China's chip firms are buying 85% of the payroll for under 6% of a revenue target

Back the headcount out of the coverage ratio and Cambricon employs about 1,107 people ^[1]. The 124 core staff whose roughly 600,000 shares have unlocked, at an average 5.57mn yuan or about $828,000 each ^[3], are around 11% of that payroll ^[2]. Their tranche also prices everyone else's: about 4,839 shares a head ^[3] at 5.57mn yuan implies roughly 1,151 yuan a share ^[4]. Carry that value across the shares promised to the wider group and the whole retention promise comes to just under 6% of the revenue figure the plan is pegged to ^[6]. For holding a design team together into 2028, that is not an expensive instrument.

The structure is the part that is awkward to copy. A rival hiring one of the covered engineers is not bidding against a salary but against unvested shares whose value depends on a corporate target the leaver stops being able to influence the day they resign ^[6]. Cash cannot be shaped that way. A raise clears the moment it is paid, so the platforms taking the pay-and-bonus route are buying loyalty with a shorter fuse.

Cambricon is running the narrow version alongside the broad one, and so is Zhongji InnoLight, which put 2.48mn shares in front of 99 key personnel at average yields above 26mn yuan ^[4]. That works out at about 25,050 shares each, or roughly $3.9m at the exchange rate implied by the dollar conversion in the source ^[8]. Breadth and depth are doing different jobs: one suppresses ambient attrition, the other holds a named handful.

Europe's problem is plumbing. Many of its relevant employers are listed incumbents, research institutes or university spinouts, and employee share schemes differ so much between member states that equity is a harder tool to reach for ^[9]. The shortfall is not bodies. On TNW's account the continent has more AI talent than the United States by headcount ^[7], and TNW's own summary is that of the three regional responses, courses are the only one that cannot be accepted immediately ^[13].

The American answer is priced rather than structured. TNW reports that Anthropic pays the highest salaries in AI research, to the point that its chief executive has worried aloud about people joining for the pay ^[11]. A grant that vests through 2028 selects for a different person than a top-of-market salary does.

Behind all of it is a demand curve, not a generosity contest. Chinese firms are poaching each other while export controls have made domestic chip design a national priority, which pushes all that demand onto a small pool of engineers ^[12]. Equity across 85% or 97% of a workforce is what a company does when it has concluded that the people it needs cannot be replaced at any salary it is willing to publish.