AI’s Capital Barrier: Why Monoliths Control the Foundation—While Entrepreneurs Win Above It

After several years of using ChatGPT as my primary LLM, I was surprised by how much Google’s new Gemini upgrade improved the experience. For my own workflows—writing, editing, image manipulation—it felt meaningfully better. And the shocking part wasn’t just Gemini’s quality. It was how frictionless it was to switch from one LLM universe to another.

Yes, I lost some personalization. But as a consumer, nothing gave me any hesitation from moving. Depending on the next leapfrog, maybe I’ll move back shortly.

This tells us something important: the switching costs in AI will accumulate around applications and distribution, not the core model IP. The capital and engineering required to build a foundation model heavily favor monolithic innovators. Nvidia moves deeper into software. Google designs its own chips. Microsoft does both. Amazon does all of the above.

A Pattern We’ve Seen Before—But Bigger This Time

My career has tracked three major tech paradigm shifts. Each time, the first encounter with the new technology was magical and transformational. My attachment to Apple, VisiCalc, Motorola, BlackBerry, Netscape, AOL Mail, eBay, and Yahoo all felt permanent—until faster-moving, more native, simpler competitors displaced them.

A familiar cycle repeated:

• The ecosystem unbundled
  
  
• Horizontal specialists emerged (chips, communications, software)
  
  
• New entrants built faster and cheaper on these components
  
  

Over the past decade, however, those horizontal layers have been reabsorbed into vertically integrated platforms. Mobile and cloud accelerated this trend—Apple, Google, Microsoft, and Amazon increasingly design or control each essential layer of their stack. AI takes this verticalization further and faster than anything before it.

AI Is the Most Capital-Intensive Technology Shift in Modern History

We tend to talk about AI as an “IP race,” but the defining characteristic of this era is simpler:

AI is a Capital racea AI is a capital race.

To compete at the foundational layer, vendors need to invest simultaneously across four deeply capital-intensive domains:capital race.To compete at the foundational layer, vendors need to invest simultaneously across four deeply capital-intensive domains:

1. Chips / Compute: Securing scarce, expensive GPUs/TPUs
   
   
2. Infrastructure: Hyperscale data centers, energy, cooling
   
   
3. Model Training: Hundreds of millions per frontier model
   
   
4. Distribution: Massive user bases to integrate AI into daily workflows
   
   

Only a handful of companies—Google, Microsoft, Amazon, Meta, Nvidia—have the capital structure, reliable access to incremental capital; at the lowest cost, distribution, and operational scale to run this race efficiently.

Their strategic advantage is growing, not shrinking. They include:

• Designing their own silicon (Tensor, Grace, Maia)
  
  
• Running their own clouds
  
  
• Training their own models
  
  
• Embedding AI into billion-user products (Google Workspace, M365, Instagram, Chrome)
  
  

This end-to-end homogeneity drives marginal compute costs down and operational leverage up. For users, switching LLMs remains easy. The real moats sit in distribution, application ecosystems, and multi-solution data platforms.

This era is not simply a broad creative explosion; It is matched with a structural consolidation event.

Where Entrepreneurs Win: The Three Frontier Opportunity Zones

The monoliths will own the foundation layer. But this creates enormous opportunity above it. New markets are emerging weekly, and the most compelling founder opportunities cluster into three domains:

1. Agentic Automation (The Workflow Master)

Problem: Enterprise systems grow more powerful and more complex—traditionally a trade-off.

AI’s Role: Domain-specific agents eliminate that trade-off.

Value: Agents that operate across ERP, CRM, Cyber, and internal data to autonomously execute multi-step workflows, surface insights, and make pre-emptive decisions.

This is the next “middleware revolution,” but intelligent.

2. Orchestration & Interoperability (The AI Control Plane)

Problem: Enterprises will use many models, each with different strengths and cost structures. Complexity and cost are exploding.

AI’s Role: Build a routing and governance layer.

Value:

• AI gateways that choose the optimal model per task
  
  
• API unification
  
  
• Reliability and cost control
  
   This is similar to the rise of cloud cost-optimization and observability—only bigger.

