At Moorfields Eye Hospital in London, a digital diagnostic system drawing on a research archive of close to a million retinal scans catches early signs of sight-threatening disease as accurately as human specialists. At the Royal Free Hospital, a companion tool flags acute kidney injury in patients far earlier than before, giving understaffed National Health Service (NHS) nurses a critical window to act. Neither breakthrough came from a Silicon Valley tech giant. Both emerged from DeepMind, the London-based AI research lab co-founded in 2010 by Mustafa Suleyman to build software for the public good.
Many of the most exciting developments in AI, the ones that make you think that technology could be a tool for progress rather than a potentially world-ending threat, are being developed by outsiders. And their outsider knowledge is shaping which problems the technology is aimed at, whose data it learns from, and who it serves first.
Consider Suleyman’s own path. It began in 2001, when he dropped out of Oxford University at nineteen to launch the Muslim Youth Helpline, a free, confidential telephone counseling network for young British Muslims whose mental health struggles and isolation were largely ignored by existing services. Growing up in North London as the son of a Syrian taxi driver and an English NHS nurse, Suleyman had seen those institutional blind spots firsthand.
That background is precisely what later connected him to medicine. Following Google’s acquisition of DeepMind in 2014, Suleyman became Head of Applied AI. Seeking a technical partner who understood public-sector strain, Moorfields clinician Pearse Keane tracked Suleyman down on LinkedIn. “I thought, Mustafa is from North London, his mother was an NHS nurse… he’s going to know Moorfields,” Keane recalled of their 2016 partnership. Rather than point its talent at advertising or trading, DeepMind’s health unit built diagnostic tools for a strained public hospital system—the very institution Suleyman had grown up inside. “We set up DeepMind because we wanted to use AI to help solve some of society’s biggest challenges,” Suleyman noted at the time.
As a scientist, you have to be resilient because science is a non-linear journey, and as an immigrant, you learn to be resilient.
When Microsoft appointed Suleyman as CEO of its consumer AI division in 2024, it placed this precise outsider philosophy at the highest level of global technology. His trajectory illustrates a core reality of modern innovation: many pioneers of the digital AI revolution are immigrants, or their children, and spent years navigating systems that were never designed for them. They watched those frameworks fail from the outside—and they wanted to do better.
An artist’s illustration of artificial intelligence (AI)
Suleyman’s work addresses who a system is built for. Yet an earlier question remains: what is that system trained on? This foundation determines what a machine can actually perceive before it ever interacts with the public. If you have ever encountered a technology that felt entirely blind to your presence, you have experienced a systemic data gap. Fei-Fei Li spent her career closing it.
Li arrived in the United States from China in 1992 at the age of fifteen. Throughout her adolescence, she split her life between two vastly different realities: working behind the counter of her parents’ suburban dry-cleaning shop in Parsippany, New Jersey, and studying physics at Princeton University. Navigating a new culture while balancing a working-class survival job directly shaped her intellectual approach. She later noted that her unique perspective on how machines learn was deeply “accentuated by [her] own journey” as an immigrant who had to learn to see and translate the world through a second language.
The immigration story, the dry-cleaning business, my parents’ health, the journey I’ve been through is so deeply human. That gives me a lens and perspective that is uniquely different.
When she joined the faculty at Stanford, the field was obsessed with building ever more complex algorithms. But Li, attuned to what dominant frameworks leave out, noticed that machines were learning only from narrow image libraries that reflected the insular worlds of the researchers themselves. “We had to change our engineering mindset,” she explained; a computer cannot understand what it has never been shown. Her solution, launched in 2009, was ImageNet—a library of roughly 14 million diverse, crowdsourced images across 22,000 categories, built to teach computers the actual breadth of the world.
What happened when others ignored that lesson made her point for her. In 2018, MIT researcher Joy Buolamwini documented that facial-recognition software from major tech firms had error rates as high as 35 percent for darker-skinned women—because the training data had left them out. To combat that bias, Li co-founded AI4ALL, whose Ignite program now provides free computer science education to BIPOC undergraduates across 180 colleges, closing the very gaps ImageNet had first exposed.
Vivid, blurred close-up of colorful code on a screen, representing web development and programming
The questions Suleyman and Li addressed—who a system is built for, and what it is trained to recognize—both assume the system gets built at all. A third condition underlies both: access to the computing infrastructure required to build AI. Ali Ghodsi spent his career working against its concentration.
