Author: openjargon

• Citadel CTO Umesh Subramanian said he personally calls top candidates, including recent college grads.
• The calls have won many candidates over, but occasionally show they're not the right fit, he said.
• While the hedge fund talent war isn't new, AI has made recruiting even more competitive.

Hedge funds are locked in a fierce talent war, shelling out pay packages worth tens of millions to secure top talent — and that means executives are going all-in to woo prospective hires, even at the entry-level.

Umesh Subramanian, Citadel's chief technology officer, leads the tech teams that power the firm's investment, research, and risk-management platforms. He told Business Insider that he personally calls some of the most in-demand candidates during the decision-making process to see where their head is at. That includes not just senior-level recruits, but also recent grads, he said.

"I get on the call with them to understand how they're going to make the decision — what is driving their decisions," Subramanian said. "And oftentimes we win those candidates."

In general, the CTO said he looks for four qualities in top candidates: intellectual curiosity; a passion for winning as a team; an interest in commercial applications and not just theory; and a strong engineering background and education.

While Subramanian directly contacts some of the most sought-after candidates, those conversations haven't always resulted in them getting hired. The CTO said that sometimes the phone calls have led him to suggest they look elsewhere because another company may be more aligned with the role they describe being interested in.

"I've had a couple of conversations where I've said, 'Look, I think you should take the other job,'" He said he thinks Citadel is a great and "highly selective" company at which to start your career, but that "it is not for everyone."

Ongoing talent wars

Subramanian isn't the only executive to personally reach out to top talent in the interview process. OpenAI's Sam Altman has been similarly been said to call candidates to convince them to join the tech company, and Meta CEO Mark Zuckerberg has previously shown up in email chains to recruit talent, as well as reportedly hosted top candidates at his home for meals.

While the talent war at hedge funds isn't new, the scramble for AI talent has added another layer of competition to the hiring process and raised the stakes for recruiting the right people — many whom are highly sought after by traditional tech giants.

Banks like Goldman Sachs and Bank of America have allocated billions to their tech budgets, with a portion of that investment going into hiring.

Subramanian said he has always been personally engaged in recruiting, and it's part of the company's culture for executives to be highly involved in seeking out top talent.

Citadel, specifically, is among the most competitive hedge funds. In 2025, the hedge fund and its sister firm, Citadel Securities, accepted just 0.4% of applicants for its summer training program, a record-low acceptance rate. Together, the two firms received 108,000 applications — a 20% increase from the previous year.

"Exceptional talent, world-class talent, is worth fighting for with everything that you got," Subramanian said.

• A Ukrainian company working on autonomous systems said autonomy in defense is "greatly overhyped."
• Much of what is called autonomous isn't really, he said.
• Autonomy can describe a host of abilities, including some where humans still play a big role.

The CEO of Ukrainian arms maker Ark Robotics told Business Insider that "autonomy in defense is greatly overhyped."

Much of what is labeled autonomous isn't, he said. "If we're talking about real frontline usage, we are talking less than 1% where autonomy is present."

"There's a gap between the hype around autonomy and what actually works on the frontline today," said Achi, who spoke to Business Insider using a pseudonym as a security precaution.

Defense forces, he said, "want a hundred percent reliability, and the AI is just not there." Artificial intelligence alone won't get the job done right now. Human oversight is still heavily required, but a shift is expected.

While autonomy may be deeply overhyped today, it may be completely unavoidable tomorrow.

Ark is working on more autonomous tech, and expects rapid improvements in the area. "I think we already are past the point of no return," the CEO said, adding that the current battlefield shows "autonomy is necessary."

Ark Robotics makes a suite of autonomous robots used by more than 20 Ukrainian brigades, and is developing a system that allows thousands of aerial drones and ground robots, including ones made by different companies, to work together with minimal human involvement. The system, Frontier, isn't fully autonomous, nor does it claim to be. "That's not where the industry is yet," Achi said.

He said autonomy is present in Frontier but described it as an early step toward network-level autonomy at scale.

"Time is essential," Achi explained. "Democratic countries aren't the only ones racing to build such products. We're doing our best, supported by our team's strong autonomy expertise and continuous feedback from the front line."

The autonomy question

Autonomous systems have become increasingly common in Ukraine as domestic companies and Western allies bolster its fight against Russia's invasion, now close to entering its fourth year. These systems help offset Russia's larger force size, reduce demands on operators, and keep drone pilots farther from danger.

Autonomous systems make decisions faster than humans and handle certain routine battlefield tasks, freeing troops to focus on higher-level work.

