The Wall Street Journal
WSJ:
Tech News Briefing
TUESDAY, MARCH 3, 2026
3/3/2026 3:01:00 AMShare This Episode
The Almost-Crisis That Struck Virginia’s Power Grid
バージニア州の電力網を襲いかけた“危機寸前”の出来事
Last year, roughly 40 data centers in Virginia suddenly dropped off the power grid, threatening to crash the system. A WSJ exclusive from reporter Katherine Blunt details this growing threat to power grids as companies across the country build infrastructure to power the AI boom. Plus, WSJ reporter Sean McLain joins to discuss Amazon’s strategy for catching up in the AI race. Peter Champelli hosts.
• roughly /ˈrʌfli/ およそ、約
• data center /ˈdeɪtə ˌsɛntər/ データセンター
• drop off the power grid /drɑːp ɔːf ðə ˈpaʊər ɡrɪd/ 電力網から外れる、電力系統から切り離される
• power grid /ˈpaʊər ɡrɪd/ 電力網、送電網
• threaten to ~ /ˈθretən tuː/ ～する恐れがある、～しかねない
• crash the system /kræʃ ðə ˈsɪstəm/ システムを崩壊させる、停止させる
• exclusive /ɪkˈskluːsɪv/ 独占記事、独占報道
• infrastructure /ˈɪnfrəˌstrʌktʃər/ インフラ
• AI boom /ˌeɪ ˈaɪ buːm/ AIブーム
• catch up in the AI race /kætʃ ʌp ɪn ði eɪ aɪ reɪs/ AI競争で追いつく
threaten to crash the system 「システムを崩壊させる恐れがある」、
build infrastructure to power the AI boom「AIブームを支える電力・設備インフラを構築する」、
catch up in the AI race「AI競争で遅れを取り戻す」
Peter Champelli: Welcome to Tech News Briefing. It's Tuesday, March 3rd. I'm Peter Champelli for The Wall Street Journal. Amazon sells what everyone wants for cheaper than anyone else. That's the formula that's rocketed the company to the top of retail and cloud computing. But can they use the same strategy in the AI race? And then early last year, Virginia's power grid was threatened when dozens of data centers suddenly dropped off. We're bringing you inside our exclusive reporting, which revealed for the first time that this dropout happened, and we'll explain why there's a growing risk that it could happen again.
But first, Amazon has fallen behind in the artificial intelligence race, and they're trying to catch up. So in December, they installed a new AI czar, Peter DeSantis, a widely respected Amazon veteran who's been at the company for nearly 28 years. In that time, he's spearheaded cloud computing and silicon chip making operations. But will it be enough to boost their offerings? The Journal's Sean McLain joins us now with more. Why is it significant that they've put Peter DeSantis in this AI role? You describe him as sort of a celebrity figure within the company.
Sean McLain: Peter DeSantis is symbolic of a strategy change going on in AI. It's really a shift from developing the tech behind their Nova AI models to now trying to speed up delivery and bring products to market, and also a change in how they're differentiating themselves in the market. So Peter DeSantis is known within Amazon as being the main guy behind launching a lot of the infrastructure that powers Amazon's cloud computing business and also its chief cheerleader and architect in their chip strategy as well. So Amazon under Peter DeSantis is trying to join both their AI efforts, their data center efforts, and their chip making efforts under one leader in the hopes that combining all those businesses will both speed up development and also speed up the delivery of products to customers.
Peter Champelli: DeSantis says he can make cheaper AI to sell to businesses looking for specific tasks rather than a general knowledge tool like ChatGPT. Break that down for me. What would that look like?
Sean McLain: DeSantis believes that right now, a lot of the AI services that companies are rolling out, there's a real cost problem to it in which the cost of those Nvidia chips, the cost of using a big model like Gemini or a big model like Claude will run up the cost to the point where that service really isn't profitable or worth doing. There's a lot of tasks that benefit from AI, but don't need the world's strongest or world's most cutting edge model to do well. So that's the strategy and the reason why Amazon is looking at shifting with AI development in a way that brings cost down.
Peter Champelli: At least one company said that the return on cost was better with Amazon's AI. Can you tell me about Nimbus Therapeutics? What was their experience like using Nova?
Sean McLain: So Nimbus Therapeutics needs a very specific thing. They need somebody to look at a bunch of potential molecules as they look for novel molecule designs that could possibly be turned into medicines, which then they could turn on and sell to drug companies. So what they found is that Nova responded just as well as a version of Anthropic's Claude, which is seen as a market leader at 1/10th the price.
