A few years ago, $100 million was a remuneration reserved for the founder of an enormously successful company. Now it\u2019s a signing bonus for a top AI engineer.<\/p>\n\n\n\n
Mark Zuckerberg has been throwing $300 million pay packages<\/a> at research scientists in the hope of building an AI talent powerhouse<\/a> at Meta. Other Big Tech companies are pillaging startups for their top employees<\/a>, paying off investors and leaving behind the hollow shells of once-promising companies for others to pick over.<\/p>\n\n\n\n One might ask: why this hiring frenzy? The first company to create artificial general intelligence (AGI) \u2014 a system that can match or surpass human cognitive abilities across most tasks \u2014 could become the most valuable enterprise in history. Sam Altman, in the old days of 2019, said<\/a> this would \u201cmaybe capture the light cone of all future value in the universe, and that\u2019s for sure not okay for one group of investors to have.\u201d <\/p>\n\n\n\n With that prize in sight, the stakes are too high for companies not<\/em> to go all in \u2014 they simply have to push their chips to the middle and pay whatever it takes to get the best talent possible in the hope of being the first to cross the finish line. <\/p>\n\n\n\n This war for talent in AI is unlike anything we have seen before in the business world. The closest parallels are all in other industries.<\/p>\n\n\n\n Professional sports teams regularly pay extraordinary sums to attract top talent. The LA Dodgers famously signed Shohei Ohtani for $700 million over 10 years. Lionel Messi\u2019s 2017 contract with FC Barcelona was worth $600 million. Quarterback Patrick Mahomes is getting $450 million over a decade with the Kansas City Chiefs.<\/p>\n\n\n\n Financial firms like Citadel, Millennium, and Jane Street also offer astronomical pay packages to talent, trying to lure prop traders, hedge fund managers, and quantitative researchers to their companies. Though the details of these deals are typically confidential, the figures are similar to the ones we see in sports.<\/p>\n\n\n\n Then there\u2019s entertainment. Movie stars and directors are regularly paid huge sums of money to contribute to projects. These are mostly time and project bound, not quite a continuing employment, but often the contract will include an \u201cexclusivity clause\u201d where the star can\u2019t work with anyone outside the studio for a certain period of time, even if the studio itself isn\u2019t taking advantage of their talents.<\/p>\n\n\n\n The race to AGI is bringing the superstar effect to the world of tech.<\/p><\/blockquote><\/figure>\n\n\n\n In all three of these domains, the firm footing the bill is able to calculate with some accuracy what sort of return on investment it can expect from the talent. In sports, for example, a team might consider how the player will affect its on-field success and then predict how that success could positively affect ticket sales, sponsorships, or merchandising. And if the cost of the talent is less than the value they could bring to the team, an offer is made.<\/p>\n\n\n\n In each industry, there\u2019s an equation, and even if it involves some educated guessing, it\u2019s solvable. These guesses don\u2019t always pan out, of course \u2014 Johnny Depp earned $20 million for \u201cThe Lone Ranger,\u201d which ended up losing $160 million for Disney \u2014 but the model pays off often enough to stick around.<\/p>\n\n\n\n Traditionally, tech hasn\u2019t been like that. It has had superstar teams, superstar companies, and even superstar founders and CEOs, but no actual superstar employees. Despite the widespread appreciation of the 10x engineer<\/a>, nobody was actually being paid 10x. Being a 10x engineer earned you admiration, not generational wealth. Equity plus salary kept you comfortable, even wealthy, but not dynastically rich. <\/p>\n\n\n\n But the race to AGI is now upending the status quo and bringing the superstar effect to the world of tech.<\/p>\n\n\n\n Many parts of the race to AGI are unprecedented. Extraordinary capital investment \u2014 tens of billions of dollars, and soon, hundreds if Softbank<\/a> and Zuckerberg<\/a> are to be believed. Extraordinary growth unlike anything we\u2019ve seen before \u2014 in less than three years, OpenAI went from $0 in revenue to $10 billion and from no users to more than a billion, across pretty much every country and organisation you could name. <\/p>\n\n\n\n If that was it, though, the AI industry would largely be following the same trajectory we\u2019ve seen cloud startups travel over the past couple of decades, just on a shorter timescale. What sets this situation apart is that the success in AI has largely come on the backs of individuals or very small teams of researchers and engineers. <\/p>\n\n\n\n Small teams are creating tremendous value.<\/p>\n<\/blockquote>\n\n\n\n The modern AI revolution started with the eight authors of Google\u2019s landmark \u201cAttention Is All You Need<\/a>\u201d paper \u2014 the \u201cTransformer 8<\/a>\u201d are now celebrities. One of them, Noam Shazeer, and researchers extraordinaire Ilya Sutskever and Alec Radford were then individually responsible for many of the breakthroughs that have led to transformer models becoming so smart and useful. OpenAI\u2019s GPT-4, the model that shook the world, was built by around 200 people \u2014 that\u2019s a group 1\/1000th the size of Google.<\/p>\n\n\n\n What\u2019s remarkable isn\u2019t just the size of these teams \u2014 it\u2019s the value they\u2019re creating. The transformer block that the \u201cAttention\u2026\u201d authors proposed in 2017 turned out to be incredibly versatile. Though originally proposed for language processing, it\u2019s now the core of systems for protein folding, weather forecasting, chip layout, and even plasma control in fusion reactors.\u00a0<\/p>\n\n\n\n With so many labs now trying to build the exact same products around their models \u2014 chatbot, image creator, coding interface, deep research \u2014 the only thing that will set them apart is how well the model <\/em>itself works. And when the model is the product \u2014 and there\u2019s proof you don\u2019t need a lot of people to build the next best model \u2014 companies are willing to pay a lot <\/em>to hire people they think will put them ahead in the race.<\/p>\n\n\n\n The emerging strategies that companies are employing in this new tech talent war already show some trends, including a few that take their cues from those other industries we\u2019d looked at.<\/p>\n\n\n\n This version of the tech industry bears little resemblance to the one that existed just a decade ago, when small companies would try to do hard things, not just because success would mean bigger paychecks, but for the love of the game. People in finance often pride themselves on working for the money, but in tech, it was also about the joy of coding, being creative, and seeing what you might be able to accomplish in a night or weekend. <\/p>\n\n\n\n That\u2019s not to say an engineer can\u2019t still find fulfillment while making nine-figures, but it\u2019s hard to imagine there\u2019s not a difference between shooting hoops with your friends and playing for a national title \u2014 the stakes reshape the game. Each win also shakes up your future with the league \u2014 better AI models means better AI coders<\/a>.<\/p>\n\n\n\n As the race intensifies, Silicon Valley is likely to continue adopting the playbooks of industries that have long treated talent as the most critical \u2014 and expensive \u2014 input. The difference is that this time, the prize isn\u2019t a trophy, a blockbuster, or an alpha-generating strategy. It\u2019s the future of intelligence.<\/p>\n\n\n\nThe top talent playbook<\/h3>\n\n\n\n
The model is the product<\/h3>\n\n\n\n
\n
AGI or bust<\/h3>\n\n\n\n
\n