Much of what is described as autonomous still requires heavy human involvement. Part of the issue is definitional — autonomy can describe everything from basic navigation to a system that identifies, classifies, and acts on a target without direction.

More advanced systems might launch a weapon autonomously or coordinate a drone swarm. Others can learn and make decisions themselves, but many still keep humans in the loop for both ethical and policy reasons.

Warfare experts say the term "autonomous" is often used inconsistently.

A report from the UK's Royal United Services Institute last year noted that there is no agreed-upon definition for how autonomous an "autonomous" system actually is.

Kateryna Bondar, an AI and defense expert at the Center for Strategic and International Studies, wrote earlier this year that "fully autonomous warfare remains an aspiration," but "significant progress" has been made in partial autonomy, especially with drone tech. In Ukraine, she said, "autonomous" is often used interchangeably with "unmanned," even if only limited autonomy is involved.

Chasing autonomy

Achi said his company is pursuing autonomy on two levels. The first is edge autonomy, where systems make decisions on their own without instructions from an operator or control center. He gave the example of an aerial drone scouting an area, waiting for a target, able to act on its own.

The second level, which Ark is more interested in, is the orchestration layer, where multiple drones or robots coordinate and give tasks. Achi said this level is where you can give the system "an abstract task," like protecting a perimeter, and it can then "orchestrate whatever is needed to do that," such as directing systems such as strike drones. Humans can step in if needed.

He said the second is "the hardest to solve" as it needs a good edge foundation to build on.

Frontier, Ark's AI-powered system, is designed to coordinate thousands of uncrewed systems, sharply reducing operator demands and keeping humans much farther from combat. In a demonstration in Denmark, a user remotely operated a ground robot 1,200 miles away in Kyiv.

Ark, founded in Ukraine and now headquartered in NATO member Estonia, aims to keep humans involved as little as needed so decisions can be made much faster.

An autonomous future

Western militaries and companies, as well as rivals Russia and China, are pushing for more autonomous systems. NATO is prioritizing autonomous technology, particularly in drone technology, for threat detection.

Among many lines of effort are programs like the US Department of Defense's push for all-domain attritable autonomous systems through the Replicator initiative and the UK's substantial investment this year of more than $5 billion in uncrewed autonomous systems.

How far autonomy can go technologically remains to be seen, and its use may be limited by ethical debates. As is, Western militaries often restrict higher levels of autonomy in lethal systems.

But Achi said there won't be any turning back, and expects all militaries to be using it in some form. He said autonomy has become "kind of prerequisite to be successful in the total drone warfare that is coming to all of us."

"You can have all these fancy drones," he said, "but what is the use of them if you can't really deploy them at scale?"

• The phrase "AI bubble" has recently been in the zeitgeist.
• Mentions in earnings calls and investor conference transcripts increased 740% from last quarter.
• Only demonstrable profits will quiet the talk, analyst Gil Luria told BI.

There's a bubble in talk of an "AI bubble."

The phrase "AI bubble" appeared in 42 earnings calls and investor conference transcripts between October and December — a 740% increase from the previous quarter, according to an AlphaSense analysis.

In the third quarter of 2025, five transcripts mentioned the term. The quarter before that, four. For all of 2024, executives and analysts used the phrase in 24 transcripts total. In 2023, it was nine. (The analysis counted transcripts where "AI" and "bubble" appeared within five words of each other.)

Now dozens of companies across semiconductors, cloud infrastructure, financial services, and industrials are fielding some version of the same question: Is the AI boom a bubble?

At UBS's Global Technology and AI Conference on December 2, AMD CEO Lisa Su told investors, "from the standpoint of, you know, do we see a bubble? We don't see a bubble."

Nvidia CFO Colette Kress similarly swatted the premise aside, describing it onstage as a "supposed AI bubble."

Her boss, CEO Jensen Huang, said as much in Nvidia's November earnings call. "There's been a lot of talk about an AI bubble," he said. "From our vantage point, we see something very different."

Aixtron, a German semiconductor-equipment maker, was asked in its November earnings call whether customers might slow capacity plans if AI enthusiasm wanes. Coface, a Paris-based credit insurer, described re-running the math on big AI-related capital requests to ensure they hold up even if "bubble-type enthusiasm" fades.

So why the sudden spike?

Analysts pointed to the sheer scale of the numbers being tossed around.

"We've been seeing these huge partnerships and people throwing out money amounts for infrastructure — like trillions of dollars — which I can't remember ever hearing before," Sarah Hoffman, director of AI thought leadership at AlphaSense, told Business Insider. That gap between the trillions being invested and the billions in revenue from many AI efforts, she said, "is what's getting people a little bit freaked out."

