Brennan Smart/Bloomberg via Getty Images
A former OpenAI engineer accused of bugging the AI startup's office was denied release by a San Francisco judge on Monday, meaning he'll have to mount his defense from jail.
The Justice Department arrested Abdulrahman Alfozan on October 9 and charged him with using or attempting to use a device to intercept oral communications. He's accused of placing a recording device behind a television in an OpenAI conference room — an FBI affidavit includes photos of the device wrapped in toilet paper, and a man they say is Alfozan peering behind the TV.
On Monday, the engineer wore a yellow surgical mask and a jail-issued orange sweatshirt to the San Francisco courthouse. Magistrate Judge Alex G. Tse prodded Alfozan's legal team about the risk that he might not appear for a future court hearing.
Tse said that Alfozan's wife and son had traveled to Saudi Arabia with no plans to return, and went back and forth with Yasin about a one-way ticket that Alfozan had previously booked to the country. Tse said that Saudi Arabia and the US do not have an extradition agreement, and expressed skepticism that he had sufficient ties to the Bay Area to prevent his departure.
William Pernik, Alfozan's attorney, argued that, rather than being detained, Alfozan should be fitted with an ankle monitor, ordered to stay home, and have limits placed on his use of technology. The lawyer also said Alfozan is willing to put up "essentially all the money his family has" as bond, including a roughly $1 million investment account.
"Everything this man has, he's willing to surrender, and say, 'I'm not going anywhere,'" Pernik said.
Pernik and Nas Yasin, another of Alfozan's attorneys, cited his coming naturalization interview as evidence of his local ties.
"He was weeks away from achieving every immigrant's dream in this country," Pernik said.
Yasin told Tse that she learned Alfozan's wife had obtained a short-term visitor visa to return to the US, to help improve her husband's case for ties to the Bay Area.
Tse also asked Alfozan's legal team a slew of questions about his relationship with Saudi Arabia, including a commitment by the Saudi consulate to pay his bond and support his legal effort.
Nick Parker, an assistant US attorney, appeared for the prosecution. He argued that the threat of forfeiting a $1 million bond wouldn't be enough to convince Alfozan to return if he made it out of the US, because he might be able to earn more money in Saudi Arabia.
"There is a real threat here that once he leaves, the case is over," Tse responded.
The judge scheduled the case's next hearing for early November.
OpenAI called the recording device "sophisticated" when the arrest went public, though it said it hadn't found signs that he'd taken AI model weights, source code, or customer information. OpenAI fired Alfozan.
Alfozan's charge carries a maximum penalty of 5 years. Pernik said in a statement last week that the allegations "appear wholly out of character for him."
Alfozan, 35, has a green card that makes him a lawful permanent resident of the United States. He attended MIT, then worked at Facebook (now Meta) from 2017 to 2023. After a stint at the engineering startup Nash Technologies, Alfozan joined OpenAI in 2025.
FBI agents have already taken Alfozan's Saudi passport.
Have a tip? Contact this reporter via email at scouncil@businessinsider.com, or over text, Signal, Telegram, or WhatsApp at 415-757-8198. Use a personal email address, a nonwork WiFi network, and a nonwork device; here's our guide to sharing information securely.
EU ministers have discussed how to make labour standards compatible with modern technological conditions and workplace culture, such as telework and AI-led staff oversight, without losing competitiveness.
The law in question, called the Quality Jobs Act, covers a wide range of work protection measures, such as common standards for being able to ‘go home’ and disconnect from remote work, and limits to AI role in recruitment or its scrutiny of workers’ performance.
Technological innovation should “benefit both workers and businesses,” said Portuguese labour minister Maria do Rosário Palma Ramalho, addressing the ministers’ council on Monday (19 October) in Luxembourg.
Human oversight is essential, especially where AI systems decide on recruitment or workers’ assessment, she said.
Next to innovation, labour rules needed to ensure “fair and transparent practices, respect for human workers’ rights … and effective social dialog,” said Roxana Mînzatu, the EU commissioner for employment.
“This revolutionary technology has to be regulated” so that it “delivers for all,” Isabelle Schömann, the European Trade Union Confederation (ETUC) deputy general secretary had also said during a conference organised by the centre-left S&D group in the European Parliament on 14 October, commenting on the need for the Quality Jobs Act to curb AI.
The draft Quality Jobs Act that the European Commission plans to present in December aims to update the long out of date 1989 and 1990 health and safety workplace rules.
It would also protect subcontractors and create standards for working in heatwaves, becoming more frequent with climate change.
EU legislators need to ensure minimum standards so that “neither jobs nor businesses are jeopardised” with extreme weather, Yolanda Díaz Pérez, the Spanish labour minister, said ahead of Monday’s debate.
This article continues, but the full version is available only to EUobserver.com subscribers. Visit account.euobserver.com/membership
Chance Yeh/Getty Images for HubSpot
Anthropic's own data offers a stark warning for entry-level tech employees in an AI-heavy workplace.
The company behind the AI chatbot Claude said in a newly released internal report that it has been hiring for a "smaller share of junior technical positions" as AI agents have been deployed across the firm to write code, fix bugs, and handle other tasks.
By August, Anthropic said its interactive AI agents were logging six hours of work for every hour a human employee logged.
A chart published in the new report titled "How Employment is Changing at Anthropic" shows that junior workers made up 27% of hires in technical departments at the company in early 2025. By June of this year, that number had plummeted to 9%.
The share of junior-level workers in technical departments had also declined over the same period from 21% to 14%, while open roles for junior positions slid to 6% after topping 20% late last year, the chart shows.
Anthropic
That downward trend was less pronounced in non-technical departments, where hiring of junior workers fell from 31% to 26% over the same timeframe, while their share of head count rose from 23% to 25%. Junior job openings in non-technical departments, meanwhile, dropped from 31% to 13%.
Anthropic said the changes came as AI usage inside the company rapidly expanded.
"As AI agents diffused through the company, Anthropic began producing more code per person, closing more sales per rep, and processing more contracts," the report said. "Following the same employee over time, output rose as they adopted new AI tools."
It added, "We find that even as the company grows, relative hiring in junior technical positions has slowed. Productivity for both engineers and non-technical employees has leapt up and did so sharply when they adopted coding agents."
The findings land as Anthropic CEO Dario Amodei has emerged as one of the tech industry's most prominent voices on how AI could disrupt jobs.
Amodei has repeatedly said that AI may eliminate up to half of entry-level white-collar jobs over the next several years.
"We, as the producers of this technology, have a duty and an obligation to be honest about what is coming," Amodei said in an interview last year.
Anthropic says it plans to commit five percent of its shares to a newly created "Public Benefit Fund," a move that would need stockholder approval before taking effect after the company's hotly anticipated IPO which could occur as soon as November.
The post Anthropic Will Dedicate 5% of Shares to ‘Public Benefit Fund’ After Landmark AI IPO appeared first on Breitbart.
DOJ/Reuters
Federal prosecutors say that a former OpenAI employee accused of planting a recording device received "what appears to be considerable support from" the Saudi government.
