The Compute Elite Class Society Has Arrived

Everyone says the AI era will produce the one-person, billion-dollar unicorn. Loosely speaking, the "solo unicorn" has already arrived.

There's an amusing rumor that Meta tried to poach Mark Chen, OpenAI's head of research, with an offer of $1 billion over four years! It didn't happen, apparently. What the Wall Street Journal reported was fairly vague on the details — the more concrete figure floating around was $300 million over four years for other poaching attempts, some of which succeeded and some of which didn't. All the specific numbers carry a whiff of rumor, since it's unlikely anyone would publish a complete poaching package, especially for a deal that never closed. But Meta's talent raids in AI have become notorious in their own right — a case of paying a fortune for a horse's bones, so to speak, to prove you're serious about buying the real thing.

Rumor: Meta Dangled a Billion-Dollar, Four-Year Package to Poach AI Talent

Meta launched a wave of poaching, plus "talent lock-in" through investments and stakes in AI startups. It started in June, when Meta paid a massive sum for nearly half of Scale AI and brought in its CEO, Alexandr Wang, to run Meta's AI lab. Wang's network turned out to be every bit as strong as advertised — he immediately kicked off an aggressive poaching campaign, pulling several senior OpenAI executives, and also spent $200 million to lure away Apple's AI lead.

But Meta's poaching wasn't actually the story people talked about most. The one that got the most attention was Windsurf, a startup focused on AI coding tools — a product I happen to like a lot myself.

The company was genuinely successful: it had already surpassed $80 million in recurring revenue and had raised over $240 million, which brought Google knocking with an acquisition offer. Google originally wanted to buy the whole company for $3 billion, but negotiations fell apart. In the end, Google paid $2.4 billion for licensed technology and brought the core team on board.

This kind of deal is, in essence, an acqui-hire in disguise — it lets Google rapidly absorb top AI talent without triggering an antitrust review.

Google didn't take an equity stake in Windsurf — it simply paid a licensing fee for technology rights and brought the talent in-house. Windsurf's original investors walked away with a solid cash return while still holding onto their remaining equity in the company. Google's official line was: "We're excited to welcome top AI coding talent from the Windsurf team." Meanwhile, the original Windsurf entity didn't actually dissolve — most of its employees stayed on, meaning that this larger remaining group was, by implication, "not the top talent." They'd probably thought joining a unicorn meant an easy win down the road — a rather brutal turn of events. It's a bit like something out of Romance of the Three Kingdoms — Guan Yu telling the sidelined general: "Liao Hua, thanks for your service — the Five Tiger Generals are off to serve Wei now."

Fortunately there's a happier postscript, since this was, after all, a top-tier company: the new CEO moved quickly and merged with another AI startup, Cognition, absorbing all the staff Google hadn't recruited. The deal worked out beautifully, thanks to strong technical complementarity between the two sides, and it seems Windsurf's investors came out ahead yet again — a win-win all around. Those remaining employees turned out to be genuinely top-tier Silicon Valley talent after all, nothing like Liao Hua — if anyone's to blame, it's simply the brutal logic of capitalist valuation. Still, a happy ending in the end.

Top AI Talent Is Now Worth as Much as All of a Company's Other Assets Combined

To be fair, though, the appeal of these giants isn't just money — joining them also means access to compute. Top talent at the mega-labs can use as much compute as they want, and compute is what it takes to break real technical ground. Elon Musk has proven this point — compute plus data, brute-force simple, and his xAI is doing just fine. Without compute, any edge in thinking or technique is nearly impossible to validate. I saw an interview recently that I found fascinating: an AI company CEO wanted to give an employee a management role and offered to add headcount to their team — in a traditional workplace, that would be a clear win. But the employee was confused: why not just give me more compute instead? I don't need more people! That, to some degree, reflects just how pervasive this compute-is-everything mindset has become.

Also — and this is only a rumor, since the deal never closed — it's worth pausing on the fact that Mark Chen is Taiwanese-American, which makes him something of a point of pride for Taiwan too. Put dramatically: he alone is worth roughly the market cap of a listed company like Chief Telecom (是方電訊). But set aside this bidding war for elite talent, and on the other side you have Big Tech continuing to lay off staff and cut hiring (presumably to free up the cash to pay for this elite talent and its compute).

Everyone's talking about AI's impact on employment, and it feels like the future of the workplace has already arrived at the cutting edge of the AI industry: a handful of people at the top are worth almost as much as everyone else combined. Whatever's left over — the work machines can't do — becomes the human's job, and "human labor" is fast becoming synonymous with "low value-added."

If even the AI industry itself isn't immune to this dynamic, then what happens to the rest of us — people who drive worse than machines, who code worse than machines? In the software and internet field I'm in, I think future opportunity will concentrate around partnering with Big Tech's ecosystems, because this AI paradigm shift was set in motion by Big Tech itself, which makes it extremely hard to challenge them head-on. But partnering with them can be an effective way to complete your own transformation amid this AI wave.

For Taiwan, the silver lining is that AI compute demand should stay strong for a good long while, which means the semiconductor industry — and everything tied to compute — should still see a fairly healthy period ahead.