Gil Luria, managing director and head of technology research at DA Davidson, said analysts and CEOs are voicing anxieties they're already hearing from investors.

Luria pointed to what he called "circular transactions" as one source of anxiety. "Nvidia investing a dollar in CoreWeave, CoreWeave borrows nine, and uses eight of them to buy Nvidia chips," he said. "That's the echoes of bubbles past."

The past eight weeks have given bubble proponents plenty of ammunition.

Tech luminaries, including Sam Altman and Bill Gates, have said there's some froth in AI spending. In earnings calls this quarter, Google, Meta, Microsoft, and Amazon all said they planned to increase their spending on data centers that are the backbone of AI. Michael Burry, whose bet against the housing market was made famous in "The Big Short," made headlines after he said he had shorted Nvidia, the world's most valuable company.

While executives have pushed back on skeptics, Luria said that only cash will quiet the conversation.

"There's nothing they can say that will change people's perception of whether or not there's a bubble," he said. "The only thing they can do is show results, and those have to be financial results."

• Christmas is three weeks away, but shoppers have more time than ever to procrastinate on gifts.
• Amazon, Target, and Walmart are each offering ultrafast ordering options as late as Christmas Eve.
• The changes are part of a broader push to make shopping even more convenient — and immediate.

Santa has more helpers than ever this year.

Christmas is still three weeks away, but shoppers now have more time to procrastinate on gifts and supplies for the holiday.

Amazon, Target, and Walmart are each racing to offer ultrafast fulfillment options as late as Christmas Eve, as each company looks to earn shoppers' trust that they can deliver in a pinch.

Target said in November that it is extending store hours from 7 a.m. to midnight in the weeks leading up to December 23, and from 7 a.m. to 8 p.m. on Christmas Eve before closing for Christmas Day.

During those operating hours — including Christmas Eve — Target says customers can get orders within two hours via curbside or in-store pickup, or opt for same-day delivery for a $9.99 fee (or free for Circle360 members).

And starting December 9, Amazon will display an "Arrives before Christmas" message on item listings, many of which will be available for fulfillment as late as Christmas Eve via delivery or one of the company's 25,000 pickup locations.

The company says it will even gift-wrap certain items.

Walmart has also been leaning hard into establishing itself as one of the fastest ways to get anything from barbecue sauce to Barbie dolls — and it's only getting faster.

The retail giant hasn't made a Christmas-specific delivery announcement yet, but last year it promised to fulfill express delivery orders placed as late as 4 p.m. local time on Christmas Eve.

More recently, Walmart said it fulfilled its fastest Black Friday order in 10 minutes, with big increases in both the volume and speed of deliveries fulfilled from its stores.

The changes are part of a broader push among retailers to make shopping more convenient — and immediate.

Target is rolling out a shopping experience within ChatGPT — including delivery options on orders with multiple items — ahead of the holidays. And, this week, Amazon announced a 30-minute delivery test in Seattle and Philadelphia, while Walmart expanded its drone delivery services in Atlanta.

In other words, the race is on to see who can deliver the cheer like Saint Nick.

• Apple has lost over a dozen executives and researchers this year, with others announcing their retirement.
• Meta hired nine of these former Apple employees.
• Others left for AI startups, OpenAI, or simply retired.

They're not called the AI talent wars for nothing.

With Big Tech firms scrambling to attract top talent, executives and artificial intelligence researchers have done a lot of job-hopping this year, and Apple employees were no exception.

The company has lost over a dozen employees who worked on its AI projects, from executives to scientists to engineers, and even its AI chief, John Giannandrea, who announced this week he was stepping down from the role.

One competing tech company has been the biggest beneficiary of the departures at Apple: Meta.

Nine of the former Apple employees listed below have landed at Meta, where CEO Mark Zuckerberg has been offering notoriously large pay packages and incentives to attract AI talent from across the industry.

Other ex-Applers jumped ship to join OpenAI.

To be sure, Apple has done some of its own recruiting too; its new vice president of AI is Amar Subramanya, who left Google in June to join Microsoft, only to join Apple several months later. Meta's top lawyer, Jennifer Newstead, is departing to become Apple's general counsel.

Here's a running list of the AI talent Apple has lost this year and where they've ended up.

Executives:

John Giannandrea

Giannandrea was the senior vice president of machine learning and AI strategy at Apple and held the role for almost eight years. He will be retiring in the spring of 2026, but will serve as an advisor to Apple until then.

Alan Dye

Dye was the head of human interface design and was a pivotal figure in designing some of the company's software. He will be joining Meta to run a new creative studio inside its Reality Labs division.

