Platform or Perish in the AI-Driven Economy

Assistance in Navigating the Minefield

Let’s begin by looking back nearly a decade to this article, “The Rise of the Platform Economy”, written by Irving Wladawsky-Berger, who is retired from IBM and an affiliate researcher at MIT.

“A platform or complement strategy differs from a product strategy in that it requires an external ecosystem to generate complementary product or service innovations and build positive feedback between the complements and the platform. The effect is much greater potential for innovation and growth than a single product-oriented firm can generate alone.” — MIT Professor Michael Cusumano

Although some forms of platform companies have been around for centuries, the type we are concerned with today are driven by the network effect. A few Big Techs have of course exploded in valuation recently due largely to expectations that LLMs and GenAI will favor their companies over all others.

“There is clearly a rising platform economy shaping our global business landscape and affecting the lives of citizens worldwide,” says the report in its concluding paragraphs. “This new form of organization seems to be a robust — some would even say dominant — form of business enterprise in the digital economy… While significant challenges lie ahead, the opportunities that platforms reveal are enormous, tapping into an unprecedented level of global Internet connectivity, and a large supply of talent and software skills… to develop the platforms of tomorrow.” — The Rise of the Platform Enterprise: A Global Survey led by Peter Evans and Annabelle Gawer (January, 2016)

At the time of the above-mentioned report, platform companies in the Bay Area had a collective market cap of $2.2 trillion dollars or 52% of the value of all the companies surveyed. Seattle ranked second with four platform companies worth $767 billion. Less than a decade later, the top U.S. tech platform companies have a market cap over ten times that amount. In addition, traditional companies have dramatically increased their own platform dynamics, in large part in reaction to losing market share to platform startups and Big Techs.

Amazon vs Walmart

Even discounting bubble valuations, increased revenue tells quite a story. Amazon for example closed out 2016 with $136 billion in annual revenue. Their annualized revenue rate in Q4 of 2025 is well over 5x that amount at about $800 billion. Amazon’s largest competitor in retail is Walmart, which has increased platform dynamics considerably in recent years, but is not growing nearly as rapidly as Amazon.

In 2016, Walmart’s annual revenue was $482.1 billion. The company enters Q4 of 2025 with an annualized revenue rate lower than Amazon for the first time, despite several platform-related businesses growing in the double digits, including e-commerce, advertising, clubs, and subscriptions. Amazon’s continuous growth represents what Jeff Bezos refers to as the Amazon flywheel—a reinforcing virtuous dynamic. That flywheel represents the greatest threat to many companies.

Today Amazon has over 100 subsidiaries and brands across e-commerce, cloud computing, media, healthcare, robotics, and AI, among others. One of the company’s greatest advantages is of course AWS, the leading cloud services business that appears likely to surpass $200 billion in annual revenue in 2026. Unlike e-commerce, elastic cloud computing that Amazon brilliantly pioneered is a highly profitable business that delivers over half of Amazon’s operating profit today (or about $40B in 2025), which enables an otherwise low margin company to invest more heavily in innovation and growth than competitors in most of its other businesses. Combined with the very large customer base and business ecosystem in Amazon’s market-leading e-commerce business makes expansion into new markets easy in comparison to others.

Amazon vs UPS

The recent story of UPS serves as yet another wake-up call for business leaders. For years, Amazon was UPS’s largest customer, but while UPS delivered its packages, Amazon was quietly building its own platform—a vast, data-driven logistics network. I remember years ago when a senior manager at UPS was questioned by media whether they were concerned about Amazon becoming a competitor, the manager said something like “we have a great relationship with Amazon”… we’re not worried at all. I immediately thought “right…you will be”. I founded my first digital business shortly after Amazon which became a niche market leader and have followed Amazon ever since.

The relationship between UPS and Amazon is quite different today. They recently agreed to reduce shipments with UPS by 50% due to the low shipping rates and impact on UPS’s financial performance.

“Amazon is our largest customer, but it’s not our most profitable customer,” Tomé said (Carol Tomé, CEO of UPS). “Its margin is very dilutive to the U.S. domestic business.”

UPS executives said about 20% to 25% of volume in their U.S. network is tied to Amazon, yet the percentage of revenue was only 11.8%. That’s a difficult position to be in, essentially subsidizing one of their largest competitors. A classic manifestation of Jeff Bezos’ “your margin is our opportunity”. As Amazon increases market share in shipping, the market share of UPS, FedEx and others will shrink unless they find or create other areas of growth.

