Microsoft’s cloud is ‘no-brainer’ for some AI startups, helping Azure gain ground on AWS

EzDubs, a developer of language-translation technology, got started the way many tech startups get off the ground. It launched on public clouds from Amazon and Google.

However, after EzDubs went through the Y Combinator startup program last year, the company made a quick pivot, adding Microsoft’s cloud into the mix. That’s because EzDubs’ founders learned of a partnership that enabled Y Combinator companies to receive $350,000 worth of credits on Microsoft Azure.

It was a “manna from heaven message,” EzDubs co-founder Padmanabhan Krishnamurthy told CNBC. The credits were particularly useful, Krishnamurthy said, because Microsoft has been at the forefront of the artificial intelligence boom, investing in OpenAI and hosting scores of projects that use the company’s large language models (LLMs).

On Azure, EzDubs was able to obtain access to the advanced graphics processing units (GPUs) needed for a new wave of AI model training, which no other cloud provider could match.

“At that point, it was a no-brainer,” said Krishnamurthy, who co-founded his company in 2022 as generative AI was taking off. “It was the exact setup we needed,” with GPU availability that “literally no one else had.”

EzDubs’ story can be heard in various forms from startups across the AI landscape. While Amazon Web Services maintains the market lead in cloud infrastructure and Google remains a popular option for companies utilizing multiple clouds, Microsoft’s perceived strength in AI is giving the company an edge, at least when it comes to startups.

In Amazon’s second-quarter earnings report on Thursday, the company said AWS revenue increased 19% from a year earlier, trailing Microsoft’s 29% growth for the latest period, though that includes other cloud services in addition to Azure.

AWS was the first of the cloud providers to dole out credits to young companies, hoping that they’d be hooked by the time the credits ran out and eventually turn into big spenders. AWS’ Activate program started in 2013, following the launch of key EC2 (compute) and S3 (storage) services in 2006, and helped cement Amazon’s dominance in public cloud.

Microsoft’s access to hefty GPU clusters coupled with its long history as an enterprise technology company ubiquitous inside IT departments is changing the narrative. And, of course, money matters.

In November, Microsoft formed a partnership with Y Combinator — known for helping spawn Dropbox, Airbnb, Stripe and other companies — that gave $350,000 in credits to startups entering the accelerator. Startups in a select few other programs, such as the Alchemist Accelerator and Alt Capital’s Generate, are also eligible.

Amazon followed in April, announcing $500,000 in credits to Y Combinator companies, including $200,000 in cloud credits and $300,000 in credits for proofs of concept using the cloud provider’s Trainium and Inferentia chips for AI, an AWS spokesperson said in an email. The current offer includes $350,000 in AWS credits, plus $300,000 reserved for tapping the custom silicon, the spokesperson said.

Annie Pearl, a Microsoft corporate vice president, told CNBC that prior to the Y Combinator partnership, only about 5% of companies in the program were building on Azure. By May, more than 50% were using Azure, she said. A spokesperson later said 58% of Y Combinator startups had taken up Microsoft’s credit offer, a figure that doesn’t reflect actual Azure usage.

AWS said it’s seeing a different dynamic play out.

“That claim just doesn’t ring true to us,” the AWS spokesperson said in an email, referring to Pearl’s statement that over half of Y Combinator startups were using Azure. “In their early stages, startups might accept promotional credits from different cloud providers, but when they mature and need to make a decision on who to trust the future of their organization with, they overwhelmingly turn to the provider with the best security, reliability and scalability.”

Amazon said in an April blog post that over 80% of startups in Y Combinator’s 2022 and 2023 batches ran on AWS.

Microsoft and Amazon are competing for startups well beyond accelerator programs. Last month, AWS doubled to $200,000 the maximum amount of credits a startup can use if it’s raised a Series A funding round in the past year, CNBC reported. In the Microsoft for Startups Founders Hub program, companies can get $150,000 in Azure credits.