Kate Adams

Adams has served as Apple's general counsel since 2017. She will transition her duties to Newstead and retire later in 2026.

Lisa Jackson

Jackson, Apple's vice president for environment, policy, and social initiatives, is set to retire at the end of January 2026.

AI leaders and researchers:

Frank Chu

Chu was the head of engineering at Apple for nearly six years before he departed for Meta. He is now a software engineer at the Meta Superintelligence Labs.

Ruoming Pang

Pang was a senior distinguished engineer at Apple who led the company's foundation models team. He joined Meta's Superintelligence Labs as an AI research scientist in July.

Robby Walker

Walker was the senior director of answers, knowledge, and information at Apple. He announced his departure a month prior to leaving in October.

Tom Gunter

Gunter joined Meta in July as an AI research scientist after over seven years at Apple, where his title was distinguished engineer.

Ke Yang

Yang was the senior director of machine learning at Apple. He was at Apple for six years and described his departure as "bittersweet" in a LinkedIn post. He is now an AI research scientist at Meta.

Chong Wang

Wang was a distinguished scientist at Apple before joining Meta as an AI research scientist.

Shuang Ma

Ma was a senior research scientist at Apple. She joined Meta's Superintelligence Labs as a research scientist in July.

Liutong Zhou

Zhou was a senior applied machine learning scientist at Apple until he left for Cohere, a Canadian AI startup, as a member of technical staff on foundation models.

Bowen Zhang

Zhang, formerly a staff machine learning researcher at Apple, joined Meta in July as an AI research scientist.

Mark Lee

Lee was a research engineer at Apple before joining Meta in July as an AI research scientist.

Brandon McKinzie

McKinzie was a senior research engineer of foundational models at Apple, before joining OpenAI as a member of its technical staff.

Dian Ang Yap

Yap was a machine learning researcher at Apple. He joined OpenAI in August as a member of its technical staff.

• UK startup Global Airlines acquired its second four-engine jet: a leased 25-year-old Airbus A340.
• It already owns an Airbus A380 and flew a handful of trips in May, but early reviews were mixed.
• Industry analysts question Global's strategy of flying fuel-thirsty quad-jets in a crowded market.

A new British airline is building its future on fuel-thirsty widebodies that most major carriers have spent years getting rid of.

UK-based Global Airlines — a carrier whose plan to fly costly quad-engine jets across the Atlantic has drawn wide skepticism — acquired a "new" plane on Tuesday: a 25-year-old Airbus A340.

It's leased from the third-party operator Hi-Fly, with the registration 9H-SUN, and is not owned by Global.

The choice has raised eyebrows as the A340 family, which first flew in 1991, is a jet more likely to be scrapped in an aircraft boneyard than painted in a fresh livery. Think of it like buying a used car with 200,000 miles on the odometer.

Global CEO James Asquith revealed the plane in an Instagram video on Tuesday, panning to the plane in a hangar in Portugal. It was the first major update for the airline since flying its fully owned Airbus A380 — another fuel-hungry quad-engine jet — in May.

Global, a privately held startup founded in 2021 with early backing from private investors, has been in and out of the spotlight in a "will they, won't they" survival saga that has analysts questioning its unusual plan to use large, expensive jets across the highly competitive transatlantic corridor.

Henry Harteveldt, a travel analyst and president of Atmosphere Research Group, told Business Insider that the A340 is a "puzzling" choice considering its high operating and maintenance costs.

"I've flown on the A340 as a passenger; it's a sound airplane, it's safe," he said. "But it's just not the most logical or cost-efficient aircraft to acquire. A well-suited twin-engine like the Boeing 777 or Airbus A330 would inevitably be less expensive to operate and maintain."

To be fair, Harteveldt said it's possible Global didn't have the capital. Asquith has previously said he has found bargain prices on old, unwanted planes, which will offset the high costs of operating them. The economics depend partly on fuel prices remaining stable.

Choosing the A340 shows Global is leaning hard into this strategy — building a fleet around gas-guzzling, harder-to-fill widebodies. Many other airlines have gone in the opposite direction.

European carriers Lufthansa, Swiss, and Edelweiss plan to retire their remaining A340s within the next two years. Cirium data shows a dozen other airlines still operate the quad-jet, though none are US-based. Back in 2003, the earliest Cirium data available, nearly 40 carriers flew the A340.

The A380 has seen similar retirements. Air France, China Southern (the previous owner of Global's A380), and Thai Airways withdrew the superjumbo as its high fuel burn and limited flexibility became harder to justify.