AWS has allowed Amazon to become the leader in R&D spending (about $90B annually, roughly the same as the UPS total revenue) and invest in many other businesses, including the largest private parcel shipping company in the U.S., passing UPS in 2022.

UPS has been a well-managed company, and is still the world’s largest parcel shipper by revenue, but Amazon is still growing rapidly and UPS is not. Walmart and UPS were the market leaders in their industries for decades and had many strengths, but it wasn’t enough to defend their market leadership. That’s the power of the platform business.

The primary obstacles to platform dominance are regulated industries. Amazon, Apple others and have made some progress expanding into healthcare and banking, but it’s much more difficult than unregulated industries. Banks and healthcare companies are responding by strengthening their own platforms.

Why LLMs Triggered an Arms Race

As the name implies, large language models require scale—far more than originally thought to deliver products people would use, much less generalized knowledge on every topic. The transformer architecture was introduced by the famous paper, “Attention Is All You Need“ (Vaswani et al., 2017), but it was the paper “Scaling Laws for Neural Language Models“(2020) by a team at OpenAI that revealed the obsession with scale at OpenAI, which spread to others.

DeepMind challenged the OpenAI claim that model size was (almost) everything in their Chinchilla paper (2022), finding that the model size and the number of training tokens should be scaled equally. It has since been demonstrated that LLMs perform worse on certain tasks as they scale (Lin et al., 2022). As I recently said in a private conversation:

“LLM chatbots are like rebellious teenagers... you can make suggestions and attempt to require them to follow the rules, but they do whatever they want anyway.”

The performance of the extremely over-hyped ChatGPT 5 was so underwhelming that it served primarily to confirm the trough of disillusionment in LLMs. A significant portion of large high-profile VC firms in SV, LLM firms, and Big Techs had already bet the farm on LLMs who were chasing an elusive chameleon known as AGI. Given current market caps, incentives, and related pressures, I’m skeptical the industry will proactively deflate the LLM bubble by managing expectations.

A Glimpse of the Present

The post below was featured by LinkedIn editors on Friday, October 31st, while I was writing this newsletter. It linked to an article in the WSJ titled: “Big Tech Is Spending More than Ever on AI and It’s Still Not Enough”. My comment in the post:

“There will never be enough money in the AI arms race, or for those chasing superintelligence with scale, but that’s becoming less of a factor for many as the enterprise market evolves. AI isn’t a religion in business as it is in a few LLM cults, but rather a very significant risk and opportunity. It’s an existential risk for many.

The more practical the better in most business cultures, but that doesn’t necessarily translate to the need for immediate ROI. For example, a logical strategy for the few companies enjoying the most cashflow and strongest balance sheets is to subsidize investments in the AI arms race, AKA predatory pricing—which is very similar to the dot-com bubble. That’s usually fine to a point, but some of the best capitalized companies are already hitting a wall.

What isn’t well understood is that the risk isn’t limited to smaller companies, which on average have not experienced nearly as much increase in productivity as the largest. LLMs favor large data stores (or swamps in some cases), which is why Big Techs are so obsessed with them, but as more mature systems become adopted like our KOS, the advantage of data scale will begin to evaporate as precision high-quality data generates greater ROI for the majority.

Until this point everyone from POTUS to most Wall Street analysts to VCs and individual investors have generally ignored the risks to Big Techs due to their enormous cashflow and strong balance sheets, but that is already beginning to change. Not only are LLM firms like OpenAI involved in highly creative financing to say the least—committing to about one $trillion more in spending than they have capital to spend, but Big Tech commitments are also beginning to exceed their ability to finance with cash, indicating much greater risk than reflected by market caps (Meta is the latest in a large debt deal as they prepare a $30 billion bond sale).

Bottom line: chasing an ill-defined movable goal post like AGI with LLMs is a classic case of the greater fool theory in action, but at the same time AI investments are paying off nicely for a growing minority of companies. The related risks and opportunities are unprecedented in modern times. Highly competent, dispassionate decision-making has rarely been more important.”

Another popular post of mine this week at LinkedIn provides a clue where I believe AI is headed, based on three decades of R&D, including 15 years surrounding our Synthetic Genius Machine (SGM).