3. AI Security & Governance (The Shield)

Problem: AI adds new vulnerabilities—prompt injection, data poisoning, emergent behaviors, and insecure AI-generated code.

AI’s Role: The trust layer.

Value:

• AI-aware threat detection
  
  
• Governance & audit trails
  
  
• Hardening AI inputs/outputs
  
  

No enterprise will deploy AI at scale without this.

The Coming Divergence: Monolith Foundations and Entrepreneurial Frontiers

ChatGPT, Anthropic, Perplexity, and others are extraordinary companies addressing one of the largest markets in tech history. But their vulnerability is structural: the trillions of dollars required to keep pace with hyperscale incumbents whose cost of capital is lower and more reliable.

Meanwhile, the next wave of entrepreneurs will win by building capital-efficient frontier solutions on top of these foundational layers—where creativity, speed, and differentiation matter far more than the scale of your balance sheet.

Great companies will still be built with old-school fundamentals:

• Strong gross margins
  
  
• Real product differentiation
  
  
• High sales productivity
  
  
• Deep customer understanding
  
  

These are the timeless drivers of durable shareholder value.

Market Jolt

I've had a long and fruitful association with Elad Baron, a serial entrepreneur whose interests range from Cyber security to consumer marketplaces. Elad has recently introduced a new company,  Market Jolt, which is a site which surfaces experts for stock trading.  He believes that today there is no efficient way for consumer sock picking experts to be properly compensated for sharing their picks. On the other end of this marketplace are people looking for transparency and guidance for their financial transactions.

'What is" vs the "what was" revolution

I've been involved with a few companies who are riding (or fell off the horse) in the relatively new Discovery space. Over the past 15 years consumers and businesses have derived great utility from search, a function which enables people to find information about what has happened or what was posted, frequently days, months or years previously. Businesses also received great rewards by advertising against the keywords used for these searches. More recently, and certainly the combination of mobile geolocation with sites such as Pinterest. WeHeartIt. Twitter  a firehose of data around what IS, vs what WAS is now available for Enterprises.

All this data, which I call the Social Highway, has many on and off ramps. The on ramps include countless blogs, data streams, photo streams, commerce streams, etc.; all delivered in real-time. Compiling the data is relatively easy, turning it into actionable information is much more difficult and is the secret sauce behind companies which built solutions optimized for real-time big data sets to measure and analyze:

• Buying intention
• Customer support
• Brand health
• Influencers (positive and negative)
• Psychographics

While this information is powerful, the real impact will only be felt as Enterprises adjust their business processes to direct the information to the right people, put in place processes for institutional learning and retention, and do this in a cross-silo manner. Unlike today's Enterprise, where each department has its own information silos, supported by people and processes, the Social Highway's on and off ramps do not respect traditional turfs. Customers may simultaneously require support, sales and marketing assistance. Influencers (and their minions) can be instantly pinpointed for special treatment, akin the TSA fast-lanes and be couponed, receive priority support and surveyed, all in one interaction. And all these interactions are no longer asynchronous.

Enterprise structure will have to be more elastic. It's a revolution in the making.

Looking after business

Last week I had breakfast with 'smart' Larry. Over a bowl of oatmeal he opined that at the end of the day, the role of an investor is to create returns. He went on to note that entrepreneurs, VC's and entrepreneurs easily get confused when caught in turbulent times, or when money flows like a river. He highlited that it's the money, not necessarily company building that is key to investor survival (e.g. building companies, but not showing great realized returns will not get you to the Limited Partner fundraising finish line). As usual, Larry is right.

It's important for entrepreneurs and investors to recognize that it's the making money part of the business that is ultimately key to survival. Often this creates positive alignment between entrepreneurs as well as multiple classes of investors. However, sometimes (e.g. when a company is not doing well) it creates a zero sum situation. When this happens, as is more common than not, the rational expectation is that folk will look after their own interests. If someone acts irrationally, it's usually the entrepreneur, who bends over backwards to take care of employees and investors.