Ghodsi was born in Iran in 1978; his family fled to Sweden when he was five. He taught himself to code, earned a doctorate in computing, and arrived at UC Berkeley in 2009 to work on machine learning—systems that learn from data rather than fixed instructions. At the time, running them demanded specialized hardware costing millions of dollars, available only to a handful of companies. To someone whose family had once been shut out and forced to start over, that concentration of power was not abstract. “We wanted to democratize that and bring it to every company on the planet,” he recalled. His team open-sourced the core software for free. That software—Apache Spark—is now the data-processing foundation most large AI systems run on, and Databricks, the company Ghodsi co-founded around it, serves more than 15,000 organizations, including more than half of the Fortune 500.
Silhouette of a man facing a dramatic city skyline under stormy clouds
These three stories are not exceptions; the same pattern shows up in the aggregate. Among the top 100 privately held US AI companies, 62 percent have at least one immigrant founder—a figure from a July 2025 analysis by the immigration platform Dreem, and one broadly echoed by independent researchers at the Institute for Progress and the National Foundation for American Policy. They include OpenAI (ChatGPT), Anthropic, Databricks, xAI, and Waymo (self-driving cars). Together, immigrant-founded firms raised $167 billion, against $68.1 billion for firms with only US-born founders—a gap of roughly 2.5 to one. These are the foundational models, data tools, and safety systems the rest of the industry builds on.
And most of these founders reached that point by clearing hurdles other entrepreneurs never face. The H-1B visa for skilled foreign workers ties its holder to a specific employer and generally can’t sponsor a founder’s own startup—so even after starting a company, many stayed dependent on an employer simply to remain in the country. Venture capital—the private money behind most tech startups—flows largely through referrals among graduates of the same universities; a 2023 study in Research Policy found that three in four immigrant founders who secured funding had entered the US through its universities, a structural disadvantage for those who came to work rather than study. Georgetown’s Center for Security and Emerging Technology similarly found 72 percent of immigrant AI founders first arrived on student visas.
And yet they clear those barriers at striking rates. The Kauffman Foundation finds immigrant founders create 150 jobs per company on average and are twice as likely to start companies as their American-born peers; the National Foundation for American Policy concluded that without them, the US would have fewer than half as many billion-dollar startups today.
The next wave of AI is already being built, and not only in Silicon Valley. At the India AI Impact Summit in February 2026, companies including Sarvam AI and BharatGPT demonstrated models built for Indian languages and local contexts. An Africa AI Village at the same summit showcased twenty applied AI innovations, from crop disease detection to medical imaging tools built for African conditions. Microsoft’s AI Economy Institute found that AI adoption in the Global South grew at half the rate of the Global North in 2025, largely depending on whether systems work in local languages. And only 5 percent of Africa’s AI talent currently has access to the computing power complex AI work requires—the gap that NVIDIA Inception, the Google for Startups Cloud AI Accelerator, and Amazon’s Impact Accelerator are built to address.
None of this is settled history; the same gaps persist right now. Stanford HAI researchers have found that most major AI language models underperform for non-English speakers, and fewer than 5 percent of the world’s roughly 7,000 languages have enough digital data to train on. The World Health Organization projects a shortage of ten million health workers by 2030. The systems that will address those gaps will be built by people who already understand which conditions go undetected, which languages go unserved, which populations have been absent from the data. Immigrants have been accumulating that knowledge for most of their lives.
By clicking “submit,” you agree to receive emails from TrooRa and accept our web terms of use and privacy and cookie policy. *Terms apply.
©2018 -2026 – TrooRa is a registered trademark of Rare Luxury Living LLC TrooRa Magazine, A Fortunest Group and is registered with the U.S. Patent and Trademark Office.
By clicking “submit,” you agree to receive emails from TrooRa and accept our web terms of use and privacy and cookie policy. *Terms apply.
©2018 -2025 – TrooRa is a registered trademark of Rare Luxury Living LLC TrooRa Magazine, A Fortunest Group and is registered with the U.S. Patent and Trademark Office.
Explore diverse creators, authentic stories, and bold ideas. Login to access TrooRa’s vibrant world of inspiration.
Not a member yet? Register Now
Tier-3 User Trying to View Tier 4 Content
Tier-2 User Trying to View Tier 3 + 4 Content
Tier-1 “Free Member Limit Reached”
Free Member Trying to View Paid (Tier 2+) Content
Guest Trying to View Premium (Tier 1+) Content
Tier-0 “Free View Limit Reached”