“One of the best ways to understand why LLMs are the wrong path is to first understand the power of E = mc2. Einstein captured the fundamental physics that applies to the entire universe -- all mass, all energy, in 5 characters. That’s the power of compression, and one of the most elegant examples of super intelligence in our existence.”

Survive and (Perhaps) Thrive in the AI-driven Platform Economy

There is no longer much debate about the power of the platform model. A combination of factors including consumer behavior, convenience, ease of use, deep analytics on customers, the network effect, and now AI is certainly sufficient to be a decisive factor. Excess capital, top-tier talent, and government capture doesn’t hurt, but acknowledging these factors doesn’t necessarily help, does it? Knowledge only becomes real-world value when actualized.

Here are five of the most important actions that can be taken to first survive and begin to thrive in the platform economy.

1) Focus on existing advantages and then enhance and expand them with the help of AI systems whenever it is pragmatic and affordable to do so. Expanding and strengthening existing moats is usually a good strategy, but it’s not always possible as many moats are easy to breach, erode, or drain by platform giants. In those cases, bolder strategies must be crafted and executed.

2) Protect and improve data quality as a top priority. LLMs are largely based on theft of intel, including IP. It’s no secret that a large number of employees admit to sharing sensitive data with consumer chatbots, or that the chatbot companies are attempting to monetize that information in any way possible. It’s therefore essential to provide employees with secure AI systems. It’s far more valuable to do so in a manner that is relevant to each person and role, but beware of the NIH syndrome—it can be fatal. Jamie Dimon recently revealed that JPM spends about $2 billion per year on AI and they are just now breaking even after a decade of sustained investment.

3) Rather than attempt to generalize knowledge (commoditized, inaccurate, and massively subsidized), hyper-focus on precision, quality data relevant to the business and employees that improves performance and provides an advantage over existing or would-be platform competitors.

4) Create your own platform ecosystems and/or expand them so they are more customer-centric, competitive, and defendable. This requires advanced technology, strong security, and potentially new native products. New products and related growth are the least realized opportunity in enterprise AI (by far).

5) Consider leveraging other platforms, but do so carefully. For example, many software companies host their apps on AWS Marketplace to increase sales. The reality is that whether one uses Marketplace or not, AWS may attempt to develop a competing product—a common behavior in platform companies. Others have moved off of AWS and reported millions of dollars in savings.

One thing to consider is that Amazon’s commission for e-commerce vendors started between 8 and 15% and is now reported to be over 50%. The same thing may occur with AWS Marketplace. It’s wise to avoid becoming reliant on any single platform, encrypt the product or service if possible, and be very careful with descriptions to make reverse engineering more difficult than it’s worth.

Our Approach at KYield with the KOS

We are absolutely committed to providing customers with a sustainable competitive advantage and have dedicated decades of deep work to do so. That philosophy covers a lot of architectural territory in an enterprise AI operating system like the KOS. One way we do this is by offering the Knowledge Network function that enables customers to create and protect their own high-quality networks and ecosystems, acting as a tailored platform optimized for and enhanced by AI systems.

By providing end-to-end data across the digital workplace in a manner that is pre-structured for optimal security and performance, businesses and individuals are empowered in a manner that was impossible to do only a few years ago, but is rapidly becoming essential for survival.

“The thing that really frustrates me is our ability to create a foundation off of which AI can work. ...We have a lot of unstructured data, and we have to figure out how to manage that data in order to be able to (take) advantage of the AI.” -- Anonymous CEO, via Alan Murray‘s CEO Brief at the WSJ.

AI is a systems challenge—human, organizational, economic and technological. System design is therefore paramount. Although we embed many interconnected functions in the KOS, the underlying priorities for the functionality include security, productivity, quality, adaptability, relevance, ease-of-use, affordability, and continuous improvement, all within a single unified system. We strengthen sovereignty in many ways, including self-tailoring and interoperability. If something were to happen to KYield, customers already own and control their own interoperable data. These are non-trivial priorities to deliver in an EAI OS architecture, and obviously extend far beyond a chatbot.

I recently produced a brief presentation with NotebookLM based on our 15 EAI principles and a cybersecurity scenario paper that demonstrates why each of the principles are necessary (executable in the KOS). Our long-term board member Robert Nielson and previous board member VADM Phil Wisecup (Ret.) helped craft the scenario. Produced with NotebookLM, the bot makes a few minor errors like mispronunciation of KYield, but since the data is from the source it’s unusually accurate. Have a look.