The world’s biggest tech companies are tearing up the rulebook in their pursuit of AI. Energy, infrastructure, talent. It’s about who gets there first.
What started as a race to build bigger models has morphed into billion-dollar job offers, reverse acquihires that hollow out startups, and spending sprees so large they rival national infrastructure budgets.
Meta alone is on track to pour as much as $72 billion into AI this year, yet last week it froze new hiring after bringing in more than 50 researchers from its rivals.
This is not just about one company’s bet. The scramble for talent is altering Silicon Valley’s culture, the economics of venture capital, and the future of innovation itself.
The moves may secure Big Tech a short-term edge, but they risk eating the seed corn of the ecosystem that fed them for decades.
Can money buy a brain trust?
In the past year, Meta has raided OpenAI, Google, Anthropic, Apple and xAI.
According to the Wall Street Journal, more than 20 hires came from OpenAI alone.
13 came from Google, 3 from Apple, 3 from xAI and 2 from Anthropic.
To land Scale AI’s founder Alexandr Wang, Meta paid $14.3 billion for a near-half stake in his company and installed him as chief AI officer.
Some recruits have reportedly been offered packages worth $100 million or more. One offer, to a co-founder of Thinking Machines Lab, was said to be worth as much as $1.5 billion.
This “pro athlete” approach to hiring has changed the tone of Silicon Valley.
Sam Altman of OpenAI has publicly mocked it as mercenary. Anthropic’s Dario Amodei has informed staff that he will not match such offers, despite Meta’s continued efforts to entice his employees.
Meta insists the numbers are exaggerated, but even if inflated, the perception is now set: the company is overpaying to replace the talent it squandered in earlier years.
Venture capital firm SignalFire reported retention rates this spring that underline the point.
Anthropic held on to around 80% of staff, DeepMind 78%, OpenAI 67%. Meta trailed at 64%.
The best researchers once saw Facebook’s FAIR lab as a top destination.
Many of those people left and founded the companies now challenging Meta’s credibility.
The reverse acquihire era
Microsoft, Google, Amazon and Meta have all embraced a new tactic to shortcut the talent war.
Instead of acquiring startups, they poach the founders and top engineers, sometimes paying large licensing fees for the company’s technology, then leave what remains to fend for itself.
Microsoft did it last year with Inflection AI, paying $650 million and hiring Mustafa Suleyman to run its Copilot business.
Google spent $2.4 billion in July to hire away Windsurf’s core team. Meta’s move on Scale AI is the largest of all, worth nearly $15 billion.
These deals avoid antitrust scrutiny, move faster than full acquisitions and give Big Tech the specific teams they want.
But they weaken the startup bargain that defined Silicon Valley: take high risk, hope for high reward.
Rank-and-file employees in gutted startups often end up with little, while founders and a handful of researchers cash out with life-changing contracts.
That erodes trust. If young engineers believe they will never see a true exit, many may simply join Big Tech from the start.
The result: fewer bold startups, fewer chances for the next Android or Annapurna Labs, and a shallower pool of ideas for the future.
Meta’s struggle to execute
The irony is that despite its record spending, Meta has little to show in terms of breakthrough models, and that’s reflected in its stock price.
Its latest flagship, the two-trillion-parameter Llama 4 “Behemoth,” was delayed and then abandoned after poor performance.
Engineers accused leadership of gaming benchmarks to make the model appear better than it was.
The majority of researchers who built the original Llama have since left.
Other AI companies, such as Mistral and xAI have reportedly snatched up a lot of these engineers, according to reports.
The company has now reorganised its AI arm again, splitting it into four units: one for frontier superintelligence models, one for products, one for infrastructure and one for long-term research.
The AGI Foundations team that worked on Llama was dissolved. Behemoth is gone.
A new closed-source model is being built from scratch, marking a sharp break from Meta’s open-source philosophy that had made Llama popular among developers.
For now, the showcase is not a model but the roster of names: Wang, former GitHub chief Nat Friedman, Safe Superintelligence co-founder Daniel Gross, and Shengjia Zhao, a co-creator of ChatGPT, who was appointed chief AI scientist.
CNN listed all of Meta’s new AI hires. But whether this star team can deliver is another question.
Meta has already had to freeze hiring to give the new structure time to settle.
The economic trap
Meta’s problem is structural. Microsoft, Google and Amazon can recycle AI infrastructure costs by selling cloud compute.
Meta has no such business. Every dollar of its AI spending has to be paid back through its own ecosystem.
The math is daunting. At the high end of its $72 billion capital expenditure forecast for 2025, even spread over five years, Meta needs to generate roughly $30 to $40 billion in new annual revenue to justify the investment.
That is on top of the $161 billion in advertising revenue it earned last year. It means the company must find a way to grow its core business by nearly a quarter in just a few years.
The most plausible route is advertising. No one else has 3.5B daily users, frictionless distribution, and an ads engine that can monetise tiny gains at gigantic scale.
Meta is rolling out tools that allow businesses to upload a single product photo and budget, then let AI generate entire campaigns.
If small and medium businesses adopt this at scale, Meta’s ad revenues could rise sharply.
But the tools need to deliver measurable conversion lift and not just slick demos.
Without that, the infrastructure spend risks looking like stranded capital.
Eating the seed corn
The broader risk extends beyond Meta. By overpaying for scarce researchers and hollowing out startups, Big Tech is weakening the ecosystem that produced their greatest assets in the first place.
Android was a tiny startup when Google paid $50 million for it in 2005.
Amazon’s chip unit, Annapurna Labs, cost $350 million in 2015.
Those acquisitions worked because the companies were intact and allowed to grow. The reverse acquihire trend instead destroys the base.
It also fundamentally alters Silicon Valley’s culture. Where once the missionary zeal of changing the world animated recruits, now the mercenary logic of cashing in dominates the headlines.
For Meta in particular, which already carries reputational baggage from past scandals, the optics are corrosive.
What comes next
The market has given Big Tech some leeway. Meta’s stock is up around 25% this year despite the spending.
Investors have so far tolerated the talent war, betting that breakthroughs will follow. But banks are starting to warn.
Morgan Stanley told clients this month that lavish stock-based pay at Meta and Google could dilute shareholder value unless it translates into real innovation.
The next phase of the AI race will show whether these billion-dollar bets buy leadership or simply bloat.
If Meta can deliver a closed frontier model that matches OpenAI and Anthropic while embedding AI into its advertising engine, the spend could pay off.
If not, the era of billion-dollar job offers may be remembered as Silicon Valley’s most extravagant own goal.
The post Why is tech at war over AI talent? appeared first on Invezz