Tag: Business

• Former GitHub CEO Thomas Dohmke joins Apiiro as a strategic advisor.
• Apiiro helps organizations monitor and secure AI-generated software code and codebases.
• Dohmke aims to enhance protections as AI coding tools increase cybersecurity risks for companies.

GitHub CEO Thomas Dohmke stepped down a few months ago. He's getting back into the arena now.

Dohmke was just tapped as a strategic advisor to Apiiro, a startup that helps organizations monitor the security of their apps and codebases.

He'll focus on helping Apiiro develop new protections for AI-generated software code.

As more AI coding tools churn out code, there's concern that this could make codebases more vulnerable to hacks and other cybersecurity risks.

Dohmke said developers now often use multiple AI coding agents, and these tools don't necessarily know all the policies, rules, and safeguards that their employers have set up to ensure their codebases and broader technology offerings are secure.

Apiiro's technology connects with companies' code-management systems, providing this crucial context for AI code generation, he said.

"That's the important part," Dohmke added, adding that Apiiro can spot and automatically fix such issues in code, helping developers comply with company guidelines, while not piling on more work for them to do.

Apiiro has raised more than $100 million from investors, including General Catalyst, Greylock, and Kleiner Perkins. CEO Idan Plotnik has launched and sold other startups in the past and has served as a cybersecurity executive at Microsoft.

Dohmke also comes from Microsoft, which owns GitHub. He said he met Plotnik about three years ago and was taken by the CEO's energy and Apiiro's mission.

Of course, the former head of GitHub has a lot of choices for what to do next. And while Dohmke gets a portion of Apiiro for his advisory services, he said he's more interested in what Apiiro is building.

As more employees use AI to develop digital products and prototypes, the startup's technology should become even more useful, he said.

Sign up for BI's Tech Memo newsletter here. Reach out to me via email at abarr@businessinsider.com.

• Technologists who specialize in AI are now Wall Street's most in-demand workers.
• The scramble is fueling a talent war with Big Tech and other sectors.
• Recruiters explain what banks and asset managers are offering to lure them.

The hottest job on Wall Street right now isn't a trader or dealmaker. It belongs to the human minds who specialize in building the digital ones.

Bank chiefs have been touting how adopting artificial intelligence at scale will help employees work smarter and cut costs, and they're putting serious money behind those AI ambitions. Goldman Sachs' CEO David Solomon said in October that he wished the firm's $6 billion annual tech budget was even higher, while Bank of America dedicated $4 billion this year to "new technology initiatives," according to the company.

A chunk of that investment is directed at building out battalions of tech professionals to realize those strategies. However, the demand for that talent is driving salaries for technologists who specialize in AI — from rank-and-file engineers to executive and C-suite tech strategists — to soar at banks, according to insiders.

"There's certainly not enough people" to keep pace with hiring demand, said recruiter Ryan Bulkoski, the global head of the AI, data, and analytics practice at the search firm Heidrick & Struggles.

Recruiters and executives across financial services companies told Business Insider the result is a full-blown talent war pitting Wall Street banks against hedge funds, private asset managers, and Silicon Valley startups.

The senior-level or C-suite AI leader who can steer corporate strategy "is the hottest job in the market" right now, forcing financial services firms to cough up "stretch offers," Chris Connors, a principal at the compensation consultancy Johnson Associates, told Business Insider.

"If you are a high-caliber talent, you are more desirable than virtually every other role in the market," Connors added, noting that some firms have extended special, non-standard awards to lure top-tier talent or prevent them from leaving. Competition across the industry is fierce, as hedge funds and other asset managers have even issued incremental equity or long-term awards, Connors said — they often range from $500,000 to $1 million, typically vesting across three to four years. Some firms have offered "significant upfront sign-on grants," often in excess of $200,000, delivered off the bat as cash or equity, he said.

For banks, especially, it has created a conundrum — how to extend big-ticket offers while adhering to their often rigid, standardized compensation bands. Hiring managers are forking over "some uncomfortable hiring packages" which are vaulting recipients into "a completely different stratosphere," Bulkoski explained, noting that top compensation packages for senior-level AI leaders are reaching the high seven- and even eight-figure range.

"Nobody's leaving for less than 30%" pay increases if they're coming from companies where they're already "heads-down and doing well," he said. "No one's going to move for the historical 10, 15, or even 20% bump in compensation."

Making money moves

AI hiring in the financial services industry grew in 2025 to 10 times its level at the start of January 2022, according to an analysis by workforce data provider Revelio Labs. To assemble the data, Revelio examined finance job postings for roles like "AI engineer," "machine learning engineer," and "computer vision engineer," among others.

Six of the largest US banks — including JPMorgan, Goldman Sachs, and Morgan Stanley — have collectively posted more than 2,000 AI roles in the past 12 months, Revelio's data showed.

That surging demand for AI talent is driving paychecks higher.

The average salary for AI professionals in finance, excluding C-suite roles, had risen to about $180,000 from $142,000 in 2020. That increase of more than 25% puts them on par with what conventional tech firms tend to pay for these roles, Revelio said.

JPMorgan and Morgan Stanley declined to comment, while others did not respond to requests for comment.

"What we have seen with banks over the past few years is they've had to really reframe and reimagine those comp ranges, especially for a lot of this type of talent," Ben Hodzic, head of North America at the recruiting firm Selby Jennings, told Business Insider. That's especially true as they're coming up against multistrategy hedge funds and high-frequency systematic trading firms that are "paying whatever they need to pay to get people," he said.

Seat at the table

But money alone isn't enough, insiders said. These technologists also want something else: power.

Insiders said the AI race has elevated the voices of technology chiefs at banks and asset managers, giving them a perch at powerful tables that steer firms' overall strategy. Bulkoski said that, for prospective AI chiefs, wielding greater levels of influence in the boardroom alongside other C-suite leaders is an appealing prospect unto itself. "It could trump another number on the sheet," he said.