In looking at the overall cloud market, industry data shows Microsoft has narrowed Amazon’s lead. AWS’ share in the first quarter of this year was 31%, and Azure was in second place at 25%, according to research firm Canalys. Three years earlier, AWS controlled 32% of the market, while Microsoft was at 19%, Canalys estimated.

Microsoft CEO Satya Nadella said on an earnings call in October that more emerging businesses were turning to Azure because of demand for OpenAI’s models.

“We’re expanding our reach with digital-first companies,” he said. “Leading AI startups use OpenAI to power their AI solutions, therefore, making them Azure customers as well.”

InKeep, whose technology lets companies search internal documents using chatbots, chose to use Azure while participating in Y Combinator in early 2023, shortly after OpenAI launched ChatGPT. OpenAI’s underlying LLMs weren’t available on other clouds.

“Especially when I started, OpenAI did have kind of the state-of-the-art models,” Nick Gomez, InKeep’s co-founder and CEO, said in an interview. InKeep also began using Google’s Cloud Platform for certain workloads.

Gomez said Azure has less downtime than other clouds and acts quickly even when dealing with compute-intensive AI models. He said data privacy is very important to customers when it comes to AI training. OpenAI had initially trained models with customer data but later stopped the practice, CEO Sam Altman told CNBC’s Andrew Ross Sorkin last year.

“People would ask all the time, ‘Are you training on our data?'” Gomez said. “Being able to say, ‘Hey, no, we don’t, we use Azure, they don’t retain it, they don’t train on it,’ stuff like that, definitely helped to put a lot of folks at ease.”

Cloud infrastructure has proven not to be a winner-take-all market. Amazon, Microsoft and Google have all steadily grown their revenue in a business that Canalys expects to expand by 20% this year to almost $350 billion.

That’s in part because large companies are increasingly using multiple clouds to ensure they’re not overly reliant on a single vendor and to take advantage of the differing services and technologies from various providers. For startups that rely on venture funding to fuel their operations, accepting credits from multiple suppliers allows them to keep their expenses in check, which is particularly important given the high costs of running AI workloads.

Accepting credits is “almost like raising money,” said Prady Modukuru, co-founder and CEO of Sync Labs, a developer of lip-synching technology.

“No one can spend $20,000 to $30,000 a month on infrastructure costs,” said Modukuru, a former Microsoft product manager.

Modukuru said Sync Labs has used Amazon, Google and Microsoft, but started with Azure earlier this year while in Y Combinator. It’s the only place where the company could find GPUs, he said.

“We would just request, and within an hour and a half, we would get access to them on Azure,” Modukuru said. “That’s what we needed as a startup.”

Earlier this year, Sync Labs learned how to run high-performance code across many GPUs by talking with Microsoft technicians during office hours, Modukuru said. AWS also makes its experts available to Y Combinator founders, a spokesperson said.

AWS has other ways of taking on Microsoft and its tight partnership with OpenAI. For example, Amazon poured billions of dollars into Anthropic, which is developing its own LLMs. Anthropic has released a model that’s at least as good as OpenAI’s GPT-4, said Daksh Gupta, CEO of Greptile, a startup helping developers work with source code.

Because Anthropic’s model is available on AWS, Greptile plans to end its use of the Azure OpenAI service and switch to AWS’ competing Bedrock tool, Gupta said.

“For the quality of experience, it doesn’t make sense to pinch pennies on it,” he said. “We spend whatever we need to spend.”

Still, OpenAI gives Microsoft a big head start in AI and is forcing AWS into the unfamiliar position of trying to play catch-up. Kareem Nassar, who co-founded EzDubs with Krishnamurthy, said OpenAI’s rapid market penetration has helped Microsoft deal with complex AI infrastructure maintenance issues.

“I know it’s been battle-tested,” Nassar said. “I wasn’t hitting huge bugs. You could just tell it has some mileage on it.”