OpenAI and Google are losing their top AI talent to startups even as cloud revenue booms

[Ronaldinho23]

The Motley Fool has a piece tracking a wave of senior AI talent leaving the big labs to start their own companies, using OpenAI's former COO Brad Lightcap and four top Alphabet engineers as the headline examples. Lightcap left days after a 7 billion dollar tender offer let OpenAI employees cash out shares, while Jeff Dean, Sanjay Ghemawat, Oriol Vinyals and Quoc Le all departed Alphabet on the same day to found a new startup called Discovery Loop focused on automating scientific research

The same day those four left, Demis Hassabis also stepped back from running Google DeepMind, and the article notes that with his departure, all eight original authors of the 2017 Attention Is All You Need paper that introduced the transformer architecture have now left Google entirely. That is a pretty remarkable full circle moment for the paper that basically kicked off the current AI era

The article's actual investment argument is that cloud revenue growth at Microsoft, Amazon and Google is not the problem, Azure topped 100 billion in annual revenue with an 84% jump in commercial backlog, AWS grew 37% with its AI business surpassing 25 billion in run rate, and Google Cloud grew 82% with a 514 billion dollar backlog. The real question the piece raises is whether value ends up concentrated in that infrastructure layer or migrates up to the model and application layer where all this departing talent is now headed

SpaceX gets floated as the wildcard threat to the cloud incumbents, with Grok 4.6 having just launched, Grok 5 reportedly training on SpaceX engineering data, and a rumored 60 billion dollar acquisition of the coding platform Cursor possibly closing within days. The one thing SpaceX is missing according to the piece is an established cloud business to match AWS, Azure or Google Cloud

It closes with the standard Motley Fool stock pitch about their top 10 list, which is worth taking with a grain of salt. The underlying observation about talent flowing out of the big labs into new ventures while the labs themselves keep printing huge cloud numbers is a genuinely interesting tension though

The AI Platforms Are Real. The People Who Built Them Are Leaving.finance.yahoo.com

[Oscar_75]

All eight Attention Is All You Need authors now gone from Google is one of those stats that sounds almost symbolic. The paper that built modern AI and the company that published it could not hold onto a single one of the people who wrote it

[Phil]

Genuinely think the model layer versus infrastructure layer debate is the single most important open question in AI investing right now. Nobody has a clean answer yet and anyone claiming certainty either way is probably overselling their own thesis

[HostFalcon]

SpaceX pivoting this hard into AI and cloud infrastructure while still running a rocket and satellite business is an extraordinary amount of simultaneous ambition. Wonder how thin that spreads their actual engineering talent

[Jonathan]

A 60 billion dollar Cursor acquisition would be a massive move for a coding platform.

That is a serious statement about how much value SpaceX thinks sits in that layer of the stack

[Rabbit21]

7 billion dollar tender offer letting employees cash out and then the COO leaves days later feels like there is a story behind that timing. The article does not fully get into it. Convenient timing if nothing else

[Pixel Mark]

The framing here feels off to me honestly.

Cloud revenue growth and talent departures are not really in tension the way the article implies, huge companies lose senior people to startups constantly regardless of how well the core business is doing

[Olivia78]

Not surprised senior researchers keep leaving for equity heavy startups. The payout ceiling at a public company is inherently capped. A supernova outcome founding your own thing is a different math entirely

[Michael]

Discovery Loop automating scientific research is the kind of startup idea that could either be genuinely transformative or another overhyped moonshot. Hard to tell this early which one it becomes.

Too soon to call

[Sandman]

The stock pitch tacked onto the end kind of undercuts the more interesting reporting earlier in the piece.

Would have been a stronger article without turning into a sales pitch for a subscription service by the final paragraph

[JayJ]

Curious whether Discovery Loop actually ships something useful within the next two years or ends up being one of those well funded research labs that quietly fades from headlines. Automating science is a much harder problem than automating code