The nature of the work is attracting a broad swathe of professionals eager to exploit the breach. "Go on LinkedIn and look at the number of people who just changed their title from data scientist to AI scientist," Bulkoski said. "It's kind of unbelievable."

Candidates for senior-level roles are skewing younger, he continued. Some AI-savvy talent in their early- to mid-20s are pulling high-six to seven-figure offers, he said. To suss out who is a fit and who isn't, Bulkoski said his team assesses "whether they have a real track record" of constructing original AI or machine learning products at scale, and if they can channel that knowledge into creating commercial or operational value.

"Most are in the early phases of developing executive maturity," he said, "but the best of them can already influence senior leaders by grounding every conversation in architecture, data, and measurable impact."

The ladder up

With access to the board and stepped-up influence and responsibility, some Wall Street insiders are betting they can offer something Big Tech companies generally can't: a steadier, more clearly defined climb up the professional ranks.

With tech firms consolidating and fears of an AI bubble mounting, recruiters say some technologists are increasingly drawn to the stability and hierarchy that the financial services industry is known for.

"There's been a lot less volatility with jobs in the finance industry compared to the tech industry," said Selby Jennings' Hodzic, adding that finance offers a more well-worn, clearly defined pathway. He said Wall Street had been insulated from the kind of mass layoffs that have roiled Big Tech. Amazon, for instance, cut about 14,000 corporate jobs in 2025.

Recruiters say the flames fanning the talent war aren't dissipating anytime soon, and that the next phase of the hiring race is already underway.

"In 2025, it's almost like Gen AI is in the past," Bulkoski said. "We're actually now working on searches that have 'agentic AI' in the title."

• Bobbi Brown said she believes that too many managerial layers in a company are 'dumb.'
• In WSJ's CMO Council Summit, she rejected the concept of layers and excess consultants.
• Jones Road Beauty reflects Brown's focus on simplicity, efficiency, and founder-led decision-making.

For Bobbi Brown, one of the perks of starting over after leaving her namesake brand was being "freed" from the constraints of a large corporation.

The beauty founder said she learned a valuable lesson from that experience: "Layers are dumb."

"The most important thing is I learned what not to do," Brown said at The Wall Street Journal's CMO Council Summit on Wednesday. "Not to waste time, energy, money."

When Brown launched her namesake brand in 1991, she was 34 years old and handled every aspect of the company.

"I used to interview every single person in the beginning. Every single person," Brown said on a recent "Master of Scale" podcast episode discussing the release of her recent memoir, "Still Bobbi."

In the years after she sold her company to Estée Lauder and signed a 25-year non-compete with the beauty conglomerate, that dynamic started to change, and toward the end of her time at the company, the brand became "very process-oriented." Brown described becoming increasingly excluded from hiring decisions as the company expanded.

Estée Lauder did not reply to a request for comment from Business Insider.

Consultants and KPIs

Part of Brown's lesson to not waste resources and energy is to "not hire a lot of consultants," she said at the WSJ CMO Council Summit.

When she founded her new company, Jones Road Beauty, she said she stripped away unnecessary layers. The founder said she's "narrow-minded with what needs to get done," but loves hearing others' opinions.

She also said she doesn't understand why traditional corporate measures of metrics, like key performance indicators, or KPIs, are needed to assess performance.

"You either did a good job and sales are good, or you did a bad job," Brown said.

Brown isn't the only leader who has expressed frustrations with corporate bureaucracy in recent years. After a spike in hiring during the pandemic, many Big Tech companies have spent the last few years cutting back staffing, trimming middle management, and flattening organizational layers.

Brown said at the WSJ CMO Council Summit that she's maintained the same entrepreneurial ethos she had as an early founder and that she's still the "same Bobbi" who was cold-calling from the Yellow Pages.

"I love being scrappy," she said.

This as-told-to essay is based on a conversation with Jenny Wood, a 45-year-old former Google executive who lives in Boulder, Colorado. She left Google in August 2024 and is now a keynote speaker, coach, and author. The following has been edited for length and clarity.

It seemed preposterous for me to ever think about leaving Google.

I started there in November 2006, when there were only around 10,000 employees, and became an executive — the director of American media relations — in 2022.

Google's amazing; I bleed Google colors. I loved the impact I was having, the future of opportunities I saw for myself, and the feedback I was getting as a leader. I'm also the breadwinner for my family.

I'd always thought I'd be at Google for another 15 years and would retire there.

I realized I couldn't sustain my life anymore

The moment that started the agony was when I was driving my son, who was 7 at the time, home from choir rehearsal in the dark, a 45-minute drive on winding roads.

Because of everything on my plate at the time — my role at Google, leading the Own Your Career program, navigating my book opportunities, and being a wife and mom — I was suffering from so much anxiety that it kept me up at night, feeling like I was letting everybody down and not doing anything well.

Mostly, I was incredibly sleep-deprived.

As I was driving, I was like, Oh my gosh. Did my eyes just flutter closed? I didn't actually fall asleep at the wheel, but it was a terrifying moment.

During my next session with my executive coach, I told her I couldn't sustain this anymore. I had taken on so many things in the name of success. She said, "Jenny, circumstances change."

Her words stopped me in my tracks and opened me up to the possibility of leaving — leaving well and quitting thoughtfully.

Thus ensued 18 months of back-and-forth about whether I should stay or go.

I used a spreadsheet to help me weigh the risks

I'm a left-brain thinker and approach the world in a very analytical way, so it was hard to feel in my gut that it was time to leave.

One thing that really helped was a spreadsheet I made, weighing actual risk against perceived risk. I broke it down into four components: physical risk, cognitive risk, emotional risk, and financial risk.

Physical risk included things like not sleeping at night, pain, and weight loss (which I gave a 1). Cognitive risk was mental stress, distraction, and mental drain (a 2). Emotional risk included potential for rejection, loss of connection with loved ones, negative self-talk, and fear (a 2). And the financial risk was things like paying my future mortgage statement and future earnings potential (a 2).

