Ziff Davis chief tests the DOJ's affordability claim against OpenAI's $750 billion compute plan

Two of the numbers in the argument do not share a unit. OpenAI's $750 billion is what the company has told investors it expects to spend on compute by 2030 ^[9], a cumulative figure. The under-$20 billion for royalties and licensing payments in the US music industry is a yearly one ^[10]. Spread the compute plan evenly across the five years to 2030 and it averages $150 billion a year, which puts a $20 billion annual licensing bill at about 13 percent of the annual spend ^[20]. Shah wrote that similar payments to news publishers would be "a rounding error to LLM companies' overall expenses" ^[19].

The mismatch runs the other way too. The music figure is what an entire national industry pays, across every licensee in it ^[10], while the compute figure belongs to one company ^[9]. Split a $20 billion pool among the handful of labs with frontier-scale capital and no single payer writes a $20 billion check ^[21]. So the comparison sets an industry total beside one firm's budget.

Shah's case is that the market already clears at some price: dozens of agreements have been announced between publishers and frontier labs, most notably OpenAI ^[11]. "The publishing industry just needs willing counterparties. This has been done before, can be done today, and is not difficult," he wrote ^[13].

The sturdier evidence for a price is what labs already buy. Shah writes that LLM companies routinely pay for data they cannot scrape, and that companies like Scale AI have raised billions of dollars to acquire training data and sell it exclusively to frontier labs ^[14]. Scale AI's data sales have observable clearing prices, and those prices hold whatever the copyright case decides.

In my view the affordability half of the government's brief is the weak half: 13 percent of an average year's compute spend is a budget line for a buyer that has already forecast $750 billion of it ^[20][9]. The music comparison is Shah's hypothetical, not a quoted rate for news content, so if a licensing regime priced well above $20 billion a year the capital argument stops being rhetorical. And whether publishers are owed anything at all is the litigation's question, which the commentary sets aside: "This opinion piece is not about arguing the law," Shah wrote ^[22].

One more caution on the record. The brief's language reaches readers here through a plaintiff's own commentary, quoted in fragments, and Fortune notes that the opinions in its commentary pieces are solely the views of their authors ^[18].