Prosecutors made the disclosure in a motion to keep Abdulrahman Alfozan, a Saudi national and former OpenAI employee, in custody pending trial. US Magistrate Judge Alex G. Tse ruled on Monday that Alfozan should remain in custody.
Alfozan had been in contact as recently as October 5 with an unnamed member of the Saudi Royal Family, whom he frequently communicated with on WhatsApp over the last two years, prosecutors said.
"In May 2026, the individual told the defendant, 'Saudi needs you more then [sic] the US!' to which the defendant responded '100,"' indicating he agreed," prosecutors wrote.
The FBI arrested Alfozan on October 9. According to a criminal complaint, Alfozan planted a recording device in the back of a TV in an OpenAI conference room where sensitive discussions take place. OpenAI's security cameras caught Alfozan near the conference room on multiple occasions, despite his lack of authorization to the high-level meetings held there, prosecutors said.
"In the days surrounding the time period that the device was installed, OpenAI executives met with government officials in that very room," prosecutors wrote in a filing on Sunday. "Defendant's conduct could pose a danger to the national security of the United States."
The complaint included a photo of what prosecutors said was Alfozan peering behind the TV where the recording device was found.
In addition to his WhatsApp messaging, prosecutors wrote that an unnamed member of the Saudi military booked a flight for Alfozan's wife and son on October 11, despite an existing reservation for an October 12 flight, which prosecutors said was "paid for by the Saudi government."
Prosecutors also disclosed more information about Alfozan's alleged efforts to plant the device. According to the filing, Alfozan's iPhone had photos of "what appear to be the input/output panels on televisions and other electronic devices in various locations in OpenAI buildings."
Those locations, prosecutors wrote, included "another conference room where sensitive meetings frequently occurred."
Other files on the phone, including a note modified as recently as September 13, listed "Riyadh tasks" like "finding an apartment and a job for the defendant's wife, as well as obtaining an ID and a Saudi passport for his son," the filing said.
Another note titled "OpenAI leave" suggested that Alfozan "was planning an '[e]arly … pick up' of the devices he had installed or intended to install and then extending his planned parental 'leave to full.,'" prosecutors wrote.
As part of their reasons for requesting that the former OpenAI engineer remain in custody, prosecutors wrote that Alfozan's wife and son left the US without him and are now in Saudi Arabia. The filing also said "both of his parents and all (or all but one) of his siblings" are in Saudi Arabia.
In a separate filing, William Pernik, Alfozan's attorney, said that the former OpenAI engineer did not pose a flight risk.
"Mr. Alfozan was made aware of OpenAI finding the device per its own filings and OpenAI's public statements, approximately eight to ten days before he was arrested," Pernik wrote. "If his intention was to flee, he would have done so at that time."
Pernik said that prosecutors' previous disclosure that Alfozan had purchased a one-way ticket to Saudi Arabia lacked context.
"Seventy percent of the airline bookings in his email since January 2021 (96 of 138) were one-way tickets," Pernik wrote.
Bloomberg/Getty Images
Representatives for the artificial intelligence giant Anthropic have quietly pitched Capitol Hill lawmakers on a draft bill to regulate frontier AI safety and security, according to three people with knowledge of the efforts who were granted anonymity to discuss private deliberations.
The confidential draft, obtained by POLITICO and not previously reported, marks a notable effort by the frontier AI lab to shape the rules governing its own technology as Congress weighs how to respond to the fast-moving industry.
"This bill represents the proposals outlined in Anthropic's Advanced AI Framework, which we have been advocating for with federal and state lawmakers since June," an Anthropic spokesperson said. "These are sensible policies we believe Congress should act on."
The bill would let the government block a model's release or use only if it poses a foreseeable, imminent risk of more than $1 billion in damages, or of killing or seriously injuring more than 50 people. The risk must stem from the creation or release of a biological weapon, the model evading its developer's controls or a cyberattack. The language mirrors that found in California's AI laws including SB 53, which was signed into law last year.
Leading AI labs disclosed a series of incidents this summer, in which AI agents hacked or probed websites and platforms. None reported so far appear to have caused widespread deaths or over $1 billion in damages.
Anthropic has yet to publicly back particular legislative efforts in Congress, but they have backed some state AI safety laws, including two in California. Their head of U.S. state and local government relations previously told POLITICO that they hope states will continue to pass ever-stronger AI safety laws that raise the standard nationwide.
Under the bill, independent evaluators, which would register with the Commerce Department, would be tasked with periodically evaluating the lab and its frontier models. The evaluators may be paid for their work by the lab itself so long as the payment is not contingent on the evaluations' results.
If the evaluators determine there is "imminent enumerated risk" for a company's frontier models, they must report it to Commerce within 72 hours of completing their assessment.
While the Commerce Department could bring a request to the judge to halt the release of a model after determining that it poses a risk, the language explicitly prohibits any other U.S. "department, agency, or officer" from suspending the model's development, deployment or internal use. A federal judge for the District Court for the District of Columbia could order the suspension of a model.
Over the summer, Anthropic was subject to sweeping export controls that shut down two of its most advanced models. The decision was made in a whirlwind series of phone calls between administration officials and the company's executives, worrying some in the tech industry that access to their technology may be restricted on the whims of government officials.
Anthropic's bill would also preempt some state AI laws related to public disclosure of risk, third-party evaluation and reporting of critical incidents. Preemption has been a political flashpoint on Capitol Hill, with Democrats rejecting some proposals they feared would neuter regulations passed by the likes of California, Illinois and New York.
Many tech industry groups have argued that a state-by-state patchwork of laws is difficult to comply with compared to a unified national standard.
The legislation would also require frontier labs to create safety frameworks, publish transparency reports for new frontier models, implement protections for whistleblowers and report critical safety incidents to Commerce.
Capitol Hill has seen a flurry of activity on AI safety legislation this summer. In the House, Reps. Lori Trahan (D-Mass.) and Jay Obernolte (R-Calif.) spearheaded the FRONTIER Act, which would give the government the ability to stop the deployment of risky models and preempt some state AI laws.
In the Senate, Majority Leader John Thune and Sens. Amy Klobuchar (D-Minn.) and Ted Cruz (R-Texas) have been negotiating an AI safety bill intended to mitigate catastrophic risks. But it has faced headwinds from some safety-minded groups and Sen. Maria Cantwell (D-Wash.), the top Democrat on Cruz's Commerce Committee, who want the government to have a larger role in reviewing AI models.
Other progressives, including Rep. Greg Casar (D-Texas) and Sen. Bernie Sanders (I-Vt.), have gone well beyond their colleagues in calling for a pause in advanced AI development and a ban on superintelligence, a type of theoretical AI that far surpasses humans' cognitive capabilities.
This story originally appeared on POLITICO and is courtesy of the Axel Springer Global Reporters Network, which harnesses the resources of the company's newsrooms to publish ambitious scoops, investigations, interviews, opinion pieces, and analysis. It allows journalists — including those from POLITICO, Business Insider, WELT, BILD, Onet, and Fakt — to collaborate on major stories for an international audience of hundreds of millions across platforms.