• return on cost /rɪˈtɜːrn ɑːn kɔːst/ コストに対する収益・効果
• potential molecule /pəˈtɛnʃəl ˈmɑːlɪˌkjuːl/ 候補となる分子
• novel molecule design /ˈnɑːvəl ˈmɑːlɪˌkjuːl dɪˈzaɪn/ 新しい分子設計
• turn into medicines /tɜːrn ˈɪntuː ˈmɛdəsɪnz/ 医薬品にする
- revitalized focus on AI「AIへの新たな重点・再強化された注力」。
- turn on? and sell to drug companies 実用化できる形にして製薬会社に売る
turn around and sell to/turn it into and sell to->turn it into medicines and sell to drug companies（〜にして売る
Peter Champelli: Is this revitalized focus on AI and Amazon's plan for spending around it leaving investors confident?
Sean McLain: I think everybody's a little bit worried about how much money is going out the door towards building these data centers and developing these AI models. Everybody is, as investors would say, priced to perfection in this market. So if you're not executing, then your stock market will reflect that instantaneously. And I think at Amazon's case, the concern is how much money is going out the door this year in particular. So Amazon says they're going to spend $200 billion on capital expenditures this year, most of which is going towards building out AI infrastructure. And that is roughly what they spent in the preceding two years. And they're spending so much that they're probably going to burn cash at least in the first quarter. And that has investors worried. Is the business going to come in fast enough to justify how much money is going out the door to get ready for these AI customers?
• go out the door /ɡoʊ aʊt ðə dɔːr/（お金が）流出する
• data center /ˈdeɪtə ˌsɛntər/ データセンター
• develop AI models /dɪˈvɛləp eɪ aɪ ˈmɑːdəlz/ AIモデルを開発する
• priced to perfection /praɪst tə pərˈfɛkʃən/ 完璧な成長がすでに株価に織り込まれている
• capital expenditures /ˈkæpɪtəl ɪkˈspɛndɪtʃərz/ 設備投資（capex）
• build out /bɪld aʊt/（インフラなどを）拡張して整備する
• preceding /prɪˈsiːdɪŋ/ 直前の・前の-> 過去
• burn cash /bɜːrn kæʃ/ 現金を消費する
Peter Champelli: That was Wall Street Journal reporter, Sean McLain. Which AI models do you use the most? If you're a listener on Spotify, be sure to leave us a comment with your thoughts.
Coming up, power grids have gotten more strained as big data centers pull more and more energy, but sudden data center drop-offs might be the bigger threat. That's after the break.
Data centers draw massive amounts of energy. And in order not to overwhelm the power grid, operators need to figure out how to keep the supply of electricity and the demand for electricity flowing very stably. So what happens when the electricity in roughly 40 data centers, enough to power a million houses, drops off all at once? The Wall Street Journal's Katherine Blunt joins our Katie Dayton to explain.
Katie Day-ton: Katherine, what happened to this cluster of data centers in Virginia last year? Do we know exactly what went wrong at this point?
Katherine Blunt: Yeah. So it's fairly straightforward what happened. There was about 42 data centers that kind of went offline all at once in response to a problem on a nearby transmission line. Data centers are equipped with technology that monitor for disruptions and power quality to make sure that they can always operate and operate well and not have damage to the data center equipment. So during this brief disturbance, all the data centers responded to it by switching to backup power. And if one or two do that, that's not really an issue, but to have so many go offline all at once created sort of the prospect of instability on the broader power grid.
Katie Day-ton: So a lot of reporting on data centers so far has been focused on what happens if they take up too much electricity from the grid. I personally wouldn't have thought it would be an issue per se if they stopped using electricity. Can you explain why it's a problem when that happens?
• per se /pər ˈseɪ/ それ自体としては
Katherine Blunt: The issue here is that on the grid, electricity and supply and demand must be in constant balance because if there's either too much supply or too little supply to meet demand, then there is the risk of cascading failure across the systems, significant damage to power plant infrastructure. It is the role of the grid operator to make sure that that balance is always maintained. So in this particular case, the grid operators called PJM interconnection. And when these data centers all went offline at once, it was equivalent to 1800 megawatts. That's pretty significant. It's like equivalent to the output of a very large power plant. And so we understand it's a problem if there's not enough supply to meet demand, but if there's too much supply flowing under the grid and that demand is no longer there, you have the same sort of risk to the rest of the system and the infrastructure.