Airbus ended A380 production in 2021 after orders dried up.

Global's A380 has been grounded for 4 months

When Global first bought its 13-year-old A380, registered 9H-GLOBL, it required thousands of hours of maintenance after spending a year collecting dust in the California desert.

"We've worked really hard the last few years to do what everyone said was impossible and put our first own A380 into the sky, through heavy maintenance, and flying passengers across the Atlantic with our operating partners," Asquith said.

He added that Global had ambitious plans and some goals may have been a "stretch too far," but it "comes with the territory of buying A380s and getting them flying."

One missed goal is timing: Global originally aimed to launch in 2024, but it still has no regular service or its own operating certificate. Its one-off May flights, operated by Hi-Fly, were to prove it could carry passengers, and the flagship A380 has been grounded since July.

Asquith told Simple Flying in December that the jet must undergo a weeks-to-months-long "D-check," meaning mechanics must essentially disassemble it piece by piece to inspect its safety and airworthiness, and then reassemble it.

Global plans to eventually start US to UK service. But doing so means taking on airlines with entrenched loyalty programs, major corporate contracts, and fleets of fuel-efficient twin-engine Airbus and Boeing aircraft far cheaper to operate than Global's aging, four-engine jets.

Beyond the nostalgic appeal of its quad-engine jets, Global has little yet to draw travelers. With only two planes, any maintenance problem can fully disrupt travel.

Younger airlines are also prone to sudden shutdowns, as seen with WOW Air in 2019 or the short-lived Nevada-based regional airline Aha! in 2022, leaving travelers with booked tickets in the lurch.

And the few flights that run weren't cheap. Its April debut fares were around $1,000 round-trip in economy, $5,000 in business, and $9,000 in first — higher than competitors on the inaugural route.

The airline bills itself as a luxury option, so premium fares aren't surprising, but uncompetitive pricing hurts its appeal — especially when the cabin fell short of the hype.

Global eventually cut fares by roughly half, but only filled about a fourth of its 506 seats: "The A380, and the A340 even, is way too much plane for a startup," Hardeveldt said.

Global's cabin and route strategy need tuning

Global's inaugural flight drew mixed reactions from bloggers and YouTubers on board.

While some praised the food and amenities, others complained about the broken in-flight entertainment and inconsistent service. And some said the seats — even in business and first class — didn't match the grand interior Global had long promised

Instead of installing a new cabin, Global kept the China Southern seats but upholstered them with Global branding and added touches, like complimentary Champagne in economy and lamps in business.

Harteveldt said it would take at least 18 months to design, certify, and install a completely new cabin.

Even if it fixes the luxury aspect, Harteveldt questioned where Global could profitably fly. London Heathrow and Gatwick are slot-restricted, limiting daily flights. Other metro airports are inconvenient or can't handle the larger quad-engine jets.

Plus, UK destinations are already served by well-known airlines. American Airlines, for example, is launching an Airbus A321XLR with an all-new cabin to Edinburgh in December. JetBlue and United are also expanding to Scotland.

"Richard Branson started Virgin Atlantic with a single [four-engine] Boeing 747 between New Jersey and Gatwick, but this isn't the 1980s," Harteveldt said. "It's a very different North American market now; airlines aren't going to cede even one-tenth of a point of market share willingly."

• Alfredo Mercedes built an AI-enabled modular recruiting platform after starting his career in defense tech.
• He runs his platform, which lives inside a VC fund, remotely from Medellín, Colombia.
• He said he's seeing AI companies struggle with overheated compensation and employee retention.

This as-told-to essay is based on a conversation with Alfredo Mercedes, the 27-year-old founder of VU Talent Partners based in Medellín, Colombia. It has been edited for length and clarity.

I'm the founder of VU Talent Partners, and I recently relocated from Orlando to Medellín, Colombia.

I started my career in the Marine Corps Reserve, where I trained as an infantry mortarman. I next worked as an executive recruiter at Daversa Partners, an executive retained search firm for VC portfolio companies, where I specialized in placing executives in cybersecurity and defense technology roles.

I left Daversa in August 2024 for a six-figure role at Defense Unicorns, a Series A defense technology company that delivers AI and open-source capabilities to national security and DoD systems.

Company pivots created uncertainty around my job. I could've moved into a leadership role, but it wasn't aligned with my personal vision. I wanted to build something of my own where people and venture intersect, helping innovative tech founders scale teams that advance democracy and security for real-world problems.

I took a layoff with severance in December. After leaving Defense Unicorns, I was unsure of my next move. Now I run an AI-enabled modular recruiting platform inside a global VC firm.

