It's the Data, Stupid
Why Silicon Valley and Wall Street are Reliving the Past ...Again
ChatGPT and Anthropic are changing the world. Just not in the way the market thinks. The market is again confusing enabling technology with the value creating asset.
I walked into a management of technology class in the winter of 1999 hosted for Fordham at the Bloomberg headquarters in New York. On the whiteboard was a single word. Google. The professor, Peter Keen, told us that it was about to become one of the most important companies in the world. He explained that their search algorithm was fundamentally different and would accelerate the utility of the internet through exponentially better search results.
Peter could see around corners and we became friends who bonded over baseball and good steak dinners in both Washington and New York. I’m glad that he saw his beloved Red Sox finally break the curse even if it was at the expense of my Yankees.
He introduced me to a couple of interesting companies that were well ahead of their time. The first was Xybernaut. Xybernaut pioneered the idea of wearable computers. The second was Webvan. Both companies flamed out for different reasons. Xybernaut for investor fraud and Webvan ran out of money.
Webvan raised $1 billion of venture capital and was looking to pioneer online delivery of groceries. Peter introduced me to Webvan and I interviewed for a job. I didn’t get it because I called their baby ugly. I challenged their logic. I questioned whether plowing $1 billion into a sophisticated national distribution network made sense before they proved the model in a single tech forward market like San Francisco or Austin. They were sitting on $1 billion of capital and they didn’t have to spend it like a kid who got hot at the craps table during a stag party in Vegas.
Nathan Bedford Forrest was famous for saying that the way to win battles was being firstest with the mostest. He was both a detestable human being and horribly wrong. Being first with capital is not the same thing as understanding an industry’s logistics and dynamics. As Omar Bradley reportedly said, “amateurs study tactics; professionals study logistics.”
Webvan got there firstest with the mostest. So did Pets.com. The sock puppet got an obituary.
Chewy and Instacart got there later, cheaper, and with a clear understanding of where the value lived.
Webvan and Pets.com confused speed and capital with sustainable competitive advantage.
The enabling technology was real. The market was real. The asset wasn’t the pipe. It was the underlying data and the system built on top of it. Like Forrest, they were horribly wrong.
Being first with capital is not the same thing as being right about where value accrues.
The cable bundle was a forced subsidy for 200 channels that nobody watched. Did anyone really need both Lifetime and the Hallmark channels? One channel with Christmas movies that inevitably end with star crossed lovers kissing in a small town gazebo was certainly enough.
Netflix launched their streaming at $7.99 per month. Cable raised their prices and added on impossible to decipher fees. Comcast, Time Warner, Optimum, and Cox Communications built the superhighway that carried the getaway car that robbed their bank.
Initially, every Netflix stream, every YouTube cat video ran over cable broadband. The cable companies conflated the value of the pipe with the value of the content and paid for it with long-term market cap destruction. While cable TV revenue shrank by $17 billion between 2017 and 2024, Netflix and YouTube grew from $11 billion to $96 billion combined, much of it delivered over cable broadband financed by their competitors.
By the time Comcast realized the underlying value of their content library and launched their own streaming service Peacock, they had already lost the race.
Last week I wrote about the New York Times and the Washington Post. The Times pivoted to digital successfully. The Times expanded their subscriber base by building a content platform to include games, cooking, sports, and product reviews. The Post has not. The Times understood that content rather than platform drove value. The Post is struggling to align its content, product and cost structure into a sustainably profitable model.
The market is making precisely the same category error for the third time in twenty-five years. The reported valuation of OpenAI was $730 billion on a revenue run rate of $13 billion. Anthropic was valued at $380 billion on a revenue run rate of $14 billion. Neither OpenAI nor Anthropic is a pre-revenue dotcom with a sock puppet mascot. The revenue growth is both real and unprecedented.
But the market is pricing them as if the large language models are the asset. It is not. The model is the enabling technology. The pipe. The data is the asset. The models matter, but only to the point of sufficiency. Sufficiency works for writing resumes and emails. After that, differentiation and value creation shifts away from the model and toward the data and the systems built around it.
OpenAI and Anthropic will print money licensing these models to the enterprise for the next few years. But gravity always wins. Deepseek and other cheaper models are already entering the market. Pricing power and margins will inevitably erode.
Epic has 300 million patient records. Workday has twenty years of curated finance and HR data. Salesforce has the commercial relationship map of the global economy. Bloomberg has forty years of market microstructure. Thomson Reuters has a century of legal precedent. These are not solely applications. They are repositories of proprietary, high-fidelity data that no one else can access or replicate. When they embed an LLM, the model becomes the interface. The curated data becomes the moat. A generic Claude API call built on publicly available internet data doesn’t know your organization’s compensation plan or your operational cost structure. Workday Sana trained on your data knows both.
Workday Sana powered by an LLM beats your ERP with a rogue finance user feeding spreadsheets into a generic unlicensed LLM every time. Because the data matters more than the model. The model has to be good enough. The market is currently valuing the middle of that stack as if it’s the top. That’s mispriced. It’s a fundamental misunderstanding of the value chain of enterprise software.
The stock market is stripping hundreds of billions from the market capitalization of enterprise SaaS platforms because it thinks they will be replaced by AI. A more likely narrative is that AI will enhance the value of these platforms over time by exposing the underlying data to improve decision making and identify trends.
Payroll has to be perfect. Financial statements have to be perfect. Can a probabilistic LLM sitting on top of a generic vibe coded infrastructure accurately calculate payroll for Aliquippa, Pennsylvania? City wage tax? School district earned income tax? State withholding? Local services tax? Every decimal? Every employee? Every pay period? With a complete audit trail?
LLMs are probabilistic engines. Enterprise financial systems are deterministic. That’s not a technical distinction. That’s the entire argument.
There is a place for a probabilistic model. It’s great at predicting the next word. Do you want an air traffic control system that’s right 95% of the time on approach to LAX? That’s what a probabilistic system can achieve today.
The end state isn’t always a winner take all cage match. In this case, the general purpose large language models are enabling technologies that enhance and extend the value of the existing enterprise application stack. Think of it as a three layer cake. Infrastructure at the bottom, Models in the middle. Curated data and purpose built systems at the top.
The internet didn’t kill Oracle. It became the delivery mechanism for Oracle’s databases and applications. LLMs won’t kill Workday or Salesforce. Workday will leverage LLMs to provide insight and efficiency to enterprise applications that power HR, finance, payroll and planning. The pipe doesn’t displace the content, it carries it and amplifies its value.
The market has confused enabling technology for the asset three times in the last twenty-five years. Google dominates search. Salesforce, Thomson Reuters, and Workday lead their respective segments. With each shift in technology, the companies that owned the data, the actual content won. This outcome is likely to be the same with AI. There will be clear winners. Claude is real. ChatGPT is real. The revenue growth is extraordinary. But the moat is data.
It’s the data, stupid.