AI’s race to value: A conversation with Abridge, Anthropic, Perplexity, and Scale AI   

When we asked top CEOs about the state of the AI market, one sentiment rang clear: “It’s still so early.”

But if the evolution of the cloud economy taught us anything, tomorrow’s AI giants will be defined by more than just superior technology — they’ll need distribution and sustainable business models. This year, a new generation of AI startups across the model and application layers are exploring innovative approaches to monetization and go-to-market strategies.

"It's still so early"

So, on the Cloud 100 stage, we gathered the leaders shaping the future of artificial intelligence: Daniela Amodei, co-founder and president of Anthropic; Aravind Srinivas, CEO and founder of Perplexity; Shiv Rao, CEO and founder of Abridge; and Alexandr Wang, CEO and founder of Scale AI.

Together, these top AI leaders shared insights into the realities facing executives and builders today. While the technology advances at breakneck speed, AI adoption faces hurdles: usability, ethical, distribution, and retaining customers. 

To overcome these challenges, startups must focus on niche applications, design user-centric experiences, and navigate the infrastructural complexities required to deliver enterprise-grade solutions. Ultimately, success in the AI space will come down to the ability to compound value over time, adapt to evolving market needs, and carve out meaningful differentiation in an increasingly competitive landscape.

_{Dive into the conversation on the state of the AI industry — quotes have been lightly edited for clarity and concision.} 

As capable as AI already is, the tech is still nascent

What changed the most about AI in the last few years? Perhaps it’s how user-friendly LLMs have become. Machine learning and algorithms have long been part of the software technology stack. But prior to 2022, they were the provenance of researchers and technical teams. Now, most people use them daily. Perhaps that has caused everyone to think they know something about them. We asked our founders, what do you most wish more people understood about AI right now? 

First off, there is just how complex and difficult it is to staff a team with the right skill and technology to put an AI tool into market. “I’m not sure people appreciate the amount of alchemy it takes to actually deliver an enterprise-grade solution,” says Shiv Rao, founder and CEO of Abridge, which offers an AI tool for clinicians. “The ingredients range from computer science and data models to input from professors, PhDs, domain experts, and people testing it in the wild.” 

"There is more work ahead of us than behind."

Daniela Amodei, founder and president at Anthropic, stresses how much further the industry has to go. “Even though the numbers are staggering, there is more work ahead of us than behind,” she says. At Anthropic, she is thinking hard about what it takes to imbue all the ethics, security, and trust necessary into today’s Claude models.

Alexandr Wang, founder and CEO of Scale AI, which offers AI infrastructure, stresses that the progress is far more rapid than is obvious to consumers. “You should not expect to see any slowdown in the performance of these models,” he says. “It’s an incredible place of technological progress.” 

Meanwhile, Aravind Srinivas, CEO of Perplexity points out that adoption of intelligent search isn’t where it needs to be. “It’s early, and people are not using these tools regularly. They need to be more intuitive,” he says. “Our goal at Perplexity is to take all the amazing models everybody else builds and shape them into amazing consumer experiences.”

AI is developing faster than industry can absorb it

There is actually a top speed at which companies can absorb money- and time-saving AI technologies, says Daniela, and we are nearing it. 

“We’ll come up with new research areas and interests and features, but then you have to turn them into something that is quickly usable by businesses,” she says. She likens it to working with both your hands, and one of them moving considerably slower than the other.

Alexandr feels that society hasn’t even discovered the best use for AI yet, and to him, that’s an exciting premise. “We don’t yet know the giant, killer applications, and I think that means the infrastructure layer is going to see many more years of growth.” It also coincides, he says, with a renewed period of global and governmental change, which will accentuate institutional interest in AI. “As I see it, the fate of the free world hangs upon whether or not AI is that defining innovation for this era. I really hope governments understand that.”

Regulated industries like healthcare are adopting AI the fastest

“I don’t think any industry is adopting AI as quickly as healthcare right now,” says Shiv. “We are announcing a new health system every week and they are adopting it because we’re in a public health emergency. Two out of five doctors don’t want to be doctors in the next two to three years, and over a quarter of nurses don’t want to be nurses in the 12 months.” That rate of burnout and attrition is causing healthcare organizations to scramble for ways to improve the quality of care and clinician experience. 

Abridge, for example, uses generative AI to transform patient-clinician conversations into real-time clinical notes, offering immediate benefits: less documentation, better patient engagement, multilingual support, and improved clinician well-being.

“The most profound feedback we’ve gotten from primary care physicians is that for the first time in their 20-30 year career, they’re making eye contact with patients,” says Shiv. 

“We hear the exact same type of stories,” echoes Daniela. “So many healthcare practitioners are exhausted and burned out from administrative work. We’re also seeing industries typically slower to adopt technology flocking to AI. Companies with big, legacy code bases, such as financial services industries, which are built on trust.” In those scenarios, AI means experts can spend time with customers.

Industries are not monoliths and AI startups are learning to niche down

Founders in a space like healthcare well know, a hospital is not a clinic is not a dental practice is not a medical device supplier. Vertical industries and unique contexts introduce new challenges for AI startups training models on specific data sets to accomplish specific tasks. But it’s also an opportunity to niche down. 

