Anthropic spends about $10,000 a head on forward-deployed engineers while AWS and OpenAI spend billions

Start with the cheapest program. Anthropic's $100 million spread over 10,000 engineers comes to $10,000 a head ^[15], or about 5% of the $195,456 median advertised salary for the role ^[16]. A figure that low works because Anthropic is training engineers already employed by its customers and partners ^[5]. AWS is building its own organization from experienced Amazon and AWS engineers plus outside hires, and Asa Kalavade, its vice president of forward-deployed engineering, said it is starting with hundreds ^[11]. CNBC did not report the period over which AWS will spend its $1 billion ^[1].

CNBC's explanation is that Big Tech, after spending billions building AI, is now spending heavily on the people needed to prove those investments are paying off ^[20]. "The Street is starting to become impatient with these billions of companies who are spending on AI without getting any business outcomes," said Jeff Christian, founder of the executive search firm Christian & Timbers ^[9]. The same week, AI-themed and memory-themed stocks including Dell, CoreWeave, AMD and Micron took significant short-term losses as concern about the economics of the boom intensified ^[10].

We think the budgets are a bet that people at the customer's site, more than model quality, limit what enterprises get from AI. Other readings fit the same numbers. Consultancies supply a large share of the demand: Deloitte, Accenture and KPMG together account for 2,007 of Lightcast's postings ^[17], at most 27.5% of the 7,300-plus total ^[18], and CNBC attributes that hiring to client demand for implementation help ^[8], so that slice answers clients directly and does not depend on what the model labs earn. Postings are advertised positions, not completed hires ^[8]. And supply may loosen. Elena Magrini, Lightcast's head of global research, said "FDE is not an entirely new role. It shares roughly half of its skills with software engineering, creating an opportunity to tap into a much broader talent pool." ^[13] Anthropic's 10,000-engineer target is about 1.4 times the 7,300 postings Lightcast counted in nine months ^[19], which suggests it is not competing for the same hires.

Training has a limit. Val Henderson, CEO of Caylent, an AWS consulting partner that embeds these engineers with customers, said "The hardest thing to teach is the customer EQ, without question." ^[14]

The thesis fails if postings keep climbing past 7,300 while customers still report no business outcomes, the complaint Christian describes.