An offer he couldn’t refuse: The young Aussie who couldn’t say no to Zuckerberg

They come from Winton, rural Queensland, a town of 800 people two hours away from the nearest IGA. They come from Perth’s most exclusive private schools, graduating with perfect tertiary entrance ranks and university medals. They come from Sydney, having quit jobs to self-study AI after seeing a friend make the leap to Silicon Valley.

Meet the new Australian expatriates at the centre of the greatest technological gold rush in human history.

Australian AI researchers and engineers are making billions in Silicon Valley.Credit: Michael Howard

“It feels like Renaissance Florence,” says Sholto Douglas, a researcher at Anthropic, describing the current mood in California. But unlike Florence, the patrons here aren’t buying art; they are buying the cognitive architecture of the future, and they are paying in equity packages that rival the GDP of small nations.

In the elite labs of San Francisco, the concentration of Australian talent is now so high it’s a running joke at parties. “People complain to me ... They’re like, ‘Oh, another Australian ... they’re practically sick of us’,” says Casey Flint, a Queenslander who is now chief of staff at the frontier lab Reflection AI. This Australian cohort is building the future of intelligence, and earning fortunes that defy belief.

The competition for AI researchers has reached extraordinary levels, bordering on the obscene. Meta chief executive Mark Zuckerberg has been personally recruiting via email, WhatsApp and invitations to his homes, with compensation packages for top-tier researchers reportedly reaching up to $US300 million over four years.​

OpenAI has countered with reported retention bonuses exceeding $US2 million and equity packages above $US20 million. Star researchers are effectively being treated like NBA superstars.

The mood amid the AI boom in San Francisco and its surrounds has been likened to “Renaissance Florence”.

But beyond the eye-watering numbers are Antipodean researchers and engineers who Zuckerberg, Elon Musk, Sam Altman and other tech billionaires are personally courting with offers that can stretch into ten figures.

These are the people helping build ChatGPT, or the next one.

Their stories reveal both the extraordinary opportunity and the troubling brain drain facing Australia’s tech sector.

Andrew Tulloch: The billion-dollar brain

Andrew Tulloch has left the AI start-up he co-founded to join Meta.

If there is a face to the sheer financial madness of the 2025 AI talent wars, it is Andrew Tulloch. The man behind the $US1.5 billion headline is not a “move fast and break things” coder. Tulloch’s background is one of pure, elite Australian academic achievement. Described by industry sources as shy and keen to retain a low profile, his resume speaks for itself.

He graduated from one of Perth’s most exclusive private schools, Christ Church Grammar, with the highest possible tertiary entrance rank (now ATAR) of 99.95 in 2007. He was school vice-captain, debating team captain, and a member of the first XI cricket and hockey teams. He studied at the University of Sydney, graduated in 2011 with first-class honours and a university medal in advanced mathematics. From there, he earned a master’s degree in statistics and machine learning at Cambridge University.

Tulloch’s staggering strategic value to Meta, however, comes from his first, 11-year stint at the company, which began after a brief period at Goldman Sachs in Sydney. At Meta (then Facebook), Tulloch was a pivotal engineer in the creation of PyTorch, the open-source AI framework Meta founded in 2016.

PyTorch is not just a project at Meta, it underpins its entire AI strategy. It is the cornerstone upon which its research and production systems are built, and a key pillar of Zuckerberg’s entire open-source AI philosophy.

In 2023, Tulloch left Meta. He did a stint at OpenAI before co-founding the start-up Thinking Machines Lab (TML) in February, alongside former OpenAI chief technology officer Mira Murati. TML immediately became a formidable competitor, raising a staggering $US2 billion seed round at a $US12 billion valuation from investors including Andreessen Horowitz.

Zuckerberg’s pursuit of Tulloch was not merely an offensive move to gain a star researcher. It was a critical defensive manoeuvre to reclaim an irreplaceable asset and neutralise a profound competitive threat.

Meta had reportedly tried to acquire TML outright to avoid losing ground in the AI race. When that failed, Zuckerberg personally targeted TML’s key staff.