• in constant balance /ɪn ˈkɑːnstənt ˈbæləns/ 常にバランスが取れている状態
• cascading failure /kæˈskeɪdɪŋ ˈfeɪljər/ 連鎖的障害
• equivalent to /ɪˈkwɪvələnt tuː/ ～に匹敵する
• output /ˈaʊtpʊt/ 出力
Katie Day-ton: How big a deal was this for PJM and also the wider data center industry?
Katherine Blunt: Fortunately for everybody who uses the power grid, this wasn't that big of a deal. PJM has contingency plans in place for something like this happening. They'd had to take some measures to keep supply and demand and balance. It didn't exceed what they were able to do quickly in this particular case. But the concern that PJM and others have is that as more of these enormous data centers come online, what if more of them trip offline all at once? Because then that would really stretch the grid operator's ability to respond quickly enough to avoid more significant impact to the system. So this actually happened twice within PJM within a matter of months. There was another incident, a very similar one that happened in July of 2024, and then this occurred in February of 2025. So it's become an industry-wide issue and the subject of a lot of discussion of how do you make it so that these data centers stay online, even if there's a brief disturbance in power quality, which happens a lot.
• contingency plan /kənˈtɪndʒənsi plæn/ 緊急対応計画
• take measures /teɪk ˈmɛʒərz/ 措置を取る
• in balance /ɪn ˈbæləns/ バランスが取れている
• come online /kʌm ˈɔnˌlaɪn/ 稼働開始する
• trip offline /trɪp ˌɔːfˈlaɪn/ （システムが）突然停止する
• all at once /ɔːl æt wʌns/ 同時に
• stretch someone’s ability /strɛtʃ ˈsʌmwʌnz əˈbɪləti/ 能力の限界に近づける
• within a matter of months /wɪˈðɪn ə ˈmætər əv mʌnθs/ 数か月のうちに
• industry-wide issue /ˈɪndəstri waɪd ˈɪʃuː/ 業界全体の問題
• disturbance in power quality /dɪˈstɜːrbəns ɪn ˈpaʊər ˈkwɑːləti/ 電力品質の乱れ
ｰｰｰｰｰｰｰｰｰｰｰｰｰｰｰｰｰｰｰｰｰｰｰｰｰｰｰｰｰｰｰ 8：35
Katie Day-ton: And what sense do you get that this is a growing cause of concern for the wider tech industry?
what sense do you get that ～
「～という点についてどんな印象を得ていますか」という意味。
Katherine Blunt: It was just the subject of conversation. There was a sort of a regulatory body that had a conference that focused on this and other risks and assembled a lot of tech folks to sort of weigh in on the wonky technical side of all of this and see how the risk can best be mitigated. But it is a growing challenge across the entire country because you're seeing data centers being developed in many, many different regions, very large ones. And there's a lot of interest, I think, in trying to get ahead of the risk and to solve for it. But it's happened a few times already and there's concern that if it happens again with more facilities, that could be really challenging.
• growing cause of concern /ˈɡroʊɪŋ kɔːz əv kənˈsɜːrn/ 懸念が高まりつつある原因
• regulatory body /ˈrɛɡjəˌleɪtəri ˈbɑːdi/ 規制機関
• focus on /ˈfoʊkəs ɑːn/ ～に焦点を当てる
• assemble /əˈsɛmbəl/ 集める・招集する
• weigh in on /weɪ ɪn ɑːn/ ～について意見を述べる
• wonky /ˈwɑːŋki/ 技術的で専門的な（ややオタク的な）
• mitigate risk /ˈmɪtəˌɡeɪt rɪsk/ リスクを軽減する
• solve for /sɑːlv fɔːr/ （問題を）解決しようとする
Katie Day-ton: Do you get a sense of what that might look like, those solutions and what the industry has been discussing?