A VC firm recruited me for my current role

In January, I was recruited by VU Venture Partners after the GP reached out to reconnect since we had crossed paths years before to launch VU Talent Partners, where I lead today. Our goal is to eliminate the chaos of scaling by building talent infrastructure that scales with startups.

I had built a reputation in defense and frontier tech recruiting. VU Venture Partners came with a unique proposition: Instead of building another search firm, what if we built a talent platform born inside the fund, powered by the same data and signals on which startups rely?

I jumped at the opportunity

It helped that VU Venture Partners was also a perfect fit, as I was a graduate of the Venture University Accelerator, and VU Venture Partners is an early-stage global venture capital fund.

I'm the founder and CEO, which comes with a 50/50 split of ownership and profit. I run the VU Talent Partners platform on a day-to-day basis, while the VC provides capital, infrastructure, and network leverage.

3 things made me feel confident to leave a high-paying career and take this risk

First, I had a personal runway. I received rental income from my house in Orlando, which I own, and I lowered my living costs by relocating to Medellín, Colombia, and giving myself some breathing room.

Living in Medellín, Colombia, is more affordable in almost every aspect of cost of living, including rent, transportation, food, and entertainment. I live in a two-story penthouse in the main district area for $1,200 a month. Groceries delivered to my house in 30 minutes cost $40. Ubering anywhere in the city costs $10.

Second, I felt the market demand. Startups are scaling leaner, and venture firms are struggling with refreshed talent sourcing. As AI became a capacity multiplier, I felt recruiting infrastructure and speed to outcomes were missing pieces, especially.

Third, I knew I would be a unique fit. With my military experience, executive recruiting expertise, and startup operator skill set, I knew I could deliver results confidently.

From my front-row seat to the AI talent wars, these are the biggest challenges I'm seeing

• Overheated compensation: I've seen AI roles with total potential compensation north of $1M, squeezing startups out of the competition.
• Signal vs. noise: Thousands of people rebrand as "AI experts," but few have shipped real systems. Filtering that talent is critical.
• Dual-use talent gap: Defense and frontier companies, in particular, need employees who can operate at the intersection of AI, government, and enterprise.
• Retention: Landing talent is one thing, but keeping them engaged when they're constantly approached with offers is another.

My story is about building a life centered on ownership, resilience, and alignment. The Corps taught me grit. Recruiting taught me leverage. Venture taught me scale. AI is teaching me code.

At 27, I'm focused and still learning, all while creating a people-first infrastructure that outlasts me by helping teams scale talent, not overhead.

• The companies making money off the AI boom aren't the ones you might expect.
• Caterpillar, Cummins, and others have made billions of dollars selling generators to data centers.
• Some analysts have raised concerns about investors overpaying, while the companies foresee years of demand.

For the last century, Caterpillar's calling card has been its trademark macaroni-yellow backhoes and bulldozers. Increasingly, customers are interested in something else: the generators powering the artificial intelligence boom.

At hundreds of data center construction sites across the US, Caterpillar — alongside Cummins, Generac, and Rolls-Royce — supplies dozens of these machines. With some capable of powering more than 1,000 homes, generators ensure that AI models and cloud software can run without interruption.

These hulking devices, it turns out, are cash cows. While the surge in AI demand has fueled hundreds of billions of dollars in spending, much of that hasn't yet flowed to the usual suspects in Big Tech. Instead, it has ended up in the hands of builders, blue-collar laborers, and companies like Caterpillar and Cummins, which manufacture unsexy but crucial equipment.

Both companies have beaten the S&P 500 over the past year — and their stock prices have spiked even more than that of many major data-center customers, including Amazon, Meta, and Microsoft. Like the California Gold Rush's picks-and-shovels purveyors before them, industrial vendors are cashing in on the AI boom.

Plenty of non-tech companies have benefited from AI spending, including companies like Primoris and Dycom Industries that lay pipes and fiber that connect data centers to natural gas and the internet. Generator companies are particularly well-positioned: As hyperscalers race to build power-hungry data centers, the demand for backup and off-grid electricity has ballooned.

"Power generation is a business that has never seen this kind of scale, in any way, ever," Tom Shepherd, a data center executive at Cummins, told Business Insider.

Data centers are a bright spot for industrial businesses

About a decade ago, Shepherd oversaw a $50 million-a-year regional power-generation business in Oklahoma that largely sold Cummins systems to customers like hospitals and wastewater plants. He recalled just one large data center in his territory, run by Google.

"There just wasn't the scale," he said.