“You have those really large health systems and then on the other end, you have the doctor down the street,” says Shiv. They are two very different personas with very different needs,” not to mention tolerances. “In big health systems, the bar for ‘good enough’ is very high. Every doctor there is an ‘edge case assassin’ — you get one thing wrong and you’re dead to them forever. But that doctor down the street, good enough may be enough. They just need relief.” 

The perennial challenge remains: distribution

When Aravind first started raising money for Perplexity, he was confident he could build a better product based on a new model that had less of a conflict of interest around the user experience than its established competitors. But a product is not enough, as AI companies learn repeatedly. 

“Investors said, yes that is cool but how are you going to get distribution?” recalls Aravind. “This is the often-told story of how Sundar Pichai became the head of Google. Everybody else focused on flashy products, while he focused on getting Chrome to become the dominant browser. This is why we are partnering with telcos, LinkedIn, Uber, and more.”

When switching costs are low, experience is everything

Aravind sees the value of Perplexity as answering people’s questions better than other search providers. People will settle on the experiences that are the best. “We always pull multiple relevant sources from the web,” he says. “That helps us be far more accurate, especially in matters of real-time knowledge.” That narrow focus has fueled their growth: Last year, they answered 500 million queries, says Aravind. Now they do this on a monthly basis. Meanwhile, they seek to monetize in ways other than the traditional sponsored search engine links — such as sponsored questions, or a publisher program.

But ultimately, they are just figuring out the details: “It’s an experiment, right?,” says Aravind. “Nobody’s really trying different models outside of subscriptions. Our belief is that if this succeeds, we can compensate content creators too by sharing the ad revenue.” 

Immature markets are forgiving of new models

In mature markets, the go-to-market motions must be more carefully planned. But in a market like today’s — rapidly expanding with unexplored territory — what may matter more is simply getting started. 

“We’re so early in the AI industry, I think a lot of business models will work. So will many approaches to getting users,” says Alexandr. “Whether it’s product-led growth, top-down sales, government sales, the model layer, the infrastructure layer, a bit of everything will succeed. If we jump forward 10 years, I think it’s obvious AI will be ubiquitous.” There is, in sum, vast room for growth and many players.

But as we’ve seen with other technological paradigm shifts, Alexandr thinks the first movers will still hold the advantage in the AI era. He equated these dynamics to the early days of the Internet and the Browser wars. “I talked to a VC about Netscape,” Alexandr says, “who told me, ‘Yeah, we originally thought we were going to monetize with the browser, then Internet Explorer crushed us. We built a wholly new business, servers, and that ended up being way bigger than the browser business we had.’”

The takeaway? Many major businesses are yet to be discovered as technology evolves and new business models emerge. However, the AI players that establish a foothold in a market and solid distribution early on will secure a competitive edge. They’ll gain the flexibility to explore new monetization strategies and continue growing.

The best response to the retention challenge? Know thy customer

Retention is a challenge in nearly every digital market, but especially those with low switching costs. How are AI startups retaining those first-time users? Perplexity is focused on getting various cohorts of first-time users to make a second query. “We track two types of retention — by week and by demographics and whether people make another query in their lifetime. Initially, we were at 80%, and I said we need to push this as close to 100% as possible. Then we’ll focus on people making five queries, and it becomes a habit.”

For Daniela and Anthropic, success is about narrowly knowing and serving their specific audience. “Claude users tend to be highly educated and use it for work or related tasks. So I joke, we’re not in the business of making snowboarding cat videos. Those are very fun. But that’s not what our user base is asking for. They want help with their writing, analyzing information, and looking through legal documents, etc.” 

If competing against a goliath, win on one narrow use case

Perplexity is competing against the largest search engines in existence. How do they plan to win? To compete only in areas where they can win, which deliver incremental value to the user. “One thing is very clear and that’s that we’re not trying to take the navigational behavior away — Googling words like ‘Reddit’ or ‘Costco,’ they can have that. That’s probably a billion searches a day by itself,” says Aravind. Instead, he and the team are targeting the detailed, complex queries where an engine like Perplexity has an advantage. “We would love to be a majority player there.” 

The Perplexity team is also careful to compete on exclusively what matters to their user — the actual experience. “Earlier, whoever tried to take on Google would attack them on privacy or monopoly practices,” says Aravind. “Consumers didn’t really get a differentiated experience from that. But now that's possible because LLMs really started working well on instruction tuning, and conversations, and being able to summarize without hallucinations.” Users are switching for situations where that really does work better.

For all their differences, every AI startup has two choices

“It’s go public or sell. It’s asymptotic,” says Aravind. “For that, you have to be profitable, or on the path to profitability. It’s better if you don’t have to pump lots of sales motions into that; the margins will be higher. If you can really change people’s daily habits, and the product spreads to a good chunk of the world, and you can monetize a good chunk of those queries, that’s what we’re trying to do.”

Many AI startups have a tailwind in this respect — they benefit from rapid advances infrastructure, from competition driving down the price of models to open source removing barriers. But then the question is, how do you turn that into an advantage over others? 

“The question it all comes down to is, can you compound?” asks Alexandr. “AI can. I think that’s what Aravind was getting at. The difference between whether or not you become a trillion-dollar company or not is, do you have the ability to just compound over an extremely long time? And I think that’s very possible with AI.”