Why the AI era signals the end of the garage founder mythHow the AWS Generative AI Accelerator is powering the startups of the future.
Oct 7, 2025 at 9:00 AM EDT
Silicon Valley has long loved the origin story of the garage inventor. Many companies that have since become household names were built by a lone entrepreneur, or a small band of dreamers, from inside a dingy room. These stories have crafted something of a creation myth — the idea that building the world’s next innovative technology just requires a few talented people and some grit.
The spirit of that sentiment remains true, particularly because AI makes it easier for anyone to create something new. These days, you don’t need to know how to write code in order to create your own app. But for the people who are creating the technologies that enable the AI era, the myth no longer holds.
For the sophisticated researchers who are getting this new tech off the ground, the work is resource intensive: Training large models requires petabytes of data and fleets of GPUs — not just ingenuity and late nights. Even startups tackling narrower AI challenges are finding that modern company-building isn’t about isolation or scrappiness in a vacuum, it’s about having access to infrastructure, expertise, and support networks. This is the ethos behind Amazon Web Services’ (AWS) Generative AI Accelerator, a key component of the company’s $230 million commitment to help accelerate generative AI startups around the world. Last year, 80 startups building both generative AI technologies and applications were selected and the new 2025 cohort was just announced, giving a new crop of 40 companies producing AI models and tooling access to up to $1 million in AWS credits, an eight-week mentorship program, and networking opportunities with AI industry leaders.
When the garage is too small
Bedrock Robotics, a company developing autonomy for heavy construction equipment, was part of AWS’ 2024 accelerator and credits its success — the company secured an $80 million funding round in July and has plans to launch in 2026 — to the resources that program made available to its team. “Bedrock Robotics couldn’t have been started successfully without support,” says Tom Eliaz, its co-founder and VP of engineering. “Neither the team, the equipment, nor the methodology could fit in a garage.”
Bedrock’s founding team — largely veterans of Waymo — brought years of valuable engineering experience, and also the knowledge that they couldn’t do it alone. The team leaned on investors, construction industry advisors, real-world partners who they could work with to test excavators, and technology providers like AWS. “For almost every aspect of the company, we sought advice and help from others,” Eliaz says. “We’re still doing that today at larger scale.”
That doesn’t mean the myth is entirely irrelevant. “The lone engineer in a garage can still change the world,” Eliaz adds. “On the other hand, at Bedrock Robotics, we’ve found that when we’re trying to create AI for the physical world, we benefit from infrastructure and advice that matches the scale of our ambition.”
If the garage was once the default birthplace of startups, the modern equivalent is increasingly the accelerator hub. Programs like the AWS Generative AI Accelerator have positioned themselves as the scaffolding for this new generation of companies.
Bedrock Robotics modifies heavy machinery with AI software, cameras, lidar, and computers that make it possible for the machines to run 24/7, even if it’s the middle of the night, or over 100 degrees — conditions that would prove difficult for solely human-operated machines. One of the main advantages of a startup is having the room to try something, make mistakes, pivot quickly, and continue to iterate. But for Bedrock Robotics, Eliaz says, “those mistakes would have been made on petabytes of data processed and days or weeks of GPU time; the financial stakes could have stopped fast iteration.”
The team needed to train and test models with massive amounts of data, in countless scenarios and across different tasks and equipment types. They knew that in the beginning they were going to get it wrong a lot before they got it right. Before the AWS accelerator program, they expected they’d need to limit spending and reign in engineers to stay on budget.
“One of the biggest immediate impacts of the AWS accelerator was a large amount of AWS credits, bringing with it the freedom to focus all our infrastructure and ML [machine learning] teams’ efforts on exploring and solving the autonomy problem, without limiting ourselves to data processed or training hours spent,” Eliaz says. The Bedrock team spent credits on compute-intensive workloads, autonomy model training, and dataset processing. This freedom allowed them to hit major milestones within six months, and set more ambitious goals during their first year building the company.
Beyond the financial support, Bedrock’s founders were able to spend a week in Seattle working alongside other companies and connect with subject matter experts, some of whom helped validate their initial infrastructure architecture. “Feeling the energy behind physical AI and the future of robotics at Amazon and among the AWS partners, along with the energy from our cohort members, was a big positive push for our infrastructure team,” he says.
For AWS, this shift from isolation to collaboration reflects a broader trend: the need to work quickly. “Whether it’s in biotech labs, creative studios, or industrial applications, the pace of generative AI innovation is extraordinary, and it’s happening everywhere,” says Sherry Karamdashti, general manager & head of startups in North America at AWS. “This year’s cohort reinforces our mission to help that innovation move faster and deliver real-world impact for customers in every industry. We’re removing the barriers and accelerating opportunities so these leaders can grow their world-changing solutions.”
While the spirit of the garage inventor narrative still exists — that notion that breakthrough innovation can come from anyone, anywhere — when it comes to creating AI technologies, the tools and costs required to bring an idea to life have skyrocketed. To get off the ground in fields like generative AI, the path forward looks less like a garage and more like a robust ecosystem: distributed, interdependent, and powered by infrastructure far larger than any one single founder. |