Tulloch possessed deep, institutional knowledge of Meta’s core AI engine and his role as co-founder of TML meant he was now using that expertise to build a next-generation competing system, backed by $US2 billion in investment. The $US1.5 billion offer (which, at the time, a Meta spokesperson said was “inaccurate and ridiculous”) was therefore not a “salary” in the traditional sense. It was a second attempt at an acquisition – this time, targeting the single human asset who built their engine, effectively neutralising the threat of him building a better one for a rival.

While Tulloch initially declined the offer, the relentless campaign eventually succeeded. In late 2025, it was confirmed that Tulloch was leaving TML to rejoin Meta for an undisclosed, though undoubtedly massive, sum. Zuckerberg had, in effect, bought his engine-builder back.

Sholto Douglas: The visionary at Anthropic

Sholto Douglas never imagined working in AI when he was competing as one of the top 50 fencers in the world during his university years in Sydney.

Despite applying to PhD programs in 2020 and being rejected, Douglas’ independent research caught the attention of researchers at Google’s AI research lab DeepMind, who recruited him on what he describes as “the most junior role that you could possibly be in” – an L3 engineer with no track record. He arrived just a month before ChatGPT’s release, when Google threw itself into panic mode.​

Anthropic researcher Sholto Douglas.

“They took 1000 people and were like, ‘Please figure out how to make a competitor to ChatGPT, your time starts now’,” Douglas says. “There was so much chaos and the field was so fresh that actually there was a lot of room for people to figure it out as you go along.” ​

Douglas is now a tech lead on Reinforcement Learning infrastructure at Anthropic, the AI lab founded by former OpenAI researchers that is now one of the primary contenders for the throne. He’s working on “scaling compute” – effectively figuring out how to feed the AI more power and data to make it smarter. He describes the work as intense but deeply mission-driven. “Everyone here is trying to bend the arc of history,” he says.

Douglas is coy about his own salary but is adamant that the reported numbers in the industry are real, as is the value that these researchers and engineers are delivering to their companies.

“Companies are spending tens of billions of dollars on compute,” he says. “If you figure out a way to make the model run 10 per cent more efficiently, you’ve actually just saved billions of dollars ... The right people can make huge improvements.”

Douglas admits the poaching is relentless. “It’s pretty CEO-driven,” he says. Tech titans like Altman, Sergey Brin and Zuckerberg personally text and call researchers to recruit them.

“You get a text from Sam [Altman] saying, ‘Hey, would you like to chat?’” he says. “You’re super nervous.”

Tristan Heywood: Making ChatGPT faster

Douglas helped pull friend Tristan Heywood into the vortex. Heywood, an engineer who grew up in Sydney and was awarded the University Medal at the University of Sydney, watched from afar as Douglas landed a job at DeepMind.

OpenAI researcher Tristan Heywood.

“It really proved it was possible,” Heywood says. He quit his job, spent three months intensely studying transformers and kernel programming, and landed a role at OpenAI in January 2023, just as the ChatGPT craze was igniting.

Heywood’s job is tangible: he makes ChatGPT faster. As part of the inference team, he optimises the massive mathematical calculations required to generate text, ensuring the model can serve millions of users without melting down the GPU clusters.

When asked about his salary, Heywood is straightforward: “I’ll tell you this, it is seven figures.”

“It’s a wild stat,” Douglas says of his friend’s ascent. “He 10 times-ed his salary in a month.”

But the money comes with a heavy psychological load. Heywood acknowledges the weight of working on technology that could reshape – or end – civilisation.

“It is possible that, like, a real AGI [artificial general intelligence] could cause extinction of humanity,” he says. “That is kind of daunting that we have that responsibility.” He worries about bad actors using the tech to develop bioweapons, a fear echoed by many in the safety-conscious corridors of OpenAI and Anthropic.

Despite the existential dread, the thrill of the race is undeniable. Heywood’s is a role that sits at the bleeding edge of human engineering.

Ben Goodger: The architect of the internet

If there is a singular architect of how humans experience the modern web, it is Ben Goodger. A New Zealander raised in Auckland, Goodger didn’t just watch the browser wars of the past two decades; he won them.

In the late 1990s, while studying computer engineering at the University of Auckland, Goodger began sending code patches to Netscape, the dominant browser of the era. “I got this ping on chat from one of the top engineers at Netscape saying, ‘Do you want a job?’” he recalls. He accepted and moved to Mountain View for what was supposed to be a one-year internship.

