The true strategy of AI companies is all written in the matter of 'hiring'.
Stop watching the press conferences.
The real strategic signals are hidden on the recruitment pages.
Researchers at Epoch AI recently did something—analyzed the public recruitment pages of OpenAI, Anthropic, xAI, and DeepMind, inferring the true direction of these top AI labs from the distribution of positions.
As a result, it exploded.
Sales positions skyrocketed, are AI companies starting to 'struggle to sell'?
In the past year, the proportion of sales and related positions at OpenAI and Anthropic has skyrocketed—
Anthropic surged from 17% to 31%, and OpenAI jumped from 18% to 28%.
The largest recruitment category for these two companies is no longer research, but sales.
What's more eye-catching is: the fastest-growing positions are technical roles that 'help customers learn to use AI'.
What are 'AI Success Engineer', 'Frontline Deployment Engineer', 'Solution Architect' ...
To put it bluntly, customers bought AI but can't use it.
The task of these positions is to teach enterprises, step by step, what AI can actually do.
Isn't this a roundabout acknowledgment - that AI implementation is far more difficult than imagined?
OpenAI is secretly building hardware, and it also has a camera
The most explosive clues on the recruitment page have arrived.
OpenAI has 15 positions aimed at a consumer-grade portable device.
The details are absurdly fine:
Hiring 'Camera ISP Software Engineer' - with camera
Hiring 'Operating System Engineer' - self-developed chip
Hiring 'Research Engineer' - run Transformer models directly on the device
This is not a mobile phone, this is a mysterious hardware that can run AI on the edge and has a built-in camera.
DeepMind is also working on XR glasses, with voice interaction as the core.
Both giants are betting that hardware is the next stop.
Robots are closer than you think
OpenAI posted 7 robot positions at once, focusing on large-scale simulation training, flexible components, and mass production ramp-up.
DeepMind is even bolder, with 9 positions aimed at humanoid robots with dexterous hands.
Stop saying AI only knows how to chat - they are growing 'bodies'.
Computing power strategy, completely diverging
The subtlest divergence is here:
OpenAI has 21 self-developed chip positions, determined to build their own.
Anthropic has no chip positions but is madly hiring for data center contracts and guiding cooling system design.
One is going for 'self-developed hard power', the other is 'outsourcing + deep binding'.
Who is right and who is wrong? I don't know. But both companies believe computing power is crucial.
xAI's style is completely different
On xAI's recruitment page, the most eye-catching is 27 positions for artificial data annotation.
Other companies outsource, xAI does it themselves.
Moreover - they dare to recruit publicly.
This indicates that xAI considers 'data quality' as a core competitive strength and is working hard on it.
Final feelings
The recruitment page is the best strategic signal source.
These positions are not 'hiring people', but telling you:
They think what is difficult (AI implementation)
They want to do what (hardware, robots)
What are they afraid of (computing power being a bottleneck)
If you don't understand the press conference, go to the recruitment page.
There are no PR scripts there, only the real battlefield.