Katherine Blunt: Yeah. So Dominion Energy is a utility company that operates the transmission lines that were involved in the disturbance and the issues that we're talking about with the demand loss. And they've been working with tech companies to have specifications in order to hook up to the power grid, their data centers have to be calibrated to be able to withstand certain disturbances so that they're not tripping off unexpectedly. Basically, it's a technical problem to solve. How do you calibrate the sensors on the data center to be able to sort of what's known as ride through minor disturbances rather than immediately switching to backup?
transmission lines /trænzˈmɪʃən laɪnz/ 送電線
disturbance /dɪˈstɜːrbəns/ 電力の乱れ・障害
demand loss /dɪˈmænd lɔːs/ 需要の急減
specifications /ˌspesəfɪˈkeɪʃənz/ 仕様・技術要件
hook up to ～ /hʊk ʌp tuː/ ～に接続する
power grid /ˈpaʊər ɡrɪd/ 電力網
calibrate /ˈkælɪbreɪt/ 調整する、校正する
withstand /wɪðˈstænd/ 耐える
trip off /trɪp ɔːf/ （装置が）自動停止する
trip /trɪp/（機械・電気系）作動して停止する、遮断する
ride through /raɪd θruː/ （障害を）乗り切る・持ちこたえる
minor disturbances /ˈmaɪnər dɪˈstɜːrbənsɪz/ 小さな電力の乱れ
switch to backup /swɪtʃ tuː ˈbækʌp/ バックアップ電源に切り替える
outage /ˈaʊtɪdʒ/ 停電
Katie Day-ton: And for the average person, how bad could an outage like this be for them?
Katherine Blunt: If it happened so that the drop in demand damaged the rest of the system, you're looking at outages for everybody. It's hard to quantify exactly what that would look like, but in that region, power plants would trip offline. They would probably be subject to some sort of significant damage. And so to repair that would take a lot. That's extreme, but that's what could potentially happen if the drop in demand was not something that the grid operator could handle.
Peter Champelli: That was Wall Street Journal reporter, Katherine Blunt, joined by our Katie Dayton. And that's it for Tech News Briefing. If you're a listener on Spotify, be sure to leave us a comment. Today's show was produced by Anthony Bansie and Julie Chang, with supervising producer, Katie Ferguson. I'm Peter Champelli for The Wall Street Journal. We'll be back later this morning with TNB Tech Minute. Thanks for listening.
What’s News
WEDNESDAY, MARCH 4, 2026
3/4/2026 5:06:00 PMShare This Episode
The Growing Middle East Conflict Risks Drawing in the U.S.’s NATO Allies
拡大する中東紛争は、アメリカのNATO同盟国を巻き込むリスクがある
P.M. Edition for Mar. 4. The geography of the U.S.-Iran conflict is expanding: the U.S. shot down an Iranian missile fired at Turkey, and also sank an Iranian ship in the Indian Ocean. Plus, oil prices stabilized today but are still up about 15% this week. We hear from WSJ reporter Benoît Morenne about why American frackers aren’t taking this as their cue to increase supply. And a record number of Americans are tapping into their 401(k)s to pay for emergencies. Alex Ossola hosts.
• conflict /ˈkɑːnflɪkt/ 紛争、衝突
• risk drawing in /rɪsk ˈdrɔːɪŋ ɪn/ ～を巻き込むリスクがある
• NATO allies /ˈneɪtoʊ ˈælaɪz/ NATO同盟国
• geography /ʤiˈɑːɡrəfi/ 地理的範囲
• shot down /ʃɑːt daʊn/ 撃墜した
• missile /ˈmɪsəl/ ミサイル
• sank /sæŋk/ 沈めた（sink の過去形）
• stabilize /ˈsteɪbəlaɪz/ 安定する
• fracker /ˈfrækər/ フラッキング（シェール採掘）企業・業者
• cue /kjuː/ 合図、きっかけ
• increase supply /ɪnˈkriːs səˈplaɪ/ 供給を増やす
• tap into /tæp ˈɪntuː/ （資金などを）引き出して使う
• 401(k) /ˌfɔːr oʊ wʌn ˈkeɪ/ 米国の確定拠出型年金制度
• emergency /ɪˈmɜːrʤənsi/ 緊急事態
risk drawing in ～
risk + 動名詞で「～するリスクがある」。
drawing in は「巻き込む」。
→ 「同盟国を紛争に巻き込む危険」。
The geography of the U.S.-Iran conflict is expanding
geography はここでは「地理的範囲」。
→ 「米国とイランの衝突の戦場・影響範囲が広がっている」。
shot down an Iranian missile fired at Turkey
fired at ～ は「～に向けて発射された」。
→ 「トルコに向けて発射されたイランのミサイルを撃墜した」。
take this as their cue to increase supply
take A as B = 「AをBと受け取る」。
cue は「合図」。
→ 「これを供給拡大の合図とは受け取っていない」。
tap into their 401(k)s
tap into は「資源・資金などを引き出して使う」。
→ 「401(k)年金の資金を取り崩す」。