Today, Google's parent company, Alphabet, has more than two dozen data centers online or in development across the US. It has revised its planned capex spending for 2025 upwards every quarter, and now estimates that it will spend at least $91 billion by the end of the year. That's a fraction of the $7 trillion that will be spent on data centers globally by 2030, per McKinsey.

For manufacturing-oriented industrial businesses that have faced threats from tariffs, rising expenses, and persistently high interest rates, data center demand is a welcome bright spot. Business Insider previously used diesel generator air permits to estimate data center growth and found that if all permitted data centers went online, their electricity use could be as much as the entire state of Florida.

Sales of Cummins' signature truck engines have slipped over the past year, but the company sold $2.6 billion of power-generation equipment to the data-center industry last year and expects that number to grow by 30 to 35% this year, CEO Jennifer Rumsey said on a conference call last month.

The power-generation segment of Caterpillar — often shortened to Cat — grew from 8.4% of its total sales in 2021 to more than 14% of its sales in the first nine months of this year. And Rolls-Royce's power-systems unit reported record revenue in February, driven partly by 46% annual growth in data-center sales.

Their stock prices have also soared. Both hit all-time highs this week; Cat closed above $590 a share on Wednesday, topping the previous high set after beating earnings in October, and Cummins shares crossed $507 on Wednesday.

Analysts are generally bullish. According to information tracked by the companies, over the past year, the percentage of analysts who think investors should buy Cat's stock has grown from 41% to 54%; for Cummins, the share has gone from one-third to half.

Some analysts have sounded alarms, but the companies aren't worried

Some indicators suggest the market took the picks-and-shovels trade too far. In October, Morgan Stanley estimated that Cat's power-generation arm was valued by investors at 60-to-100 times its operating income. At Nvidia, the same multiple was 25 times; at GE Verova, another electrical-equipment business, the multiple was 28 times.

Melissa Busen, who runs the electrical power division at Cat, says she's not worried. The company's $39.8 billion order backlog is almost triple what it was five years ago, and Busen said the largest customers are making plans for years down the road.

If one of them has hardships, she said, another will take its place. She noted that Cat has been supplying the data-center industry since the early 1990s and has "been able to build true partnerships" with the new crop of hyperscalers.

Cat broke ground last year on a $725 million expansion of its engine factory in Lafayette, Indiana, where it currently employs 1,900 people. The expansion will more than double the company's output of the engines that data-center developers are demanding, and is set to come online in 2027. Expansions are also underway at Rolls-Royce plants that build data-center generator parts in South Carolina and Minnesota.

Data center construction in Asia and Europe is also a component of demand, and many generators are manufactured outside the US. Cummins said earlier this year that it planned to invest $200 million across its power-generation manufacturing sites in India, England, and the US. The company plans to double the production capacity of the giant, 95-liter engines that drive data-center generators.

The money isn't just in backup generators that sit idle most of the time, either. In the race for power, utilities aren't able to connect companies to the electric grid fast enough. Increasingly, Cat and Cummins have been called on by Microsoft, Amazon, hyperscalers like Coreweave, and regional players like Joule for natural gas-burning generators and turbines that can run 24/7 for months or even years.

Cat's Busen said customers don't just want to talk about engines and turbines — they want to talk about "a complete, integrated solution" to their electrical needs, including things like batteries, inverters, and control systems.

The biggest limiting factor for now, said Shepherd, is "the physical ability of our society to build this kind of infrastructure."

Have a tip? Know more? Reach Jack Newsham via email (jnewsham@businessinsider.com) or via Signal (+1-314-971-1627). Use a personal email address, a nonwork device, and nonwork WiFi; here's our guide to sharing information securely.

• Salesforce CEO Marc Benioff is rebranding many products under the Agentforce name.
• Benioff says customers prefer AI agent terminology over cloud, based on recent focus groups.
• Salesforce might rename the entire company Agentforce, Benioff told Business Insider this week.

Salesforce CEO Marc Benioff is done talking about the cloud and is renaming everything after its AI software Agentforce — perhaps even the whole company.

Salesforce slapped the "Agentforce" moniker on many of its products and services recently. And when the company reported earnings this week, there were updated names for several offerings. Sales is now "Agentforce Sales." The Service offering is "Agentforce Service," and Platform became "Agentforce 365 Platform," and so on.

A Salesforce employee recently told Business Insider that the company might change its official corporate name to Agentforce. This was mentioned partly in jest. However, during an interview with Benioff late on Wednesday, the CEO confirmed this is a possibility.

"It might," Benioff he told Business Insider. "That would not shock me."

