against escalating the trade dispute, saying tariffs and counter-tariffs harm both Americans and Canadians, and urged diplomacy. On AI and data centres, which have prompted town halls in communities such as Langdon and Indus, she said Alberta is developing consistent province-wide standards so large data centres are built far from residential areas. She also noted ground was recently broken on the Dawson's Landing school in Chestermere with Rocky View Schools.

MarketBeat's stock screener lists five real estate stocks to watch today: Blackstone (BX), Crown Castle (CCI), Equinix (EQIX), Prologis (PLD) and Digital Realty Trust (DLR), which had the highest dollar trading volume in recent days. The article notes that real estate stocks, including REITs, may offer dividend income and price appreciation but are affected by interest rates, property values, economic conditions and market cycles. Blackstone is an alternative asset manager, Prologis is a global leader in logistics real estate, and Digital Realty Trust operates data center, colocation and interconnection solutions for industries including AI, cloud and financial services.

JPMorgan CEO Jamie Dimon warned that rising borrowing costs in 2026 mean many companies may not be ready, especially when they must refinance or borrow and face higher credit spreads. He argues that tech firms borrowing heavily to build AI data center infrastructure are competing with the U.S. Treasury for funds, pushing rates higher for households with mortgages, car loans and credit cards. The Joint Economic Committee reported that national debt hit a record $40.25 trillion at the start of October. Citing Bloomberg, JPMorgan strategists found deeply distressed loans rose 62.5%, from $40 billion in 2025 to $65 billion in the third quarter, the highest since March 2020. Freddie Mac data showed the 30-year fixed mortgage averaging 7.40% and the 15-year at 6.73% as of October 8, 2026, and the article urges readers to review and pay down their debts early.