In the spring of 1848, Sam Brannan walked through San Francisco waving a small bottle of gold dust. James Marshall had discovered gold at Sutter’s Mill that January. Brannan gave the news a tangible form: something people could see, carried through the streets. [California Geological Survey]
On December 5, President James K. Polk confirmed the discovery to Congress, citing military reports. His message also described necessities becoming more expensive, labor commanding extraordinary prices, and ships losing their crews to the mines. The rush was already changing the cost of participating in it.
A bottle of gold could establish that the opportunity was real. It could tell a prospective miner very little about the return on the next patch of ground. The location, recoverable quantity and expense still had to be worked out.
That gap between evidence of value and knowledge of its distribution is what interests me in today’s scramble for private business data. A valuable collection can draw attention to an entire category long before anyone knows how much of the remaining supply is commercially useful.
One recent argument on X brought the problem down to a price: the cost of making company records anonymous.
In an October post, Erik Newsham put that cost at roughly $35,000 per terabyte. A reply from Thomas Unise arrived with a much smaller figure: “35k/tb? Google is $1k/tb.”
The figures were not comparable quotations. Newsham’s spending was self-reported, with its full scope undisclosed. Google’s published prices separate inspection from transformation and vary by method. A cloud-service rate cannot, by itself, price an entire delivery process.
Still, the exchange exposed a useful question: when somebody buys “AI data,” what exactly are they paying for?
Newsham described a requirement that sounds small until it fails: different names, abbreviations and misspellings referring to the same person must remain connected after that person’s identity is removed. Otherwise, the records may stop telling a coherent story.
That is the industrial problem beneath the gold-rush language. Company files can contain valuable information, scattered across systems and mixed with repetition, missing context and material that cannot be supplied to a buyer.
Raw data resembles iron ore. Its commercial usefulness depends on the concentration of relevant information and the cost of recovering it.
Moving between China and Silicon Valley, I am interested in the business that forms around that recovery process. Who assembles the supply? Who determines what counts as useful? And who retains the savings when processing becomes cheaper?
My central argument is that the valuable scale in this market may come from covering a business problem thoroughly, rather than accumulating the largest archive. A supplier that can repeatedly assemble the right evidence at a predictable cost has a plausible business. A supplier with a large inventory still has something to prove.
The paid analysis below follows the economics from scattered records to usable supply: why buyers are spending, how aggregation changes costs, and where bargaining power is likely to settle.