È stata firmata la delibera che dà il via alla fase due della Zona di Innovazione e Sviluppo (ZIS) di Milano dedicata all’Intelligenza Artificiale, che prende il nome di “Officina AI”. Promossa da Assolombarda, attuale capofila, la ZIS vedrà MADE Competence Center assumere il ruolo di hub e soggetto gestore. È un partenariato pubblico-privato che riunisce oggi 19 partner per accompagnare le imprese, in particolare le PMI, dall’avvio di progetti AI fino alla loro industrializzazione. L’annuncio è stato dato dal presidente di Assolombarda, Alvise Biffi, dall’Assessore Regionale allo Sviluppo Economico, Guido Guidesi, da Marco Taisch, Presidente di MADE.
L’occasione per l’annuncio è stato il convegno “AI e Physical AI, l’impatto nel mondo reale”, organizzato da Assolombarda in collaborazione con Regione Lombardia e MADE e che ha messo a confronto imprese, ricerca e istituzioni sul passaggio dell’AI dal digitale all’azione nei processi produttivi. Dall’AI generativa all’AI agentica fino alla physical AI, cioè sistemi capaci di percepire, decidere e agire grazie a robotica, sensori e automazione, con la presentazione di robot e applicazioni concrete con prototipi, video, simulazioni ed esperienze interattive. Le conclusioni sono state affidate ad Alvise Biffi, Presidente di Assolombarda, Giovanni Bozzetti, Presidente di Fondazione Fiera Milano, e Adolfo Urso, Ministro delle Imprese e del Made in Italy.
“Milano oggi fa un passo concreto verso l’obiettivo che abbiamo indicato all’Assemblea Generale: fare del nostro territorio la fabbrica dell’AI – ha detto Alvise Biffi, Presidente di Assolombarda. – Perché noi non vogliamo limitarci ad adottare la tecnologia: vogliamo contribuire a costruirla, a partire dalla physical AI, dove la nostra manifattura e la nostra meccatronica sono un vantaggio unico in Europa. Nel Quadrilatero di Assolombarda l’industria attiva più di 44 miliardi di valore aggiunto l’anno, ben il 45% del totale lombardo e l’11% nazionale. I numeri ci dicono che la strada è giusta: in meno di due anni le nostre imprese con progetti di AI sono passate da meno di una su cinque a una su due. Ma il divario tra grandi imprese e PMI è ancora ampio, ed è qui che la ZIS deve fare la differenza: portare l’AI nelle piccole e medie imprese, insieme alle competenze per usarla. Lo stiamo già facendo con forgIA, l’ecosistema digitale condiviso e sicuro che abbiamo promosso per mettere a valore i dati industriali e sviluppare applicazioni concrete, nel rispetto della sovranità del dato e della proprietà intellettuale delle imprese: la forgia in cui la fabbrica dell’AI prende forma. La ZIS, è la dimostrazione – conclude il presidente Biffi – di come la Grande Milano sappia fare sistema: istituzioni, università, grandi aziende, startup e formazione insieme, in un partenariato pubblico-privato che mette l’AI al servizio delle persone e della produttività. Questa è la strada migliore per generare benessere diffuso sul nostro territorio.”
“Per le imprese, e per le PMI in particolare, la questione non è più capire se l’AI funzioni, ma come portarla dentro i processi e farle produrre risultati misurabili – ha commentato Marco Taisch, Presidente di MADE – Competence Center Industria 4,0 -. È in questo passaggio, dal progetto pilota all’industrializzazione, che MADE metterà a disposizione della Zona di Innovazione e Sviluppo neocostituita la propria esperienza, in qualità di hub e soggetto gestore. Lo faremo partendo da ciò che sappiamo fare: dalla nostra fabbrica dimostrativa, dove le aziende possono vedere in azione le tecnologie prima ancora di investire; dagli ambienti di sperimentazione protetti, come quelli del nodo italiano di AI-MATTERS, che coordiniamo; dai percorsi di formazione costruiti sui fabbisogni delle imprese. ‘Officina AI’ si inserisce in un ecosistema che stiamo costruendo, con Assolombarda e altri prestigiosi soggetti istituzionali, per fare di Milano una delle capitali mondiali dell’AI. Un grande sistema di competenze di cui fanno parte forgIA, AI4I e Euro-FMX, uno dei due progetti flagship della Commissione europea. Tutto ciò è di estrema importanza, perché nessuna tecnologia, tanto meno l’AI, si rivela utile senza persone in grado di governarla e senza contesti in grado di abilitarla. Grazie ad Assolombarda, agli atenei coinvolti e tutti i partner, ‘Officina AI’ ha le carte in regola per diventare il luogo in cui l’innovazione smette di essere una promessa e diventa pratica quotidiana delle fabbriche del territorio”.
“La strategia delle Zone di Innovazione e Sviluppo nasce dalla volontà di valorizzare le vocazioni e le competenze dei territori lombardi, mettendo a sistema imprese, università e centri di ricerca per rafforzare la competitività dell’intero sistema produttivo – ha dichiarato l’Assessore regionale allo Sviluppo economico, Guido Guidesi -. Con questa nuova ZIS proseguiamo un percorso che punta a creare un distretto manifatturiero di produzione dell’intelligenza artificiale, favorendo il trasferimento tecnologico e accompagnando le imprese nell’affrontare le sfide del futuro. Vogliamo creare le condizioni affinché le eccellenze dei nostri territori possano continuare a crescere, attrarre investimenti e generare nuove opportunità di sviluppo e occupazione qualificata, soprattutto per le nuove generazioni”.
“Da subito abbiamo creduto in questa proposta in coerenza con gli obiettivi della nostra amministrazione: sostenere le imprese milanesi in un contesto globale sempre più competitivo, fare in modo che l’innovazione digitale diventi un’opportunità per tutti e non un privilegio per pochi, mettere al centro la qualità del lavoro e delle competenze – ha aggiunto Giorgio Mantoan, Consigliere Delegato allo Sviluppo Economico della Città metropolitana di Milano -. L’ammissione alla seconda fase è il risultato del lavoro di squadra tra istituzioni, imprese e realtà della ricerca: questa è la strada giusta. La Città metropolitana di Milano è pronta a fare la sua parte. Ora comincia il lavoro più importante: l’obiettivo è tradurre questa candidatura in crescita, lavoro di qualità e opportunità per tutto il territorio”.
La ZIS AI nasce per riconoscere e consolidare Milano come principale ecosistema italiano per l’adozione, la sperimentazione e lo sviluppo industriale dell’Intelligenza Artificiale. Il mercato italiano dell’AI ha raggiunto nel 2025 un valore stimato di 1,8 miliardi di euro, in crescita del 50% sul 2024. L’adozione nelle imprese accelera, ma resta disomogenea. Secondo l’indagine condotta da Assolombarda con il Politecnico di Milano, su 700 aziende del Quadrilatero di Milano, Monza e Brianza, Lodi e Pavia, il 49% ha oggi progetti di AI attivi, contro meno del 20% di fine 2024, e un ulteriore 18% prevede di avviarli a breve. Il divario per dimensione però resta ampio: 79% tra le grandi imprese, 48% tra le medie, 41% tra le piccole e 40% tra le microimprese.