OpenAI executive Ben Goodger.

He never really left the frontline. Goodger went on to ship the first version of Firefox in 2004, breaking Microsoft’s stranglehold on the web. He moved to Google, where he was instrumental in creating Chrome, the browser that now runs the world. Eighteen months ago, he joined OpenAI to lead engineering on Atlas, a project that aims to do to the browser what ChatGPT did to search.

“It puts ChatGPT at the heart of the browser,” Goodger says of the new tool, which allows users to let the AI “drive” their web experience rather than just retrieving information. He describes the shift as moving from a world of search queries to one where “habits take over”, allowing the AI to click around, fill out forms, and execute tasks on the user’s behalf.

Goodger compares the atmosphere at OpenAI to his early days at Google – “it’s not the same, but it rhymes” – describing a culture of enthusiasm to build and release products tempered by the heavy responsibility of serving 800 million weekly users.

Despite his legendary status in Silicon Valley, Goodger says he retains the thick skin of an Antipodean expat. He dismisses the fervour around 10-figure compensation packages. “I don’t spend too much time thinking about crazy compensation packages,” he says. “I’m spending time with my team.” He views the exodus of talent from Australia and New Zealand not as a crisis, but as a cultural trait.

OpenAI’s new AI browser Atlas.

“We’re explorers,” he says of the Kiwi and Aussie diaspora. “We want to go out and see what the world’s like ... it’s just kind of a fact of life.” While he notes the billboards in San Francisco are now exclusively advertising AI products, Goodger believes the tech scenes in Australia and New Zealand are richer than when he left 22 years ago. “The world doesn’t really care where you come from,” he says. “The world wants to see what your idea is.”

Casey Flint: The AI strategist

If Tulloch, Douglas, Goodger and Heywood are building the brains, Casey Flint is helping build the body. Raised in Winton, a rural Queensland town with a population of 800, Flint’s journey to the C-suite of a frontier AI lab is perhaps the most unlikely of all.

“We had a fly-in, fly-out doctor,” she says of her childhood. “We were two hours away from the nearest IGA.” Her story illustrates both the pull of Silicon Valley and what Australia risks losing.​

Growing up without permanent doctors, dentists or psychologists, Flint became fascinated with how technology could transform healthcare. That interest led to a biochemistry degree at the University of Queensland and eventually a six-year career at Uber, where she worked her way from intern to strategy roles spanning Brisbane, Sydney, Amsterdam, Japan and South Korea.​

Casey Flint was at Uber before joining Square Peg as an investment associate.Credit: AFR

A pivot to venture capital at Square Peg gave Flint a front-row seat to Australia’s AI ecosystem, and its limitations. She realised she wanted to be “in the room when it happens”.

“I’d go to SF [San Francisco], I’d have these conversations with researchers and founders here, and then I’d come back and I’d get pitched a concept that was like six to 12 months old by SF standards,” she says. “I found that a bit fatiguing, and really wanted to be at the very frontier.” ​

After a stint at Amazon Web Services focusing on AI investments, Flint joined ReflectionAI as chief of staff, working with former DeepMind and OpenAI researchers building what they describe as “superintelligence”. The team recently launched Asimov, a code research agent for engineering teams.​

Flint worries about AI’s economic implications for Australians like her family back in Winton.

‘I come from a rural part of Australia … If you have all of this economic value and knowledge work basically being done overseas ... how are Australian businesses capturing value from that?’

Casey Flint

“I come from a rural part of Australia,” she says. “I don’t think there are nearly enough people thinking about the economic implications. If you have all of this economic value and knowledge work basically being done overseas ... how are Australian businesses capturing value from that?”

She believes a new “FAANG” – comprising tech giants Meta (formerly Facebook), Amazon, Apple, Netflix and Alphabet (formerly Google) – is emerging and is determined that “Reflection” will one day form part of the acronym.

James Groeneveld: The rower who rebuilt the engine

James Groeneveld was never meant to be fixing billion-dollar server stacks in Silicon Valley; he was supposed to be in a hospital ward back home. The Brisbane native headed to Harvard on a rowing scholarship with a clear, parental-approved plan: “I’d come back one day [as a] doctor, my mum would be happy, and everything would work out.”