Breaking things down helped me get out of a catastrophizing mindset of thinking, This is a ridiculous idea, and made me think much more practically about how this might be possible.

I had to change my mindset to escape the golden handcuffs

The golden handcuffs are very real.

It wasn't just my salary, bonus, and equity; it was all of that future income as well. I would log in to my Google stock portfolio system — which tells you what you've earned and what you'll earn when your stock vests — and my palms would sweat. It was really hard to walk away from that number.

But ultimately, if you're in an executive role at any Fortune 500 company, you're probably making more than you need to live on. I guess it depends on your lifestyle; I live pretty frugally. Even so, I still couldn't imagine my income and net assets not continuing to go up and to the right every single year until I retired.

That was a mindset I had to move past.

It took my husband and me having seven conversations with our financial advisor — which ended up being more like therapy — for me to feel comfortable and confident that I could do this.

My advice on quitting well

I ended up leaving Google in August 2024. I cried after I turned in my badge and computer and as I drove away — happy, sad, and bittersweet tears. It was 18 years of my life!

When I came home, my husband and kids had written all these phrases that I'd said before through my leadership and coaching work, and arranged them in a heart shape on the window in the kitchen.

The heart is still up, 14 months later.

Quitting Google has been a massive change. I don't want to make it sound like it was easy; it was the scariest and hardest thing I've ever gone through in my professional life.

But I'd say I quit extraordinarily well. Here's my advice for others.

1. Mind your truths and tales

A truth is a verifiable fact, while a tale is a story you create to make sense around the facts. We often tell ourselves negative tales, and they don't serve us well because we believe what we think.

To get past my fears, I had to separate the truths from the tales, and then rewrite those tales to be more empowering.

For example:

Tale: I will lose my entire identity if I leave Google.

Truth: I'll no longer be employed by Google.

A more empowering tale: Part of the reason I'm leaving is because I want to have a huge book launch and possibly be a bestselling author. That's an incredible new identity to adopt!

Or, tale: We will run out of money and have to move to a smaller house, away from the gorgeous hiking trails that are behind our current house.

Truth: I will not get a paycheck with the Google logo on it every two weeks.

A more empowering tale: I've worked really hard to put myself in the best financial position possible to make this a reality.

One tale I told myself was that my kids would never forgive me for leaving Google because they love the secret game room, the climbing wall, and the free snacks and candy — Google's a really cool place for a parent to work. But I know they're really proud of what I'm doing.

What matters more to them is that now I'm done with work every day at 2:40 p.m., I drop them off and pick them up from school almost every day, and they're probably going to start traveling to places like Disney World and Vegas with me for keynotes now.

There's no question this was the right decision for my family.

2. Prioritize your dynamic dozen

Before I quit, I made a spreadsheet on my personal computer of people I wanted to stay in touch with and their email addresses. I was also posting on LinkedIn frequently and building an audience.

I set up what I call the "dynamic dozen" — 12 people you want to meet with in the next 12 weeks. This is great if you're trying to switch roles within your company, if you're looking for a new job, and also if you want to quit. It could be 12 people in 12 weeks, or 30 people in 30 days.

Mine was probably closer to around 60 people in 60 days, because I wanted to leverage all of the relationships I had: people who might want to bring me as a keynote speaker in the future, or people who might want to buy a hundred bulk copies of my book two years from then.

When you leave a company, your network always remains, so double down on that before you peace out. Have honest, intentional conversations, put time on someone's calendar, and reach out to people, even if it's been years since you had a working relationship.

I had to push past the fear that no one at Google would want to work with me once I was on the outside, that I'd be irrelevant. My work is a lot about how to thrive in a corporate environment, so I wondered, If I'm no longer in one, will any of my content still be valid?

Now, my number one client is Google. The vast majority of my coaching clients are Google employees, and a huge chunk of my speaking revenue is from Google speaking engagements or consulting.

3. Move, then map

Once, I was hiking in Montana with two friends, and the trail diverged into two paths. I'm always trying to optimize, so I started peppering the park ranger with all of these questions: "What is the perfect path? Which one will be more cardio? Which one is a lake view and which is a mountain view? Which trail is muddy?"

And from 50 feet ahead, my friend yelled, "Jenny, it's all beautiful! Just start walking!"

I'm always trying to map out everything perfectly — how much income I'd make, how quickly I could build a business, what I'd be if not a Googler.

You can't do that. Fear adds friction, which slows you down without actually minimizing risk.

Trying to map every little possible component also takes the joy out of the process. Action makes progress; thinking provides clarity. When you move and then map —or at least move and map in tandem — you're going to be set up for so much more success.

If you quit your job for an unconventional path and want to share your story, please reach out to this reporter at janezhang@businessinsider.com.

• Brian Zhan is a young VC who made early bets on Skild AI, Dyna Robotics, Periodic Labs, and Reflection AI.
• Zhan will be employing an unusually bold strategy of writing $30 million checks to seed-stage companies.
• Zhan believes VC is undergoing a transformation that requires firms to pounce with big checks very early.

At just 29, Brian Zhan, who practices what he describes as "nerdy investing," has already made early bets on buzzy AI companies now valued in the billions.

After quietly leaving Silicon Valley VC firm CRV this summer, Zhan recently joined Striker Venture Partners, where he and veteran investor Max Gazor plan to upend traditional VC playbooks with a $165 million fund that aims to make massive seed bets on the next generation of AI startups.

"We're only investing in 10 companies per fund, and we're going to show up with checks up to $30 million," Zhan said. "That takes conviction."

$30 million is what companies typically raise in a later funding round once they have achieved product-market fit, but Zhan believes VC is undergoing a radical transformation that requires firms to pounce with big checks when an early kernel of an idea presents itself.