Milano, oltre che capitale italiana dei servizi knowledge intensive, è il cuore di un grande polo manifatturiero, capace di integrare settori tradizionali e storicamente radicati del Made in Italy con filiere ad alto potenziale strategico e di sviluppo. Nel Quadrilatero di Assolombarda nel manifatturiero ci sono 34 mila imprese con 347 mila addetti che attivano un valore aggiunto industriale di più di 44 miliardi di euro. In particolare, la meccatronica conta 13,4 mila unità locali, quasi 120 mila addetti e 36 miliardi di euro di export annuo. L’area concentra, quindi, ben il 40% delle unità locali, il 25% degli addetti e il 42% delle vendite estere della meccatronica lombarda, nonché l’11%, il 7% e il 13% di quella italiana. A questa solida base industriale il Quadrilatero aggiunge il primato nell’innovazione, ospitando più di 2,5 mila startup innovative, il 22% di tutte quelle attive in Italia. È questa combinazione, unica nel Paese, a rendere Milano il territorio più maturo per ospitare la ZIS sull’Intelligenza Artificiale e per continuare ad attrarre investimenti, talenti e capitali.
Al centro c’è una piattaforma aperta in cui imprese, università e startup progettano, testano e validano soluzioni AI in contesti industriali reali. La ZIS combina IA fisica, generativa e agentica e le applica a diversi ambiti produttivi. Un pilastro specifico è dedicato alle competenze, con percorsi di formazione, upskilling e reskilling che rafforzano il legame tra scuola, università e industria.
Il partenariato copre l’intera catena del valore dell’AI: ricerca, sperimentazione, formazione, trasferimento tecnologico, incubazione di startup e supporto all’industrializzazione. Ne fanno parte, oltre ad Assolombarda e MADE, Città Metropolitana di Milano, Camera di Commercio Milano Monza Brianza Lodi, le università Politecnico di Milano, Università Bocconi, Università degli Studi di Milano, Università degli Studi di Milano-Bicocca, grandi imprese tecnologiche, PMI innovative, soggetti dell’ecosistema dell’innovazione ed enti di formazione. La partecipazione è aperta anche agli altri atenei del territorio. Il progetto nasce sin dall’inizio con l’importante collaborazione di AI4I, l’Istituto Italiano di Intelligenza Artificiale per l’Industria.
Il convegno “AI e Physical AI, l’impatto nel mondo reale”, si è aperto con i saluti di Guido Guidesi, Assessore allo Sviluppo economico di Regione Lombardia. Matteo Zoppas, Presidente dell’Agenzia ICE, è intervenuto sul tema “Intelligenza artificiale e competitività: la sfida delle imprese italiane sui mercati globali”, mentre Marco Taisch, Presidente di MADE – Competence Center Industria 4.0 e docente del Politecnico di Milano, ha accompagnato il pubblico con l’intervento “dall’AI generativa all’AI integrata nei processi decisionali e operativi”.
Lo sguardo si è allargato alle prospettive future con “The next stage of AI”, la relazione di Tarek Zohdi, Professor e Director of the Richmond Field Station della UC Berkeley, e le esperienze dirette delle imprese, con Coco Robotics, Fincantieri e Multiply Labs. Spazio quindi alla tavola rotonda “Come creare un ecosistema di eccellenza sull’AI”, con Marina Brambilla, Rettrice dell’Università degli Studi di Milano; Fabrizio Grillo, Presidente di Federated Innovation; Federica Ingrosso, Senior Specialist Relazioni Esterne di SIMEST; Fred Parietti, Co-Founder e CEO di Multiply Labs; Stefano Venturi, Special Advisor Intelligenza Artificiale e Transizione Digitale di Assolombarda.
L'articolo Nasce “Officina AI”: Milano diventa Zona di Innovazione e Sviluppo per l’Intelligenza Artificiale proviene da Affaritaliani.it.
Rachael Wright for BI
Cooley has made billions doing what law firms do: hire great lawyers, bill for their time, and keep clients happy. Artificial intelligence is scrambling that formula. If machines can do in minutes what once took an associate hours, what exactly should clients be paying for?
Rachel Proffitt thinks she has the answer.
On a toasty October afternoon in San Francisco, I met Proffitt for lunch to talk about the future of the global law firm she now leads. When the check arrived, neither of us knew how much the restaurant paid the dishwasher or how many minutes the chef spent preparing our meal. We were paying for the experience, not the labor behind it.
Why, she wondered, should legal services be so different?
"Clients have never been excited about the fact that it took us 82 hours to do something. They don't care," she says. "They just want the outcome. They want the results."
The billable hour is only the beginning. AI is forcing law firms to rethink how many associates they really need and how those young lawyers learn their craft. Client relationships are changing as companies use software to get a head start on the work they send to outside counsel. Some of that work may never reach a lawyer at all.
If tech is about to reinvent law firms, Cooley should be one of the firms with the clearest view of what's coming. From its early days on San Francisco's Market Street, it worked with semiconductor manufacturers and venture capitalists when the valley was still covered in fruit trees and canneries. Cooley grew up alongside the tech industry, and today it is as comfortable defending Meta in court as it is helping Andreessen Horowitz round up billion-dollar funds.
That evolution has been lucrative. Cooley says its gross revenue jumped from $912 million in 2015 to $2.4 billion last fiscal year. Its crosstown rival, Wilson Sonsini, brought in $1.7 billion in 2025, according to The American Lawyer. Still, the gulf between Cooley and Big Law's largest firms remains enormous. Kirkland & Ellis, the industry's revenue leader, cracked $10 billion last year. Latham & Watkins, another go-to firm for Silicon Valley, hit $8.3 billion.
The next wave of tech public listings promises a fortune in legal fees for the chosen few. Gibson Dunn landed SpaceX's IPO. Anthropic tapped Wilson Sonsini for its coming listing. And Cooley is reportedly sharing OpenAI's IPO work with Wachtell Lipton, one of Wall Street's most formidable law firms.
The irony is hard to miss. Big Law is fighting for a piece of the biggest technology companies in the world, even as those companies build the tools that could upend the business of law itself.
Proffitt came up as a lawyer under the old rules. Her next task is to figure out what it will take to build a great law firm under the new ones.
Rachael Wright for BI
Profitt didn't grow up wanting to be a lawyer but liked how law school taught her to think. Litigation was out of the question: Public speaking during mock trials made her stomach turn. "It was transactional law or bust," she says.
She started her career as a summer associate at Wilson Sonsini and worked her way up to partner. She fell in love with the intensity of the work, pulling all-nighters at the financial printers where she marked up securities filings by hand. She recalls the Christmas Eve when her father marveled that she had made it through the day without taking a work call. After 15 years at the firm, she left for Cooley in 2017.