That plan evaporated in his freshman year, derailed by a computer science course that left him enamoured with the power of code. Today, instead of hospital rounds, Groeneveld is dissecting the digital brains of Character.AI, one of the world’s most heavily used consumer AI platforms.

Character.AI chief intelligence officer James Groeneveld.

Having cut his teeth at Google Brain managing the massive fleets of chips used to train early AI models, Groeneveld joined Character during a period of chaotic hyper-growth. His mandate sounds terrifying: rebuild the entire technological foundation of the company while millions of users were actively using it.

“We were building the planes while flying,” Groeneveld says of the six-month sprint to replace the prototype code written by brilliant – but commercially inexperienced – researchers. “They’re brilliant mathematician types ... But not the best systems programmers building for scale.”

Groeneveld argues that the astronomical salaries defining his industry aren’t just hype; they are a reflection of a sector where the rule book hasn’t been written yet. He describes the industry as being at the “very beginning of the onion”, lacking the established frameworks that exist in traditional web development.

“There are no best practices, really,” he says. Consequently, companies are forced to pay for the few engineers who understand the raw fundamentals of how these models interact with hardware. When a single training run costs $1 billion, the engineer who ensures it works is worth their weight in gold. “Tying the compensation to the outcome makes sense,” Groeneveld says.

“If one person can have that large of an economic outcome ... 2 per cent of that is nothing in the grand scheme of things.”

Despite his success in the US, Groeneveld remains a believer in the grit and determination of Australian talent. He points to rising stars such as the “Unsloth” brothers (Daniel and Michael Han), who began optimising AI models from a Sydney bedroom, as proof that Aussies are punching above their weight. For now, however, the centre of gravity remains in San Francisco, which is a long way from the rowing sheds of Brisbane.

Can Australia and New Zealand compete? Should they try?

The list of Australian AI leaders continues to grow. There is Joel Pobar, a veteran engineering leader who recently returned to Meta’s superintelligence division after a stint at Anthropic. Clare Birch, a former quantum chemist and Blackbird investor, is now helping build the ecosystem at Mira Murati’s Thinking Machines as chief of staff. And Ben Brooks, a fellow at Harvard’s Berkman Klein Centre, is shaping the global regulations that will govern these technologies.

Joel Pobar is an Australian engineering leader who recently returned to Meta’s superintelligence division after a stint at Anthropic.

The consensus among the expatriates is clear: for now, the cutting edge is in San Francisco. You just have to be there. “It would be hard in Australia,” Heywood says. “OpenAI is ... very consequential. Everything they do is a lot more impactful.”

However, they believe Australia is not out of the game, provided it stops trying to build its own ChatGPT and starts playing to its strengths. “Australia is one of the best places in the world to build compute,” Douglas says, citing the nation’s solar capacity, stable rule of law and massive superannuation capital. “The world is going to be accelerator constrained for the foreseeable future.”

Flint echoes this “lower stack” strategy. She believes Australia should become an energy and mineral superpower for the AI age, supplying the copper for the chips and the green energy for the data centres. “We need a lot of energy to power data centres ... Australia could become an energy superpower.”

Heywood also has an idea for Australia, one that wouldn’t necessarily stem the brain drain but might help maximise the benefits from AI.

“Australia should ask American tech companies to release open-source models,” he says. “If the [Australian] government is negotiating a big contract, they can push for that as part of negotiations. This would let Australian companies build on top of strong AI models, and also reduces the centralisation of the technology in the US.”

For now, though, the brain drain continues. San Francisco’s pull – with its “Renaissance” vibe and its eye-watering compensation packages – is simply too strong.

“I keep trying to tell other friends,” Douglas says, urging more Australians to make the leap. “Whatever you’re doing, come and do it here.”

For these expats, the gamble has paid off. They are the architects of a new intelligence, working on the most important technology of the century, and being paid fortunes. Australia may have lost them, but it certainly hasn’t forgotten them – even if it can no longer afford them.

It can still hold hope they one day make their way back.

David Swan travelled to San Francisco as a guest of Salesforce.

Get news and reviews on technology, gadgets and gaming in our Technology newsletter. Sign up to receive it every Friday.