"Seed-stage investing is moving earlier and earlier," he said. "Seed rounds today are often just 22-year-old founders with a great idea."

Zhan, who studied computer science at Northwestern University and started his career writing code at Facebook, is emblematic of a shift taking place in venture, where investors switch firms more often and technical backgrounds are prized above MBAs because many of the most promising new startups have few business metrics to evaluate.

"It was on the Facebook data team where I saw a pattern that really stuck with me," he said. "The best technical minds at the company were founding startups and then struggling to raise meaningful capital, despite having résumés that Silicon Valley normally drools over."

He started angel investing in his spare time, writing small checks into projects led by former Facebook engineers. At CRV, Zhan transitioned to a full-time VC role, making early bets on Skild AI, Dyna Robotics, and Periodic Labs.

"Today, these companies are some of the best-known companies in AI, but back at the seed round, a lot of VCs just didn't really know who these founders were," Zhan said.

His biggest win so far is the model startup Reflection AI, which he views as proof that VCs should not be afraid to pay up in seed rounds.

Zhan co-led Reflection AI's seed round just last year at a $200 million valuation, a very expensive price that many other VCs passed up. Last month, the company raised $2 billion at a valuation of $8 billion.

"Founders want someone who can speak their language," said Misha Laskin, CEO and cofounder of Reflection AI. "Brian is rare in that he's actually done the work, is incredibly well connected, and is insightful about what's happening at the frontier."

Excited about AI for science

Zhan grew up in Hong Kong, Beijing, and Palo Alto. His older sister is Stephanie Zhan, a general partner at Sequoia Capital, who was also an early backer of Reflection AI and Skild AI.

"I would say we have very similar tastes, and sometimes we land on the same companies," Brian Zhan said.

Asked if they will be discussing deals around the Thanksgiving table, he said: "No, we keep work separate."

While many VCs spend their days in coffee shops meeting as many founders as possible, Zhan finds it more useful to devote hours a day to reading the latest AI research curated by his own custom AI assistant.

He says he was drawn to reunite with Gazor, an MIT Ph.D. dropout who has appeared four times on the Forbes Midas list, because of their similar intellectual style.

"We spend all morning reading papers and trying to identify the trends and who has the most exciting ideas," Zhan said. "We take very few meetings every day. We just spend more time figuring out what the frontier is, and when we reach out to founders, we have already made up our mind to go back to them."

Gazor, who is probably best known for his early investment in Airtable, helped recruit Zhan to CRV in 2023 and wanted to work with him again.

"I saw him become one of the most trusted advisors to some of the most high-profile AI founders, and I was really in awe of how he related to them," Gazor said. "He's got a fantastic trajectory ahead of him."

Zhan says he is most excited about founders who are harnessing the power of AI for science, such as William Fedus, a former OpenAI researcher who recently cofounded Periodic Labs to automate scientific discovery.

"Imagine if we could use AI to dramatically shorten drug discovery timelines?" Zhan said. "I think that AI for science will become as big a category as AI for robotics is today."

• Amazon Web Services lost some AI customers to competitors due to capacity issues.
• Some clients shifted projects after AWS Bedrock didn't meet capacity demand this summer.
• AWS Bedrock faces rising competition as cloud and AI rivals offer better performance.

This summer, Amazon's cloud business struggled to keep pace with surging AI demand and missed out on real revenue for its flagship AI product.

Amazon Web Services's Bedrock service sits at the center of the company's AI push. It lets developers tap into powerful models, including Anthropic's Claude and Meta's Llama.

But over the summer, Bedrock hit "critical capacity constraints" that drove some customers to rival services, such as Google's cloud service, according to an internal July document obtained by Business Insider.

The shortages led to tens of millions of dollars in lost or delayed revenue. Epic Games shifted a $10 million Fortnite project to Google Cloud after AWS failed to provide enough quota for Bedrock, according to the document. (Quota limits control how much intelligence customers can access via AI cloud services).

Oil trader Vitol weighed moving some projects away from AWS, a decision that risked a $3.5 million revenue hit amid "prolonged quota approvals," the document also warned. Other customers, including Atlassian and GovTech Singapore, were waiting on quota increases this summer, delaying at least $52.6 million in projected sales, the document also disclosed.

Bedrock was "experiencing critical capacity constraints that are threatening customer adoption and potentially causing substantial revenue loss across multiple industries," the July document stated.

The fallout underscores the financial toll of AWS's capacity crunch, and it explains why the biggest cloud companies are rushing to build as many AI data centers as possible right now. High demand is a good thing, but if you can't satisfy this and customers go to rivals, that's a frustrating problem.

Indeed, Amazon CEO Andy Jassy has repeatedly stressed the need to ramp up cloud infrastructure, particularly AI chips and data center power. It's unclear whether the company has fully resolved these issues. Three current and former employees said the capacity crunch remained one of AWS's top concerns through September.

An Amazon spokesperson said Bedrock is "experiencing rapid growth" and AWS is adding capacity to meet that demand. Reviewing customer feedback is a core part of Amazon's culture, which helps the company improve its products and services, the spokesperson added.

"At Amazon, we're vocally self-critical because that's how we drive continuous improvement and deliver better results for customers," the spokesperson said in a statement. "This internal candor is a feature of our culture, not a flaw. We're grateful for all customer feedback—including challenges they encounter—because it helps us make Bedrock even better, and that's exactly how you build a scalable, sustainable business that serves customers well over the long term."

A Google spokesperson declined to comment. Representatives for Anthropic and Epic Games didn't respond to requests for comment.

'Accelerating capacity'

Expanding data center capacity, as with other cloud providers, is one of AWS's top priorities.

During an October earnings call, Jassy said AWS had been "focused on accelerating capacity the last several months," adding more than 3.8 gigawatts of power over the past year, more than any other cloud provider. AWS has doubled its power capacity since 2022 and plans to double it again by 2027, he noted.