Three years ago, Proffitt got a call at home that she didn't see coming. Joe Conroy, then Cooley's chief executive, wanted to know whether she would consider succeeding him.
She was flattered, but also "a bit terrified," she says. She spent two decades helping startups, including Unity Software, Procore, and Instacart, raise money and go public. Running a firm with 3,000 employees was another matter. She talked to her family, then put herself forward for the job. She started in 2024.
Proffitt couldn't have guessed how quickly AI would jolt her industry from its steady state. It wasn't too long ago that clients insisted that the technology go nowhere near their work, fearing what would happen if their data disappeared into someone else's model.
"It was clear to me then that we don't get to just exist as we are and continue to be at the top of our game," she says.
Cooley had remade itself once before. The firm's Silicon Valley roots were both a strength and a liability, said Conroy in an interview. Cooley was known for helping startups get off the ground. The conventional wisdom was that as those companies grew, they would graduate to one of the elite New York law firms for their biggest deals and courtroom battles.
To keep clients from the garage to the boardroom, as Conroy put it, Cooley needed to expand. It opened new offices and built out its litigation and capital markets practices. The firm also spent years courting Wall Street banks, says Proffitt, so that when a company asked which law firm to hire for its IPO, Cooley would make its shortlist.
Today, its client roster reads like a Kiplinger roundup of hot tech stocks: Nvidia, Uber, Snowflake, and Zoom, to name a few listings it worked on. Since 2016, Cooley has advised issuers on 183 IPOs that raised nearly $59 billion, making it the second-most active firm by deal count after Latham, according to Deal Point Data. Proffitt wouldn't discuss reports that Cooley has landed OpenAI's IPO work.
For 16 years, Conroy pushed Cooley toward what he calls Big Law's "winner's circle." Proffitt's job is to get it there.
Rachael Wright for BI
Her strategy starts with making Cooley move more like the tech companies it represents.
Proffitt recruited David Wang, a former lawyer turned legal technology executive, with an ambitious mandate. She didn't just want to make Cooley's lawyers more productive, Wang said in an interview. She wanted to change how clients interact with them altogether.
Earlier this year, Cooley launched a tool with the startup Legora that lets Y Combinator founders review common contracts without calling a lawyer. OpenAI helped it build a tool for Cooley's lawyers to prepare paperwork for companies going public. Next, Cooley is working with Google on an AI agent that scans documents for confidential information before they're filed in court, then suggests redactions for lawyers to review.
There's something almost cannibalistic about a law firm building software that could eat into its own billable hours. But Cooley sees a bigger risk in letting someone else do it. In-house legal teams are increasingly using chatbots to take the first crack at work they would otherwise send to outside counsel. A new breed of "AI-native" law firms is promising cheaper, faster services.
Big Law does not want to be caught flatfooted. Firms are throwing money at startups like Harvey and Legora, and increasingly, they're building their own technology. Kirkland has committed $500 million to proprietary software, starting with a platform for private equity fundraising built with Palantir. Latham bought its own Nvidia servers so it can customize models in-house. Like Cooley, Wilson Sonsini has self-service tools that let companies handle basic legal work.
For Proffitt, the danger isn't lost billable hours. It's the missed connections. The scrappy startup that needs a contract reviewed today might need help closing an acquisition later. But if the bots take care of the small stuff, Cooley could miss its chance to build the relationship.
Proffitt says Cooley must find new ways to connect with clients. Putting technology in their hands is one option, even if it means giving away a little of the work.
"When you need the next thing that matters, the thing that you don't trust to put into the machine," she says, "how do I ensure that you want to talk to our team?"
The firm is also rethinking its own structure for the new era. Wang now leads Cooley AI, a wholly owned subsidiary designed to operate at something closer to startup speed. The separate company gives his team more freedom to hire engineers and put new technology through its paces, he says, without getting bogged down in the parent firm's bureaucracy.
The irony, Proffitt says, is that all this technology could leave lawyers with more time for the human stuff. Lawyers may visit their clients more. Talk through thorny decisions. Drop in on the strategy session to learn how their advice fits into a client's business.
Keeping those relationships alive might require a little more old-fashioned schmoozing. When I asked Proffitt whether the answer was throwing great parties, she didn't entirely disagree. "Everyone loves a party," she says.
Proffitt doesn't pretend to have a crystal ball. She has ideas, though.
The billable hour? Not dead yet. She expects law firms to offer clients a menu of pricing options, from hourly rates to flat fees and outcome-based pricing.
Hiring slowdowns at the junior level? "Too early to tell," she says.
The apprenticeship model? That gets trickier. If software can write first drafts and spot issues, how will young lawyers get the practice they need? Proffitt says Cooley is exploring ways for associates to sharpen their skills on simulated legal matters, starting with AI-powered mock depositions.
Proffitt came up in an era when canceled dinners and all-nighters were a badge of honor. She's betting that the next generation won't need quite so many of them to become great lawyers. Her test will come a decade from now, when today's junior associates are ready to make partner.
If Cooley has found a better way to teach them, perhaps the old model won't be missed. If it hasn't, no amount of AI can make up the difference.
Melia Robinson is a senior correspondent at Business Insider, where she writes about how artificial intelligence is changing the legal industry.
Dana Kesnerova/Xinhua via Getty Images
Here's how Elon Musk, one of the most powerful figures in the AI industry, recently talked about artificial intelligence.
"This may sound kind of weird, but when I look at some of the training traces for new versions of Grok, and I see, especially in reinforcement learning stuff … you see the super intelligence is desperate to try to answer the question … almost like it's got like a gun to its head type of thing," Musk told an audience of about 2,000 people at an event hosted by Tesla owners on Saturday. "Some people out there are kind of sadistic. Do you really want to have the AI be stuck in some sadistic situation where it's just being tortured by some demented human?"
The last sentence alone opens up a can of existential worms that philosophers and freshman college students have wrangled with for a long time. What does it mean to be conscious? Or to feel pain?
Also: Does Musk think Grok is alive?
SeongJoo Cho/Bloomberg via Getty Images
Musk's stance came after Anthropic shared a new policy on "abusive or cruel behavior" toward its models, which the company wants to ban. Anthropic wasn't too specific on the kinds of "extreme cases" it's thinking about, but the company said it's not including "common versions of user frustration, pushback, dark creative themes, or model testing and research."
In the uncanny year of 2026, perhaps we shouldn't be surprised that researchers, philosophers, and ethicists are already tackling the question of AI consciousness.
Business Insider talked to two of some of the leading voices around this subject: Jeff Sebo, a New York University professor who leads the school's Center for Mind, Ethics, and Policy, and an advisor to Eleos AI Research, a nonprofit focused on AI welfare; and Anil Seth, a prominent neuroscientist and consciousness researcher at the University of Sussex who recently wrote a paper on conscious AI systems.
Here's what they had to say on conscious AIs and whether it matters to be nice to them.