As Tessio said in the Godfather " it was only business"

Constrained, shared, and pirated media

One of the great things about the technology industry is that, when a technology paradigm shifts, hard and fast rules for success begats soft and slow companies which can't or won't adust to them. IBM, Yahoo and Microsoft are examples of survivors which are, did, or will reinvent themselves to deal with the PC, social, and mobile paradigms. One of the early rules of the social web was really a short hand equation:

For every person who posts;
Ten will comment
Eighty-nine will silently read

The explosion of Twitter,Vine, and Snapchat turned that equation on its head, and while doing so, created a massive wealth opportunity for a few entrepreneurs and their backers, opened up new categories for innovation, and took a great deal of 'friction' out of the user experience.

A few months ago I read a great post by Andrew Chen on constrained media. He argues that a class of applications which have intentional limits (140 characters, 6 seconds of video or 10 seconds per photo) blows apart the above rule of thumb. These constraints have the glorious attributes of simplifying product management to concentrate on an easy experience, creates its own context, and most importantly to me, makes casual posting acceptable. It's no problem if you don't have the skills to write the great American novel, 140 characters, replete with abbreviations is more than welcome! Tearing down the walls which hindered sharing, engagement and virality are good things.

One area that he did not mention, and it fits within his theme, is the concept of reblogging. Pioneered by Tumblr and fast followers Pinterest and WeHeartIt, these services greatly reduced the friction associated with repurposing (some may say 'borrowing') photos and other IP via one click snipping. It made it incredibly easy for people to express themselves and visually fashion their online identities. While doing so, it created legions of followers, shared media, and engagement.

A holy grail for wealth creation.

Apple thoughts

Yesterday, at its World Wide Developer Conference, Apple presented a series of product-line extensions ranging from an updated operating system, to a streaming music service, and a series of hardware tweaks. By the end of the day, the press was somewhat disappointed and  the stock market was mixed to down.

I think it's important to revisit the DNA of Apple's staggeringly great run over the past decade. In fact, you can point to a similarity over the Company's lifetime. Apple tends to invent, what in hindsight were obvious solutions, which are so self-evident in their presentation as to be accepted immediately by tens of millions of people. As the solutions tend to 'invent' markets, they begin with near 100% share, which the force of gravity and well capitalized competitors continually erode. PC's are in the single digits and the phones have about a 1/3 share. The tablet, being relatively newer, still has share over 60%, but it's declining too.  Revenues are going up, as the markets are growing, but share is declining.

The company is far better at inventing new markets than protecting their position in existing ones. It's not that they are disinterested in doing so, but unlike what Microsoft pulled off 20 years ago, the key candidates for customer 'lock-in' are applications which are now browser, rather than operating system based.  Presenting a family of solutions such as iOS, Airplay, iCloud, iTunes etc helps to influence people to stay in the family; it's necessary, but not sufficient for lock-in. This is good for consumers, and I believe good for Apple. They have to keep innovating to not only stay ahead, but to support a premium position.

Market defining opportunities do not present themselves every year, let alone the time it takes to properly engineer the right solution, at an acceptable price. As a consequence, the company will have up and down performance, and will continue to lose customers in some segments, while gaining share in others. That's just a fact, and again, it's better for them to accept this than to bring out products before their time. After all, it took three CEO's and five years to recover from the Newton debacle.

An original thinker...Scott Belsky

Behance is a community of creative folk founded by Scott Belsky. He recently sold the company to Adobe and now leads their community efforts. He is an original thinker whose passionate about his product and his constituency. Recently, I met with him, and saw his presentation at NJTech Meet-up. Fortunately, here's a version of it given at Internet Week in NY.

Here's some snippets to whet your appetite:

• Meritocracy, innovation and access are not natural to the web
• The problem is the web is verticalizing. This limits creativity and inspiration. Behance leveraged that 95% of creatives follow other fields and would appreciate broad exposure
• Creative meritocracy is about credible mass vs critical mass. Not a crowd sourcing fan. Good to hear a dissenting voice
• I try to hire great people and view my role as to be the wind at their backs

Scott Belsky at NY Internet Week