Jassy added that Amazon will remain "very aggressive" in scaling up capacity to meet booming demand, noting that AWS can monetize new infrastructure almost immediately. Bedrock, he said, is already showing potential to grow as large as EC2, one of AWS's most successful cloud products and a key profit engine.

Part of Bedrock's shortages may stem from prioritizing large clients. In October, Jassy said most of Bedrock's workloads run on AWS's in-house AI chip, Trainium, but that usage so far has come mainly from "a small number of very large customers." He added that more mid-sized companies are expected to adopt the next-generation Trainium in the coming months.

Amazon is expected to reveal more details about Bedrock and its broader cloud strategy during its annual re:Invent conference in early December.

'Urgent need'

The July AWS document said the capacity crunch was hitting customers across industries, including finance, gaming, and tech. Companies such as HelloFresh, Zalando, and Ryanair were among those affected.

At the same time, "slow capacity approval and denial of spiky workload requests" prevented firms like Stripe, Robinhood, and Vanguard from moving AI workloads from Anthropic to Bedrock, the document noted.

"These constraints are forcing customers to explore alternative providers like GCP, OpenAI, and Anthropic, signaling an urgent need for AWS to address its Bedrock service quota and performance challenges to maintain competitive positioning in the rapidly evolving GenAI market," the document stated.

Quota limits in Bedrock are based on how many AI tokens you can process in a minute, or the number of API calls you can make in a given time period. (Tokens are how AI models break queries down into digestible data chunks. Industry pricing is based on how many tokens are processed. APIs are application programming interfaces, a common way applications share data).

In recent weeks, investors have grown uneasy over the tech industry's massive AI spending, with fears of a potential bubble weighing on markets.

Amazon's AI capacity issues are a double-edge sword here. On one hand, these challenges suggest customer demand is still very strong. On the other, it's another reason for big tech companies to keep spending heavily, potentially fueling the AI bubble even more.

Amazon has said it plans to pour $125 billion into capital expenditures this year, and even more in 2026. AWS revenue climbed to $33 billion last quarter, up 20% year over year, marking its fastest growth since 2022.

Performance issues

It wasn't just capacity woes driving customer workloads away from Bedrock. Latency and missing features also played a major role.

Customers using Anthropic's Claude models through Bedrock opted to switch to Anthropic's own platform or Google Cloud because of "ongoing capacity, latency, and feature parity issues," according to the July AWS document. Companies such as Figma, Intercom, and Wealthsimple were among those migrating their workloads "due to one or several of these challenges."

The UK's Government Digital Service considered a move to Microsoft's cloud because Anthropic's Claude 3.7 Sonnet model ran slower on Bedrock, the document added.

Thomson Reuters also chose Google Cloud over Bedrock for its CoCounsel AI product after finding AWS's service was 15% to 30% slower and lacked key government compliance certifications, the document showed. In May, executives raised these concerns with AWS leadership, including CEO Matt Garman and compute VP Dave Brown, leading both companies to agree to monthly review meetings.

Joel Hron, CTO of Thomson Reuters, told Business Insider that the company recently moved "one component of an AI workload to Google Cloud to prioritize latency." He added that Thomson Reuters still runs substantial workloads on AWS and Anthropic as part of its multi-model, multi-cloud strategy.

'Increasing competition' from Google

The July AWS document also noted that Bedrock was losing ground to Google's Gemini models, which boast five to six times larger quota limits and, in many cases, better performance.

When comparing accessing Claude via Bedrock against Gemini Pro, the internal report said the Google model outperformed "across multiple benchmarks." The document also noted that Gemini Flash, a smaller, cheaper Google model, "delivers comparable quality at a fraction of the cost." (And this was before Google's Gemini 3 launch, which improved AI performance further for the internet giant).

Some startups jumped ship because of this. Financial startup TainAI shifted 40% of its Claude workloads from Bedrock to Gemini Flash, saving $85,000 a day, while Hotel Planner was planning to move to Google Cloud or OpenAI, the document noted.

The broader concern, according to the document, is that AWS lacked a cohesive product vision for AI inference, the main space in which Bedrock competes. Rivals such as Databricks, FireworksAI, and Nvidia's Dynamo were quickly pulling ahead, it noted.

Without a clear strategy or compelling long-term vision, AWS risked missing out on one of the most lucrative opportunities in the AI market, it warned.

"We are still missing an inspiring long-term vision and a holistic strategy," the document 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.

• We asked top investors at firms like A16z and NEA to name healthcare's most promising startups.
• Most of their picks are using AI, automating in critical areas like healthcare hiring and payments.
• Meet the 18 healthcare startups investors have been watching in 2025.

Healthcare investors are chomping at the bit to fund hot startups tackling administrative burdens, workforce shortages, and high medical costs with new technologies.

We asked 10 investors from VC firms, private equity shops, and family offices to identify the most promising healthcare startups of 2025.

Top backers from firms like Kleiner Perkins, Andreessen Horowitz, and Oak HC/FT were asked to make two picks each: one healthcare startup from their portfolio, and one they have no financial interest in.

Nearly every startup nominated is utilizing AI in their products, from automating hospital back-end tasks to assisting patients with questions about their payments. Healthcare AI startups have dominated digital health VC funding so far this year, capturing nearly $4 billion of the $6.4 billion raised across the industry in the first half of 2025, according to Rock Health.

Here's the full list of the most promising healthcare startups of 2025, according to investors.

In July 2024, Magdalena Robinson was laid off from her job as a vice president of talent acquisition at a media agency. The news came as a shock — by all accounts she was doing well — but she saw a silver lining: Maybe it was finally time for a career break. She took the rest of the year off, soaking in the extra time with her husband and teenage daughter. Then she started applying for jobs in January.

Eleven months and 300 applications later, she still doesn't have a single offer.