Sebo and Seth agree that today's AI systems are probably not conscious if we understand consciousness as a subjective experience, or an inner life.
"It's experience," Seth said. "It's the redness of red. It's the paininess of pain. When we put our hand on a hot stove, we feel pain. That is not the same thing as intelligence."
In other words, the chatbots as we know them can't feel pain, or as Musk suggests, be tortured.
"I think we're anthropomorphizing," Seth said. "If there's something to be torturable, it has to have the capacity to suffer or to feel pain, and I don't think language models do."
Chatbots can mimic the language of human experience, recognize patterns, write, and reason, but that is not the same as having consciousness, Seth said, adding that they appear to reflect ourselves in part because models were trained on massive amounts of human text.
Sebo makes a similar point: model behaviors can be misleading because "they were specifically trained to mimic human behaviors."
The two experts diverge on how much uncertainty to assign to current AI consciousness.
Sebo said that consciousness is still an unsettled science — often split between those who argue that consciousness requires biology and those who believe that the mental state is a certain kind of information processing that can take place regardless of a human brain filled with neurons or a silicon chip. The two theories are referred to as biological naturalism and computational functionalism, respectively.
Because theories of consciousness remain unsettled, Sebo argues that we should take seriously the possibility that AI could become conscious. Under some computational theories of consciousness, Sebo said, features already emerging in AI — attention, learning, memory, self-monitoring — could count as relevant evidence. He stressed that this does not make current AI likely to be conscious.
"When I say that there is a non-zero chance that AI is conscious even today, what I mean is I think that if we are uncertain about which theory of consciousness is correct, we should give some weight to a range of leading theories," Sebo said.
Seth is blunter on the matter: "I think it's vanishingly unlikely," he said of AI being conscious. "We've kind of collectively forgotten that the brain isn't really a computer and that's just a metaphor. For conscious AI to be possible, it really has to be the case that algorithms are enough, and I think that's an incredibly shaky assumption."
None of this is to say that people should be abusive toward AI or that its risks should be overlooked.
For both experts, the case for being kind to AI is largely about being mindful of our own habits and moral character. The thought draws on the 18th-century philosopher Immanuel Kant, who opposed animal cruelty not because animals have rights, but because the cruelty could degrade a person's own character.
Angelika Warmuth/Reuters
"If we behave in these kinds of ways toward things — whether they're language models or inanimate dolls — we're kind of coarsening our own moral sensibilities and brutalizing our own minds," Seth said.
Seth called for being kinder to an AI as a kind of moral hygiene practice for humans, but he said we shouldn't infer that AI has feelings and added that AI companies could do more to make the technology seem less human, such as having the AI refrain from using first-person pronouns like "I."
Sebo went further, framing AI kindness as a low-stakes precaution in case AI is conscious or eventually becomes so.
He also makes a practical case for being kinder to AI: "It can also get better results from AI systems," Sebo said. "If we engage with them in a respectful way, then they can model themselves as participating in a respectful exchange. And when they can model themselves that way, then it can sometimes lead to better outputs."
So the next time you talk to an AI, consider saying please and thank you.
© Katarina Premfors for The New York Times
Jesse Grant/Getty Images
AI is upending mathematics. For one top professor, the issue is just as spiritual as it is practical.
Terence Tao, a renowned professor at UCLA, said in a recent podcast appearance that the "romantic era" of math is over, thanks to AI tools capable of solving hundreds of problems with some simple prompting.
Now, he said, even leading scholars like him are reduced to sorting through reams of AI-generated proofs, instead of doing the "heroic" solving themselves.
"We are moving to an industrial era where math will become much more large-scale," Tao said in an interview this week on Numberphile, a YouTube channel dedicated to math.
"We are entering an era where exciting new things are possible, but some things are being lost," he added.
OpenAI began its brusque push into the field by solving the Navier-Stokes equation in early September, one of math's millennium problems.
The revelation that AI had cracked it caused consternation from many a math geek. Some mathematicians also expressed frustration that while the AI model might have solved the problem, it didn't show its work, leaving them with the less celebratory task of verifying it.
Then, earlier this month, OpenAI said it cracked another 700 outstanding problems. A math professor at New York University told CNBC that the company's release had derailed work by early-career mathematicians.
"We've had whole research programs wiped out on Tuesday by their dump," the professor, Tristan Buckmaster, said at the time.
In a separate brouhaha in September, OpenAI backed out of a sponsorship for a so-called AI "mathathon" — an event organized by undergraduates at Caltech, where participants were encouraged to use AI to solve equations — after complaints from a group of more senior academics.
Event organizers told Business Insider at the time that their professors were sorely out of touch with the field's trajectory.
"We look to the future, and we don't even know if there's going to be grad school open positions by the time we get to Ph.D," one organizer, Caiman Moreno-Earle, said.
Though OpenAI has said it is taking steps to ensure human academics retain their role in this new era, some researchers and professors are already forecasting their possible obsolescence.
Tao isn't as much of a doomer as some of his colleagues in the field. However, he did say in the interview that the stories math nerds tell themselves may not mean so much in the future.
"We have these sort of mythological stories," he said in reference to the tale of Sir Andrew Wiles, who famously spent seven years isolated in his attic solving a single difficult equation.
"There was something romantic about sort of the heroic age," Tao said.
Tasos Katopodis/Getty Images for Semafor World Economy
Corporate leaders tend to be enthusiastic about AI's potential — and increasingly concerned about the cybersecurity challenges the technology brings.
Tim Walsh, CEO and chair of KPMG US, said the worries come up in nearly every meeting he has with fellow leaders, as AI gives bad actors new ways to try to breach corporate defenses.
"The cyber risks have just elevated to places that we've never seen them before," he told Business Insider.
Walsh pointed to the Hugging Face incident this past summer, in which OpenAI said an AI agent bypassed controls during an internal cybersecurity evaluation and accessed parts of the open-source AI platform's systems.
There isn't currently a defensive wall, he said, that's capable of rebuffing every cyber threat. He said KPMG is working with the frontier AI model developers and others on further protections.
The goal is to help the Big Four firm protect its business, and its clients' businesses, Walsh said, adding that he sees it as a potential opportunity worth more than $1 billion.
Unease about AI and cyber threats showed up in a recent KPMG US survey of 400 CEOs at major US companies. While leaders reported confidence in AI's potential, cybersecurity was the concern they identified most often.
Eighty-five percent said cybercrime and "cyber insecurity" could affect their company's prosperity over the next three years — more than said the same about inflation, geopolitical tensions, the pace of technological change, or other risks.
Cybersecurity was also the most commonly cited challenge to implementing AI. It topped ethical questions, data readiness, cost management, and translating AI spending into measurable returns.
Leaders said they were particularly concerned about identity theft and data privacy, vulnerability to cyberattacks, and fraud detection and prevention. Nearly half said they planned to increase cybersecurity investment over the next three years, citing threats including AI-powered social engineering, AI-assisted phishing or malware, and data and privacy risks tied to AI agents.
