The Talent-to-Valuation Ratio: Why Alphabet Lost $269B Over Four Researchers

Google lost $269 billion in market value in one week.

The talent to valuation ratio

In one week in June 2026, Alphabet lost about $269 billion in market value, more than the total worth of most companies on earth. The cause was almost embarrassingly human. A few of Google’s most respected researchers announced they were leaving for rival labs, and investors reacted as if something fundamental had broken inside the company.

I keep coming back to that number because of what sits right beside it on Google’s books.

Alphabet plans to spend somewhere close to $190 billion on AI infrastructure in 2026.

So in the same stretch of weeks, the market watched Google commit nearly $190 billion to chips, data centres, and compute, then knocked roughly $269 billion off the company because four or five researchers walked out the door. If you want one sentence to explain how the AI industry is priced today, that is it. The infrastructure is the cost. The people are the asset.

What actually happened that week

On 18 June, Noam Shazeer announced he was leaving Google for OpenAI. Shazeer is about as far from a replaceable engineer as you can get. He helped write the 2017 paper “Attention Is All You Need,” the work that introduced the transformer, the design behind almost every large language model in use today.

The “T” in ChatGPT is his. He was one of the leads on Gemini, Google’s flagship model family. And here is the part that stings: Google had paid about $2.7 billion in late 2024 to bring him back through the purchase of his startup Character.AI. Less than two years later, he left again.

A day later, John Jumper said he was leaving for Anthropic after nearly nine years at DeepMind. Jumper led the work on AlphaFold, the system that predicted the structure of more than 200 million proteins and won him a share of the 2024 Nobel Prize in Chemistry, alongside DeepMind’s own chief executive Demis Hassabis. Think about what that means. A Nobel laureate, arguably the person behind DeepMind’s single greatest scientific result, chose to do his next chapter at a competitor.

Those two exits set off the selloff.

Alphabet shares fell as much as 7% on 22 June 2026, one of the stock’s worst days in over a year, and by one measure the largest single day loss of market value in the company’s history.

Then the week got worse. Bloomberg reported that two more researchers seen internally as central to Gemini, Jonas Adler and Alexander Pritzel, were also moving to Anthropic, and a researcher named Arthur Conmy said publicly he was joining Anthropic to work on AI safety. So when a headline says “four researchers,” the honest answer is that, depending on who you count, it was four or five in a single week, and four of them went to the same place.

Why a few names moved a quarter of a trillion dollars

The obvious objection is that Google employs nearly 195,000 people. A research group that deep does not empty out because a handful of stars leave. Hassabis made exactly this point, and he is not wrong. There is constant movement between the big labs, and Google still has the broadest research bench of any of them.

I think the market was reacting to something more uncomfortable than the loss of any single person. It was reacting to direction. Talent in this field does not move at random. A 2025 analysis by the venture firm SignalFire found that DeepMind engineers were roughly eleven times more likely to leave for Anthropic than the other way around. That is a current. It flows away from the company with the most compute and toward the smaller labs.

Two forces make that current strong right now. The first is money, though not in the form of salary. Anthropic filed confidentially to go public on 1 June after raising about $65 billion at a valuation near $965 billion, which put it ahead of OpenAI for the first time. Both companies have said they plan to list.

For a senior researcher, owning equity in a private lab just before it goes public is a kind of upside that Google’s stock grants, however large, simply cannot match. The second force is friction. Bloomberg reported that shortly before Shazeer left, computing power assigned to one of his projects was handed to another team in London. Inside a company of 195,000 people, even the most important researcher waits in a queue and fights internal politics for resources. At a focused lab, they do not.

Timing made it all land harder. The day before the drop, Microsoft’s chief executive Satya Nadella gave an interview arguing that AI models are becoming commoditised and that the industry should lean less on a few giant providers. Place that next to Google’s $190 billion infrastructure plan and you get the precise fear that rattled investors.