"I don't know what's going on right now," she tells me. She has a great resume lined with reputable employers and a steady ascension to bigger titles. And as a recruiting executive, she thought she knew how to navigate the modern job market. Was it her age? Her experience? Some black hole sucking up all her applications before they ever reached a human being? "I think about what I've done in my professional life," she says. "And I'm like, am I that unemployable?"

It's a question I've heard from dozens of professional jobseekers I've spoken to over the last two years. They're accomplished and articulate and seem like exactly the kind of candidates companies say they want. Why are so many of them struggling in an economy with low unemployment?

One explanation is the unusual nature of the current market. Many businesses stopped hiring after overstaffing in the pandemic, leaving too few openings to go around — especially in white-collar, middle-manager roles. But a hiring slowdown alone isn't enough to capture just how hellish the white-collar job search has become. Something deeper is going on.

At the heart of it is Robinson's application count: 300. A few years ago, a number like that would have floored me. Now, it's utterly ordinary. Two other people I spoke to recently told me they've applied to more than a thousand roles.

To see how bad it's gotten, I asked Greenhouse, one of the leading providers of hiring software, to take a look at their data. Last quarter, the average job opening received 242 applications — nearly triple the amount in 2017, when the unemployment rate was at a comparable level. That means that when someone submits a resume today, they have the abysmally low 0.4% chance of actually getting the job. (That's just for the average job. At the big-name employers, the odds are even worse.) Applying to a job in 2025 really is the statistical equivalent of hurling your resume into a black hole.

You'd think employers would be loving this, getting to pick from such a big pool of candidates. But they're just as frustrated. Recruiters are swamped with resumes, many from people who aren't remotely qualified for the roles they're hiring, and they're completely unequipped to sort through the deluge. The result is something far more dangerous than a hiring slump. Gridlocked to the point of paralysis, the job market isn't working.

"Nobody's happy with the current situation," says Greenhouse CEO Daniel Chait. "Something broke in the technology."

---

This isn't the first time a market's grown so overcrowded it stopped functioning. Economists even have a name for it: congestion. Big markets hold the promise of creating better matches, but they also tend to devolve into total chaos.

"Congestion is the bane of a lot of markets," says Alvin Roth, a Nobel Prize-winning economist at Stanford who's helped design programs to better match students with schools, organ donors with patients, and hospitals with new doctors. "Successful marketplaces have to fight hard to defeat congestion."

Take what happened in the early days of the romantic equivalent of job sites, online dating. Gaining access to a seemingly infinite number of potential partners, men bombarded the inboxes of the hottest women. Overwhelmed, those women didn't reply. So the men messaged even more women, sending generic "hey"s they could recycle over and over again. The women hated that even more. The result? A vicious cycle of spam and rejection. Few people found love.

Over the last three decades, something similar has played out in the labor market. Before the internet, job seekers had to pore over the classifieds in newspapers, trek to Kinko's to print out their resumes, and mail them off — and then wait weeks or months to hear back. Job sites like Monster and CareerBuilder changed all that, centralizing job listings into one searchable forum and letting people apply on the spot. Every innovation since then — from email job alerts to LinkedIn's Easy Apply button — made applying ever more seamless. That very convenience, though, also encouraged people to send out more applications. Recruiters tried to keep up with the influx, learning to skim resumes in mere seconds.

On Greenhouse, the applications-to-recruiter ratio is now 500 to 1, four times what it was four years ago.

The tipping point, Chait says, came when ChatGPT arrived just as companies slammed the brakes on hiring. Faced with fewer callbacks, candidates turned to AI to churn out more applications, customizing resumes and cover letters to each job at the click of a button. Some even deployed bots to mass-apply on their behalf. In a recent Greenhouse survey, 74% of candidates said they use AI in their job search. The surge in volume increased competition for each job even further, prompting people to cast wider nets and apply to roles they were less suited for.

Recruiters, meanwhile, were left to manage the flood alone. And as companies shrank their HR teams in the white-collar downturn, the workload only grew. On Greenhouse, the applications-to-recruiter ratio is now 500 to 1, four times what it was four years ago. Stretched thin, recruiters resorted to emergency triage — say, skimming only the first 100 resumes, or considering just the candidates with referrals. That's come at the expense of missing out on so many qualified candidates. Which has led to a bizarre observation I'm hearing from more and more companies: They're getting no shortage of eager applicants, but they're still having a hard time finding good hires.

"The forces that make it cheap to send more applications are working faster than the forces that allow you to quickly process many applications," says Roth. "We're deep into congestion."

---

Chait started seeing the warning signs in late 2022. But it was last year that he remembers thinking: "OK, this is really happening." Suddenly the crisis was everywhere: in Greenhouse's data, in its surveys of jobseekers, in conversations with customers. So he marshalled his staff to try every fix they could imagine.

One of the first was a new website where candidates can manage their applications to Greenhouse's clients. There, it introduced a feature called Dream Job, which lets people mark one application a month as a job they especially want. The idea is that recruiters don't just want qualified applicants. They want to know — amid the sea of people applying with a single click — who's actually serious enough that they'd likely accept an offer.

Online daters might recognize the concept as the "rose" on Hinge or the "super like" on Tinder — gestures borrowed from a landmark study in market design. Dream Job launched in June, and the early data is promising: Employers have been five times more likely to hire Dream Job applicants than standard ones.

Human beings are still good at a lot of things machines aren't. We're sort of wired to figure out who we are like and who we should like.Alvin Roth, Nobel Prize-winning economist

Other intermediaries of the job market are trying their own fixes. LinkedIn, for instance, introduced its own "rose," called Top Choice, to its premium members (Top Choice candidates, the platform says, are 43% more likely to get a recruiter message). It also shows people whether they're a high, medium, or low match for the roles they view ("try exploring other jobs," it gently advises low-match candidates). And this year it's been testing daily limits on Easy Apply submissions.

While these measures might help on the margins, to me they feel like slapping band-aids on a severed limb. In the age of AI-generated mass applications, companies will never be able to keep up until they have tools that can evaluate candidates at scale — and so far those tools have been lacking.