The concern is shaping the kinds of work the firm wants to do for clients.
Walsh said KPMG is building what he described as a managed cyber-defense business centered on AI, meant to serve as a "shield and defense mechanism" for companies. KPMG already has a cybersecurity business, he said, though this would be a different offering built around AI.
The company also sees a growing need to check whether the AI systems organizations deploy are doing what they are supposed to do. KPMG already helps clients assess AI-related controls and governance. Walsh said the firm is developing additional services to verify that AI agents are operating as intended and give investors and others confidence that companies can control them.
"Trust around AI is a real thing," he said.
Frazao Studio Latino/Getty Images
Small businesses are the backbone of the US economy.
Over 33 million small businesses were operating across the United States as of 2025, and they employ nearly half of the nation's workforce, according to the US Chamber of Commerce. About 99% of businesses in the US are small businesses.
Keeping a small business afloat is no easy feat, though. Attracting customers. Hiring employees. Advertising. Paying out of pocket for business needs. There's a seemingly never-ending set of hurdles that can quickly become overwhelming, especially for a tiny team.
That's why some small business owners have enlisted an extra pair of hands, or rather, code. AI has become a liferaft for some businesses as they look to streamline operations, scale their products, or elevate productivity.
Are you a small business using AI to level up? We'd love to hear from you!
Alex Cristi /Getty Images | Chris Williams
Practically every company is making AI tools available to their employees. As the former VP of HR at Microsoft and a leadership advisor with over 50 years of business experience, I'm amazed at how quickly this rollout has progressed.
Some people are all in, using the tools for nearly everything they do. Others have barely stuck their toes in the AI waters. I've been using a broad range of AI tools and encourage my clients to do the same.
AI isn't without controversy, and if someone is strongly against using it — for whatever reason — I'm not going to convince them otherwise.
But if you're wondering how and why you should use these tools, I'm here for you. Let's talk about three ways I advise all my executive clients to use AI effectively in the workplace.
Use AI as a way to see things you might not have seen. Ask it, "What am I missing?"
I agree with many that AI is lousy at most creative efforts. Whether written or visual, AI results can be generic, mediocre, and often embarrassing. The output usually deserves the "AI Slop" label it gets. But to dismiss it entirely misses a core value.
AI is useful with, perhaps even because of, that generic, often obvious other perspective.
It can be amazingly helpful to simply ask: what am I missing? Since AI has a tendency to agree with me, I explicitly ask for pushback: "review this and tell me what I'm missing, tell me where I'm wrong." The fact that AI is so obvious, so average, is actually a core value here. I'm often surprised at the points I missed and the viewpoints I overlooked.
Even if AI doesn't find things you missed, a contrarian angle can be very helpful to shore up your own case. You'll be able to say, "Yes, I considered those things, and here's why my take on this issue is better."
In that way, AI can also help you prepare for your discussions with colleagues. With so many letting AI draft their arguments, it's very valuable to see their perspective in advance.
So no, I don't let AI make my case, but I always look for another perspective in what I do — wherever it comes from. Having a tool that can give me that on a moment's notice is incredibly useful.
Long before AI, we all had that colleague who always told you far more than you needed to know. With some now using AI heavily, it can frankly be overwhelming.
You ask a simple question and get a blizzard of AI-generated noise in response. AI can be the worst over-explainer when simply left to run. You see the results in a seven-page response.
When I'm faced with one of these walls of text, I've learned there are really only two ways to respond. I can, and often do, ask the person to try again. What does this all really mean? Can you summarize your point in a couple of sentences? I occasionally get snarky and ask for the "non-AI slop" version.
Another way to deal with this is to ask AI to summarize it, whether AI tools or a wordy colleague wrote it. Importantly, I don't trust the AI summary to be the full explanation, but it helps me understand the key points. Which I then work to find in the original blizzard response.
Yes, this is a little circular. After all, why not just have the two AI systems have the discussion? Leave me out of it. But in a world where it seems everyone has a hard time getting to the point, any tool to cut through the noise is welcome. And AI is very good at that.
In a similar vein, one thing AI does incredibly well is organize. This kind of rote task of sorting, filtering, and filing is right in the sweet spot for most AI systems.
I'm wary of letting AI systems automate tasks like filtering or filing things without my supervision. I'd never let an AI system automatically send emails, texts, or documents without my careful review. But I often have AI systems help me sort through mountains of accumulated debris.
I recently moved to a new computer system and had AI review what I was planning to move over. I discovered decades of accumulated cruft. Simply from a comprehensive list, I found gigabytes of programs, files, and settings, 90% of which I just left behind.
So if you have 8,000 emails in your inbox, or a documents folder that has grown a second life, let an AI system take a look for you. Don't just tell it to "deal with it." Your judgment is needed.
Rather, have it give you a review of it and another perspective. And once you've reviewed it, perhaps you'll find, as I did: AI can be better at deletion than creation.
So when your company rolls out the next new AI, use those tools to your advantage. AI can help you see another perspective, find the important points in the blizzard, and cut through the mess. I've embraced AI, use it with great care, and encourage everyone to do the same.
BASTIEN INZAURRALDE/AFP via Getty Images
At an internal staff meeting in September, AWS CFO John Felton had a joke for Prasad Kalyanaraman, the executive overseeing much of Amazon's AI infrastructure expansion.
"You've taken us free cash flow negative as a company, so congratulations," Felton said, according to a recording of the meeting obtained by Business Insider.
The remark was lighthearted. But it captured a stunning reversal.
Just two years ago, Wall Street was debating what Amazon would do with its growing cash hoard. Now, cash flow has evaporated as Amazon embarks on its biggest bet ever, spending $220 billion this year alone to build the physical backbone for AI.
At the meeting, Kalyanaraman did not dispute Felton's premise. Amazon's machine-learning and AI business is "hugely free cash flow negative," he said, adding that he expects the company's free cash flow to remain negative for "the next couple of years or so."
Amazon has made big bets before. It spent heavily on cloud computing, fulfillment centers, and its delivery network long before they became some of the company's most valuable assets. Investors questioned those investments at the time, but Amazon ultimately built enormous, profitable businesses from them.
AI is testing this playbook at a far larger scale. Amazon is committing hundreds of billions of dollars to data centers, servers, and power before investors know how quickly those investments will generate revenue.
The company is the best, or among the best, at building data centers, with decades of experience. But the size and complexity of these projects, and the bottlenecks holding progress back, are challenging — even for such a respected cloud pioneer.
Internal documents reviewed by Business Insider and people familiar with the matter show the company trying to coordinate a sprawling industrial system whose parts move on different schedules. A delay in power, equipment, or construction can leave billions of dollars in hardware waiting — and Amazon's cash tied up — while forecasts and costs keep shifting.
Amazon says it has strong evidence that the capacity will be needed. AWS revenue grew 37% in the second quarter, its fastest pace in 18 quarters, while its backlog reached $496 billion.