There's precedent for such radical moves in the industry, as companies shift focus to new technologies or strategies. Mark Zuckerberg famously changed Facebook's name to Meta. Google's cofounders also changed their corporate name to Alphabet. Jack Dorsey's Square became Block.

From cloud to AI agents

Salesforce was born in the cloud, so that tech buzzword has been attached to many products over the years: Sales Cloud, Service Cloud, Data Cloud.

Now customers don't want to hear about the cloud anymore, Benioff said in the interview. He learned this during recent focus groups ahead of the company's annual Dreamforce conference in October. Instead, clients want to talk about AI agents.

"I dropped the word cloud totally," he said. "I did it at Dreamforce. Did you notice I never used it in the keynote? We learned in focus groups customers don't talk about cloud anymore. They just want to talk about their agentic interface."

Salesforce renamed Data Cloud, its data platform, to "Data 360" and announced the change during Dreamforce. Data 360 is central to the company's 2026 strategic plan, which Benioff also discussed with Business Insider during the interview.

Salesforce has made a huge bet on AI agents. Some insiders have cast doubt on the initiative, but Wednesday's earnings showed some positive signs as the company raised its revenue forecast.

Have a tip? Contact this reporter via email at astewart@businessinsider.com or Signal at +1-425-344-8242. Use a personal email address and a nonwork device; here's our guide to sharing information securely.

• Legal tech startup Harvey has raised $760 million this year, reaching an $8 billion valuation.
• Harvey has been widely adopted by law firms, and lawyers say it helps them deliver work faster.
• Now the startup needs to prove it can boost law firms' profits.

Legal tech startup Harvey this year has bagged $760 million in funding and soared to a valuation of $8 billion. Now it needs to prove that it can actually boost profits for the law firms that pay for its services.

Harvey, which on Thursday announced the close of a $160 million funding round, says its products free lawyers from drudge work so they can spend more time on strategy and client counseling. But at this point, Harvey's ROI shows up in feelings more than spreadsheets.

Law firms want the payoff to show up on financial reports and client bills, says Zach Abramowitz, a former lawyer turned technology consultant for the legal field. Right now, he says, the payoff "inures first and foremost to the user" — the attorney who suddenly feels like they're working with a supercharged second screen.

AI adoption has skyrocketed across the corporate world. Still, few companies are deriving measurable value from it. According to a Boston Consulting Group survey this year of more than 1,250 global firms, only 5% of companies were seeing real returns on AI.

Investors have thrown $3.2 billion at startups in the legal field this year, according to a Business Insider analysis. Harvey alone has scarfed up nearly one in four of those dollars.

Harvey builds software for analyzing and drafting documents using legally-tuned large language models.

The startup has proven that lawyers want what it's selling. Over half of the hundred highest-grossing law firms in the United States have Harvey licenses. And the startup is quietly making inroads with some of the world's largest enterprises, including Walmart and Comcast.

Every organization is seeking to nail down the value of artificial intelligence, said Winston Weinberg, Harvey's cofounder and chief executive. Time savings, he said, are the "first horizon of measuring ROI."

"The average Harvey customer is saving considerable time by using Harvey regularly," Weinberg told Business Insider in an email. "Longer term, industry-wide, you'll want to see metrics on transformation, profits per partner for firms, and ultimately revenue metrics for in-house teams."

To prove it isn't just selling workplace satisfaction, Harvey recently hired legal-intelligence shop RSGI to run a study of 40 Harvey customers, from A&O Shearman and Paul Weiss to the legal departments of AT&T and National Grid.

Most respondents said Harvey paid off in months, not years, with lawyers churning through work faster. And most said they'd be "upset or disappointed" if the tool vanished. One person joked: "Take away my coffee before my Harvey license."

What eluded firms was a clear way to turn all that enthusiasm into hard metrics.

Only about a fifth of participants — six law firms and two in-house teams — said they had a formal way to track the return on their investment in Harvey tools. Most said they relied on adoption stats and anecdotal "power user" stories to track Harvey's usefulness.

When asked how they measured value, 83% cited internal adoption, 75% intensity of usage, and 58% time savings. Only 18% pointed to cost savings.

One participant shrugged off the question with: "How would you measure the value of Microsoft Word?"

Harvey's customers, at least, seem patient. A third of law firms surveyed said building a formal ROI framework was a lower priority than simply giving lawyers access to Harvey.

Harvey has clearly answered the question: "Do lawyers actually use this?" The open question is whether usage translates into higher profits or just nicer Tuesdays.

Have a tip? Contact this reporter via email at mrussell@businessinsider.com or Signal at @MeliaRussell.01. Use a personal email address and a non-work device; here's our guide to sharing information securely.