Job seekers now widely assume employers are already using AI to reject them. In reality, many of the companies I've spoken to have been hesitant to adopt AI screening, fearing they'll run afoul of regulations banning potentially discriminatory hiring practices. Their caution makes sense: Already, a group of job seekers has sued Workday, alleging its AI-powered technology disproportionately screens out older candidates. In May, a federal judge let the case move forward as a collective action, paving the way for more people to join the lawsuit. "We deny the allegations in the suit," a Workday spokesperson wrote in an email. "Our products, both AI-enabled and not, are built to help our customers manage this increasing volume of applicants with a focus on human decision-making."

Greenhouse is taking a different approach. It's testing a tool that, for every job opening, allows hiring managers to essentially teach the system the kind of candidates it's looking for, ranking which attributes matter most. The company plans to roll it out to a broader set of customers next year. "We can help you do it more rapidly with an algorithm," Chait says. "But it's not like there's a secret Greenhouse algorithm that's going to tell you who to look at."

If that works, it could be an improvement on the current market, where so many qualified resumes go unseen. Still, it assumes hiring managers know what they're looking for — and that what they want is in fact what the company needs. Roth, who's spent much of his career studying different labor markets, notes there's still so much we don't understand about how we decide who to hire — especially the intuition behind those decisions that might be hard for software to replicate. "Human beings are still good at a lot of things machines aren't," he says. "We're sort of wired to figure out who we are like and who we should like and who's on our team and could be on our team."

Robinson, the recruiting executive, prided herself on that very intuition throughout her career. For her, a good hire doesn't emerge from a mechanical formula of credentials — it's about recognizing the potential of who the candidate could become. "You can't be so close-minded to say, 'Oh, you have to check all the boxes,'" she says.

Being on the other side of the table now, it's the kind of potential she longs for someone to see in her. But every rejection has made it harder and harder to hold out that hope, and in this broken job market, she can't afford to wait forever. For now, she's taken a job at a grocery store so her daughter won't have to quit figure skating. "It breaks my heart that I can't provide what I was providing," Robinson says. "I just want to work."

---

Aki Ito is a chief correspondent at Business Insider.

• Michael Burry of "The Big Short" has pivoted from investing to financial writing.
• Burry's first two posts on his new Substack discuss his history as a blogger and skepticism of AI.
• The market veteran warns of overinvestment and says Nvidia is the Cisco of this tech boom.

Michael Burry of "The Big Short" fame has pivoted from investing to writing, launching a paywalled Substack called "Cassandra Unchained."

The blog is now Burry's "sole focus" and promises a "front row seat to his analytical efforts and projections for stocks, markets, and bubbles, often with an eye to history and its remarkably timeless patterns."

Burry has published two initial posts, one titled "Foundations: My 1999 (and part of 2000)" and the other titled "The Cardinal Sign of a Bubble: Supply-Side Gluttony."

The former recalls his time as a neurology resident at Stanford University Hospital, where he wrote about value investing at night.

"As I devote myself to Cassandra Unchained, I find myself on an old road not taken," it reads. "I feel lucky, and I am grateful for the opportunity as I walk it again."

The second post aims straight at the heart of the AI boom, which he calls a "glorious folly" that will require investigation over several posts to break down.

Burry goes on to take aim at a common argument about the difference between the dot-com bubble and AI boom — that the companies 25 years ago were largely unprofitable while the current crop are money-printing machines.

At the turn of this century, Burry writes, the Nasdaq was driven by "highly profitable large caps, among which were the so-called 'Four Horsemen' of the era — Microsoft, Intel, Dell, and Cisco."

He writes that a key issue with the dot-com bubble was "catastrophically overbuilt supply and nowhere near enough demand," before adding that it's "just not so different this time, try as so many might do to make it so."

Burry calls out the "five public horsemen of today's AI boom — Microsoft, Google, Meta, Amazon and Oracle" along with "several adolescent startups" including Sam Altman's OpenAI.

"And once again there is a Cisco at the center of it all, with the picks and shovels for all and the expansive vision to go with it," he writes, after noting that Cisco plunged 78% during the dot-com crash. "Its name is Nvidia."

Burry finished that post with a quote from Buffett's late business partner, Charlie Munger: "If you go around popping a lot of balloons, you are not going to be the most popular fellow in the room."

The market veteran unveiled his blog on X, where he goes by "Cassandra" in reference to the priestess from Greek mythology who was cursed to speak true prophecies but never to be believed. Warren Buffett described Burry and John Paulson as "Cassandras" for predicting the mid-2000s housing market crash that ignited a global financial crisis.

Burry recently terminated Scion Asset Management's SEC registration, closing his hedge fund to outside cash. He returned to X after more than two years in late October, marking his comeback with a coded post suggesting the AI boom is a bubble and the "only winning move is not to play."

• A Swedish startup is making a low-cost interceptor to counter drone swarms.
• The single-propeller Kreuger-100XR, designed to loiter in the sky, costs just a few thousand dollars.
• Nordic Air Defense was recently selected by European authorities to compete in a contest.

Sleek, black, and sporting foldable wings, the footlong Kreuger-100XR looks almost like something you'd find in a toy store aisle.

The rocket-shaped device is what a new Swedish startup hopes will be the future of point air defense against drones.

The Kreuger-100XR, with XR standing for Extended Range, is Nordic Air Defence's debut product. It's a mostly carbon fiber interceptor that weighs about a pound and is meant to be deployed in large numbers as an inexpensive, no-frills way to hunt down and destroy uncrewed aerial systems.

European authorities are taking notice. NAD is barely two years old, but was one of four finalists selected this year to compete in a counter-drone contest held in Portugal by Frontex, the European Border and Coast Guard Agency.

Startups like NAD are one of the many ways Europe is rebalancing to face Russia's ability to mass-produce long-range attack drones. They'll compete with leading contractors that have long dominated the market, such as the French-headquartered MBDA, which is making its own eight-motor interceptors to be rolled out at scale.