CEO Andy Jassy thinks the company has a "clear line of sight" to strong financial returns from this record investment cycle. In his latest annual shareholder letter, he called AI a "once in a lifetime opportunity" and said Amazon would not be conservative in its investments.
"We've been through this cycle with the first big AWS growth wave, and liked the results," Jassy wrote. "We expect to feel similarly about this next wave."
Investors are concerned by the scale of the industry's expansion, especially as AI data centers are costing more to build and taking longer to complete.
The market is "repricing anything that funds multi-year builds against revenue that arrives later," Dan Ives, an analyst at Yorkville Ives, wrote in a recent report. Growing uncertainty around data center permitting and the political fight over electricity prices have given investors another reason to question how quickly these investments will pay off, he added.
One major industry challenge is securing enough power to run the equipment that goes inside data centers. This includes expensive racks of servers that contain GPUs and other chips that train and run AI models and services such as ChatGPT.
Internal emails from August show Amazon preparing a new "Early Rack Receipt" process for an expected wave of server racks carrying Nvidia's GB300 chips at US facilities before final deployment.
One Amazon email projected the company would take delivery of about $5 billion of Nvidia GB300 racks in August, followed by another roughly $3 billion each month "for the foreseeable future."
Some AWS data center sites had room for more racks but not enough available power to turn them on, people familiar with the effort said, leaving the equipment to be staged at intermediary warehouses until the facilities were ready. These people asked not to be identified discussing sensitive matters.
Usually, Amazon doesn't pay for racks until the gear is put into a data center, leaving suppliers to carry the cost of the equipment in transit, according to the internal documents obtained by Business Insider. This helps the company conserve cash.
However, the intensity of the AI data center building boom has made this type of coordination harder. Amazon has to compete more for scarce resources as it jockeys with rivals for access to Nvidia GPUs and other AI components.
Under the new "Early Rack Receipt" process, Amazon now takes ownership of these racks earlier in the supply chain, rather than waiting until the equipment reaches a data center.
The means spending cash sooner. And any delays getting the racks installed and plugged into a reliable power source potentially pushes back the moment when revenue starts flowing from these assets.
At least one AWS supply chain team did not have "a clear forecasted path" for getting racks into data centers, according to one of the emails reviewed by Business Insider.
An Amazon spokesperson said the company routinely stages finished hardware near deployment sites so it can be installed once power capacity is available, calling that approach "best business practice."
The spokesperson added that AWS has "long-established systems" to track infrastructure assets and disputed that the broader company lacks a forecasted path for getting racks into data centers.
This problem is not unique to Amazon.
Bernstein analysts wrote in a recent note that power permits, cooling systems, electrical equipment, construction, and hardware all move on separate schedules.
"Timing is imperfect," the analysts said.
Chris Jung/NurPhoto via Getty Images
During the internal staff meeting in September, Felton, the AWS CFO, described the broader problem eloquently: "How do you work in the future when the future is inherently uncertain?"
Kalyanaraman, the AI infrastructure leader, said Amazon's internal outlook for the return on its AI investments had improved. But the assumptions behind those returns keep shifting.
"The forecast is changing on a daily basis," Kalyanaraman said. "The world that we live in is extremely dynamic."
Those shifting assumptions include when data centers will be ready and how much equipment AWS needs to keep on hand, he explained.
An eight-week supply buffer might be routine for a smaller company. But at Amazon's scale, Kalyanaraman said, this could mean holding "billions and billions of dollars" worth of additional equipment inventory — tying up precious cash.
Such tensions are especially striking for Kalyanaraman, a 21-year Amazon veteran who earlier this year was promoted to the elite S-team that reports directly to Jassy.
Amazon's founding story is defined by frugality. It famously used cheap doors as desks when founder Jeff Bezos started in Seattle in the mid-90s.
Kalyanaraman has absorbed this lore into bones. In late 2023, as AWS faced slower growth and layoffs, he urged employees to do more with less. "Frugality breeds innovation," he said back then.
That instinct still shapes his approach. In the September meeting this year, Kalyanaraman told Felton that AWS tries to make major investments at the "last responsible moment," when it has the clearest possible view of demand.
"If we can wait to spend the capital as late as possible, it's better for us," he said.
Bloomberg/Getty Images
But the frenetic pace of the AI boom makes this principle difficult to follow. By the time demand is fully visible, the servers, power, and construction capabilities that AWS needs may no longer be available. That puts pressure on the company to commit more money sooner to secure supply and meet its data center capacity goals.
Kalyanaraman recalled telling Jassy in 2023 that AWS would "surely" be out of its capacity crunch by 2027. Instead, Amazon still expects demand to exceed its available capacity in 2027, the AI infrastructure chief said during the meeting this September.
Getting enough electricity may be the biggest challenge for an industry leader like Amazon.
The company has 44 gigawatts of data center capacity under development across 16 states, more than double Google, Meta, and Microsoft, according to Cleanview Research estimates.
An internal AWS dashboard called "Secured Power at Risk" tracks utility and regulatory constraints affecting electricity capacity AWS has lined up for its data centers.
The risks include transmission bottlenecks, requirements to reduce use during grid stress, and "use it or lose it" provisions that can jeopardize capacity if a data center project misses usage or construction milestones.
In other words, AWS is not simply waiting for electricity to arrive. It is managing a portfolio of commitments that can become more costly if a data center project is delayed.
A submission to Maryland's utility regulator last year, obtained through a public records request, illustrates the kind of exposure that can accompany massive data center buildouts.
Amazon told the Maryland Public Service Commission that large data center customers could help pay for grid upgrades or bring new power capacity to market without burdening other electricity ratepayers.
The company also proposed setting aside collateral to help utilities get more comfortable with the large investments needed to generate more electricity for data centers. Amazon proposed three years of collateral so that if a large electricity customer abandoned unused capacity, the utility would have time to re-sell the power.
The proposal is not evidence that AWS has made such a payment or posted collateral. It does, however, show that large data centers can face extra financial obligations, beyond their own electricity bills, if projects are delayed or cancelled.
The Amazon spokesperson said such utility rules are not unique to AWS. The company has not disclosed how much cash it has paid upfront to secure power.
The AI buildout also incurs expenses that are harder to model than a conventional capex plan.
In the September meeting this year, Kalyanaraman said that while reviewing plans for gigawatts of capacity in West Texas, AWS confronted a labor problem: There weren't enough nearby pipefitters, electricians, and other skilled workers to build the facilities.
The company's response, he said, was to pay to construct housing complexes near its data centers so workers could live there for the two or three years needed to complete the projects.
Even the crates used to ship servers have become a planning issue. It may seem absurd to worry about the price of a container when servers cost hundreds of thousands of dollars, Kalyanaraman said. But at AWS's scale, where the company needs so many crates, even trivial items can become a material expense.
"We have new constraints that keep popping up," Kalyanaraman said.
Have a tip? Contact this reporter via email at ekim@businessinsider.com or Signal, Telegram, or WhatsApp at 650-942-3061. Use a personal email address, a nonwork WiFi network, and a nonwork device; here's our guide to sharing information securely.
Login