This Dial‑Up Era of AI Is Just the Beginning — From Infrastructure to Implementation

The screech of a dial-up modem and the hum of an NVIDIA Blackwell GPU are a quarter-century apart—but the historical echo is unmistakable: today’s infrastructure frenzy mirrors the early internet boom, laying foundations for transformative technologies still yet to emerge.

Back in 1999, as a young City and Wall Street investment banker at Credit Suisse First Boston, I experienced firsthand the buzz generated by the Freeserve IPO, an Internet Service Provider (ISP), which launched with little more than a “free online access” pitch—and yet floated at roughly £1.5 billion. Investors weren’t buying software; they were buying the handshake that got people online.

A few months later, Cisco briefly overtook GE as the most valuable public company on Earth—all because it sold the routers powering the early internet. Both milestones came years before everyday internet use looked anything like it does today.

The new tech stack

Fast-forward to July 2025 and the chorus is familiar, but involves more capital and pace. NVIDIA just smashed through a $4 trillion market cap. TSMC controls over 90 percent of the packaging capacity that every frontier model depends on. AWS is clearing $30 billion a quarter in cloud sales as enterprises shovel compute in, mostly ahead of knowing what to build with it. Once again, the market is cheering the laying of the pipes—the construction of the skyscrapers defining our future economic landscape has barely begun.

The Internet Stack – From ISP to SaaS

Layer 1: Internet Access & Infrastructure (1995–2000)

Back then, the product was connectivity itself. Freeserve signed up a million UK users in months, proving that “free” access could still be profitable. Across the Atlantic, AOL had 20 million subscribers by 2000—the scale that fueled history’s largest merger at the time. Meanwhile, Cisco’s routers became the shovels of a broadband gold rush, peaking at a $500 billion valuation just as the dot-com bubble crested. The scarcity wasn’t content; it was the ability to just get online.

Layer 2: Portals & Early Websites (1997–2001) 

With user growth exploding, bandwidth no longer guaranteed returns. Instead, attention became scarce. Yahoo! soared to nearly $475 per share by January 2000, becoming the place for every new net user.

Web 1.0 saw the shift from a 'read-only' web to the beginnings of e-commerce as the likes of Amazon and eBay—still glorified mail-order shops with shopping carts—rode the same wave. Logistics and payment solutions followed. At the end of 1997, Amazon was valued at $1.5 billion, whereas it is now valued at $2.3 trillion (151,900% growth). Each new layer, application and use case added more value.

Layer 3: Social & User-Generated Explosion (2003–2012) 

Then came platforms where users built each other’s value—the centralised Web 2. MySpace, Friendster, and ultimately Facebook created sticky networks. Facebook’s $100 billion IPO in 2012 overshadowed companies that once felt “unassailable.” Meta is now valued at $1.9 trillion.

Networks, not pipes or platforms, generated enormous value.

Layer 4: Cloud, Mobile & SaaS (2006–2015) 

My own businesses were involved in the adoption of both cloud and smartphones. The smartphone put the internet in pockets. Cloud made infrastructure globally programmable via APIs and swapped on-premise compute and storage ownership and management to an outsourced, subscription model. AWS (a division of Amazon), launched in 2006, quietly achieved $30 billion-per-quarter scale long before anyone asked why it wasn’t shipping CDs. Salesforce proved SaaS sales could hit $5 billion annually while customers just clicked “subscribe.”

Cloud and 'apps' attracted billions of venture capital investment into businesses that were UIs with a database that could scale with demand.

Layer 5: Decentralisation (2016– ) 

Just as Web 2.0 transitioned power from static portals to social platforms, Web3 now aims to shift power and ownership from centralised tech giants back into the hands of users. Driven by blockchain and decentralised technologies, this layer is re-architecting the internet’s foundational structures. If previous layers of the internet stack were about connecting, interacting, and sharing content, decentralisation is about owning and governing it—potentially reshaping value flows and power dynamics just as profoundly as earlier shifts did.

What the Internet Stack Teaches Us Once the layer below commoditised, economic gravity pulled profit one tier upward. Dial-up gave way to portals; portals flattened into platforms; platforms were eclipsed by SaaS. That climb wasn’t a choice—it was technological progress pulled by economic gravity. We now refer to each layer as a wrapper of the one below.