If models and compute are heading toward cheap and interchangeable, then spending close to $200 billion to own the whole stack buys you a very expensive ticket to the same race everyone else is running.

The talent to valuation ratio

Here is where I will put my own view plainly. For roughly a decade, the working assumption in AI was that the moat was infrastructure. Compute, data, distribution, and capital were the hard things to get, and researchers were treated as skilled labour sitting on top of that moat. June turned the assumption upside down.

Look at what talent now costs, because the prices carry the argument. In 2025 Meta built its superintelligence lab by offering packages reported at up to $300 million over four years, with some payouts in a single year climbing above $100 million.

Mark Zuckerberg reportedly offered one researcher, Andrew Tulloch, as much as $1.5 billion over six years. Tulloch said no. Microsoft’s head of research, Peter Lee, put it bluntly to Bloomberg: a top deep learning expert now costs about what a top NFL quarterback prospect costs.

Google paid more than $500 million for DeepMind back in 2014 mostly to get its founders. It paid $2.4 billion in 2025 to fold the Windsurf team into DeepMind. Those numbers look like what you would pay for infrastructure, and companies are paying them for individual people.

Now hold that beside the $269 billion. Markets today treat Google’s enormous infrastructure spending as a cost to be justified, sometimes even as a worry, while treating the loss of a few researchers as a real threat. The reaction to who leaves is sharper than any reward for what gets built. That asymmetry is the entire argument. When the market responds more violently to a resignation than to a budget, it is telling you which of the two it believes is scarce.

The commoditisation point that frightened everyone actually strengthens the case rather than weakening it. If compute gets cheap and models start to look alike, then infrastructure stops being a differentiator at all. What is left to explain why one lab pulls ahead of another is the small group of people who decide what to build with all that identical hardware. The cheaper the hardware gets, the more the people are worth.

The case against my own argument

I do not want to oversell this, because the opposite view is real and I hold a piece of it too.

Google did not lose its actual business in June.

Search still prints money, YouTube is still dominant, Cloud is still growing, and the company designs its own chips.

Nothing about that changed because a few people updated their profiles. The $269 billion was a repricing of risk and sentiment, not a judgement on whether Google can build. Breakthroughs in this field tend to be collective, and we have a habit of crediting a few famous names for the work of large teams.

There is also a nuance the talent thesis likes to skip. The value of a great researcher depends heavily on the infrastructure behind them. A brilliant person with no compute ships nothing. As one analyst told CNBC, if the cost of building these models fell by ten times tomorrow, the pay for the people who build them would fall too. Talent and infrastructure work as multipliers of each other rather than as rivals. The genius and the GPUs only matter together.

Where this leaves us

So which story is true? I think both are, and the tension between them is the actual insight.

Infrastructure has become necessary without being sufficient, and it is no longer special. Every serious lab now holds comparable compute and comparable capital. When the inputs converge like that, the thing that explains who wins is whatever remains once you subtract them, and what remains is the people. That is why a resignation moves the stock more than a spending announcement does. We have shifted from a phase of AI limited by capital to one limited by talent.

The deeper risk for Alphabet has little to do with whether Jumper or Shazeer can be replaced as individuals. It sits in what their choice signals to everyone else. Google’s most valuable asset in this race may be the hardest one to find on a balance sheet: the belief, held by the best researchers in the world, that Google is where the frontier is being built. That belief lives nowhere in the accounts, yet it is built entirely from where people choose to work. In June, a few of the most respected people in the field chose to work somewhere else.

If the talent thesis is right, the $190 billion is best understood as the price of being allowed to compete for the people, rather than as a wasted bet on hardware. The companies that win the next decade will be the ones the best researchers actually want to join. Spending the most on chips has already shown that it does not guarantee that.

That is the talent to valuation ratio. The market has started pricing the people above the machines, and the week Alphabet lost $269 billion is the clearest sign yet that the repricing has already arrived.