The startups have an edge up their sleeve. Defense primes, though more established and better connected, typically charge a premium and are often seen as large, cumbersome organizations. NAD has about two dozen employees, while MBDA and US-headquartered Lockheed Martin, for example, have 18,000 and 121,000, respectively.

Fastest in class, NAD says

NAD says the Kreuger-100XR's propeller can push the craft to speeds above 220 mph, and that its optimal range is about two miles at an altitude of around 3,300 feet.

That distance is relatively short for typical air defense, but NAD is banking on low cost as a selling point.

The firm declined to disclose the exact price of the interceptor, saying it costs a few thousand dollars each. Traditional short-range intercept missiles, by comparison, can typically cost anywhere from $400,000 to millions of dollars each; the shoulder-fired Stinger, for example, costs $480,000.

Jens Holzapfel, NAD's director of business development, told Business Insider at the company's Stockholm office that the interceptor is designed with simplicity in mind.

A police officer should be able to easily carry several Kreuger-100XRs in a backpack and then launch them from a handheld device.

"In testing, we can either throw it like a dart or use a catapult," Holzapfel said, balancing a prototype on his fingers.

He added that only someone with an exceptionally strong arm can hurl the interceptor fast enough for an effective launch.

Making cheap counters to drone swarms

Holzapfel said the Kreuger-100XR's main cost goes toward two key components. The first is a camera that enables the interceptor to autonomously acquire its target, while the second is an optional 250-gram explosive warhead for destroying larger aircraft such as the Shahed-136 attack drone.

A cheaper version with a laser seeker, in lieu of a camera, is also planned. It will require a laser designator, whether operated by a human or an autonomous device, to track its target.

"With the XR, its wings also allow us to loiter, so we don't need a straight-shot scenario," Holzapfel said. With a radio connection, the interceptor is designed to remain airborne for at least 20 minutes before acquiring a target.

In that sense, the Kreuger-100XR can act like an autonomous drone — and that's NAD's long-term vision. A cheap, fire-and-forget interceptor could become a fundamental defense against enemy drone swarms, an anticipated threat for which militaries around the world are preparing.

"The economy of war is at such a point that it's about scalability," Holzapfel said. "If we are being attacked by a swarm of drones, we need a swarm of counter-drones."

Ukrainian drone manufacturers have also been developing their own interceptors: quadcopters capable of flying fast enough to catch the Shahed.

"But they tend to be slower or more expensive," Holzapfel said.

While many interceptor drones use four motors, which can be more costly, the Kreuger-100XR uses a single prop engine, helping NAD drive down costs and increase the device's speed, he said.

Yet it may be some time until a Kreuger-100XR counterswarm becomes a reality. Holzapfel said NAD is still working on friend-or-foe programming to prevent multiple interceptors from targeting each other.

Shaheds and recon drones

The firm is positioning the Kreuger-100XR to fight the Shahed, which Russia has been using in droves to bombard Ukrainian cities and overwhelm air defenses.

Kyiv also needs more defensive options against unjammable, small drones that use cable connections, as well as medium-range reconnaissance drones like the Orlan.

NAD plans to conduct trials in Ukraine soon, but has yet to test its product in battle. An operational altitude of 1,000 meters will also be a challenge for the Kreuger-100XR, because many Russian Shaheds and Gerbera decoys are known to cruise at 2,000 meters or above before diving down at their targets.

But Holzapfel said the Kreuger-100XR's "fastest-in-class" speed of 220 mph can still give it an edge against the Shahed-136, which typically flies at speeds around 115 mph.

"We are aiming toward the Shahed scenario for sure," he said. "And that's why we're also considering the air trail scenario, where we'll be chasing Shaheds from another airborne platform and attack them even from above."

Or the Kreuger-100XR could intercept the drones during their terminal attack, though it would be harder to accurately strike the Russian drone in that scenario because of their higher speeds.

To build an interceptor

Holzapfel said NAD is completing its prototyping phase and moving toward mass production. Although most of the Kreuger-100XR's manufacturing should be automated — particularly for the fuselage and wings — it will still require humans for some assembly tasks, such as mounting the explosive warhead.

NAD aims to establish initial manufacturing operations in Sweden, but anticipates that it may be asked to build the Kreuger-100XR on-site for foreign clients.

"If we're in Britain, they probably want to have Brits assembling these," Holzapfel said.

The startup is planning for its supply chain to have as little dependence on China as possible, sourcing off-the-shelf components from manufacturers in Europe, North America, Japan, or South Korea.

For now, NAD makes its interceptors on the ninth floor of an office building in Stockholm, where its team of 23 sits among a mix of 3D printers, prototypes, and workshop tables.

Nestled in the corner of a room is one of the startup's test-launch catapults: a few metal poles and some string.

After its founding in March 2024, NAD initially worked on another interceptor. The Kreuger-100XR's predecessor, the Kreuger-100, is a propeller-driven, wingless device designed to eliminate drones solely by hitting them in flight.

However, it's ideally suited for smaller, commercial drones, such as the hobby drones that Ukrainian and Russian troops face regularly on the front lines today.

Holzapfel said prospective clients were more interested in capabilities for fighting larger targets, such as Shaheds, and so the firm pivoted toward the XR variant during the summer.

So far, NAD has raised about $4.4 million and is looking to begin another seed round early next year, he added. Afterward, the firm hopes to recruit software engineers to train artificial intelligence that can manage a defensive swarm.

Despite pressure in the industry for startups to send their products to Ukraine as soon as possible, Holzapfel said NAD is waiting until it's sure that the Kreuger-100XR can deliver.

"Everyone is saying: 'You've got to be in Ukraine, you've got to be in Ukraine.' But we don't want to just be tourists and waste anybody's time," he said. "We weren't going to be there until we were really confident in our product."