The AI Stack – From GPU to Value (2023– )

Layer 1: Silicon and Supply Chains 

Today’s AI frenzy around chips and GPUs has echoes of the dot-com boom—Cisco briefly became the world's most valuable company simply by selling the routers powering the early internet.

Right now, GPUs from NVIDIA and AMD dominate headlines, powering trillion-dollar valuations purely on infrastructure-level hype. Behind them sits an essential manufacturing layer: chip foundries like TSMC. Yet this excitement is just infrastructure noise—like cheering the modem rather than the website.

The real AI wealth will flow further up the stack, to companies and applications we haven't even imagined yet.

Layer 2: Cloud AI Compute 

If GPUs are today’s equivalent of Cisco’s routers, the hyperscale cloud providers—Microsoft Azure, Amazon AWS, and Google Cloud—are playing the role of the early ISPs. They're the new gatekeepers of compute, racing to deploy massive GPU-powered data centres that deliver raw AI horsepower on demand. AWS alone posted a record quarter at nearly $31 billion, powered largely by enterprises scrambling to secure cloud compute before they fully understand what they'll build with it.

The real economic engine is still waiting higher up the stack, where integrated, orchestrated AI applications will change the way individuals and organisations work and compete.

Layer 3: Foundational Models 

Foundational AI models—like those built by OpenAI and Anthropic—are akin to early browsers, providing access to intelligence in the race to dominate the AI future. AI, however, evolves exponentially, and non-human intelligence will transform everything at a pace beyond the evolution of the internet stack.

OpenAI's massive $8.3 billion raise at a $300 billion valuation, and Anthropic's pursuit of $170 billion, demonstrate intense excitement and demand for groundbreaking technology. But intelligence itself, and access to knowledge, will trend quickly towards commodity pricing. Like Netscape and Internet Explorer, these foundational models are vital gateways—but unlikely to remain lasting monopolies.

Recognising this, these labs are launching personal and corporate AI agents, exploring AI-native hardware, and even spinning up multi-billion-dollar consulting arms to help organisations implement the tech. They know enduring value lies in embedding intelligence into real-world workflows and solutions.

Layer 4: Tools for Builders - Fastest Growing Businesses in History 

Platforms like GitHub Copilot, Cursor, and Lovable are turning AI into rapidly scaling products. GitHub Copilot already has over 20 million users and is embedded in 90% of Fortune 100 engineering teams. Cursor exploded to $500 million ARR in under 18 months, while Lovable reached $100 million ARR in just eight months.

Today they're primarily serving technical users, but they've only scratched the surface of AI’s broader potential across the entire workforce.

Layer 5: Orchestration and Implementation 

This is where AI’s complexity collides with business reality. Gartner forecasts 15 percent of day-to-day decisions will be handled by agentic AI by 2028, yet 61 percent of enterprises piloting agents struggle with siloed bots. Implementation platforms build bridges here, ensuring policy, compliance, and workflow integration. This is where immense operational value is created.

Layer 6: Everyday AI and Robotics - Still Unwritten 

AI’s most transformative products—essential to daily life and business—are yet to be built, achieve scale, or even launch. Just as the internet’s true power emerged through apps and social networks, AI’s ultimate impact will embed seamlessly into cognitive tasks and physical labour. Tomorrow’s trillion-dollar companies are currently sketched in slide decks. The real AI revolution will unfold quietly, embedding intelligence invisibly into every aspect of life and business.

Why the Upper Stacks Win

McKinsey estimates generative AI could add $2.6–$4.4 trillion annually across 60+ applications—mostly unlocked in customer operations, marketing, engineering, and R&D. That’s where tangible ROI lives—not in rented GPUs or billion‑parameter LLMs. 

Scarcity is shifting: compute will get cheaper, models will commoditize, but integration—that stitch of tools into trusted workflows—gets tougher every quarter.

What to Do Now

If you believe the market ends at GPUs, think again. The rails are being built—but the wealth will flow through platforms that make outcomes first-class. Whether you’re building today’s tools or integrating tomorrow’s systems, there’s massive opportunity above the compute crease.

If you’ve seen AI ship real ROI—not just prototypes—comment below with what made it stick. 

What layer are you focused on? Are you looking ahead, and upwards?

Follow me on LinkedIn for more.

Thanks for reading.