The Next Internet Will Not Be a Page. It Will Be a Warehouse Full of Machines

The market will begin to see that the internet stopped being a network of sites and became an industry of physical scale.

The market is still trying to crawl out of the wreckage. Banks broke, credit evaporated, investors discovered that liquidity is not a permanent property of nature, and executives who used to speak of innovation began to speak of survival. In moments like this, collective imagination shrinks. After a crisis, almost everyone wants to buy what looks safe, cheap, known, accounting-based, auditable, domesticated. Wounded capital looks for ground. The problem is that the best asymmetries often appear exactly when capital is too busy looking for ground to notice the terrain is changing.

We are in March 2009. The internet has already won culturally. Everyone knows that. And because everyone knows that, almost no one is still asking what the internet is becoming. That is the trap. When a thesis becomes consensus, it changes shape. The big money is rarely in the phrase "the internet will grow." That phrase already belongs to the past. The useful question is different: what kind of infrastructure will be necessary when the internet stops being a place where people visit pages and becomes a permanent system of computation, commerce, media, storage, communication, security, software, and data at industrial scale?

The first internet was sold as experience. Portals, emails, blogs, search engines, social networks, online stores. The user saw pages. The investor saw traffic. The advertiser saw attention. But the second internet will be less visible. It will not have a pretty face. It will be measured in servers, switches, fiber, racks, square meters, electricity consumption, latency, cooling, bandwidth, redundancy, semiconductors, colocation contracts, interconnection, and the ability to move data without collapsing.

The first internet looked light because the end user did not see the weight. The second internet will make fortunes exactly because that weight does not disappear. It only changes owners.

The market's mistake will be to keep calling this "the internet." The word has become too small. The next cycle will not be internet. It will be infrastructure for computation at scale. The difference sounds semantic, but it is not. Internet is what the consumer perceives. Computation at scale is what allows thousands of companies to deliver software, video, search, payments, social networks, commerce, and services without owning every machine they use. Internet is surface. Scale is subsoil.

Nvidia and AMD enter this story as obvious companies, but not yet for the reason the crowd usually imagines. In 2009, most people will look at them through the lens of PCs, games, semiconductor cycles, graphics cards, notebooks, and traditional competition. That matters, but it is not enough. The larger point is that any economy that comes to depend on abundant computation will eventually pressure the entire processing stack. CPU, GPU, accelerators, memory, networking, storage. When demand for computation stops being episodic and becomes permanent, capacity suppliers stop being ordinary cyclicals and begin touching a deeper nerve of the digital economy.

But the most interesting place may not be in the most discussed chip. It may be in the path between one server and another.

The vulgar investor thinks of computation as if each machine worked alone. That image is old. The next internet will be a distributed machine. The value will not be only in processing. It will be in coordinating. In moving data. In reducing latency. In preventing congestion. In connecting thousands of servers as if they were parts of an organism. In making local failures unable to bring down entire systems. In allowing private clouds, public clouds, providers, corporate customers, and networks to meet with enough speed and security that no one notices the effort.

Arista may still be a discreet name for many. But the category it represents is enormous: networking for modern data centers. If computation at scale grows, the internal network of data centers stops being secondary plumbing and becomes the circulatory system. The machine is not only the server. The machine is the coordinated set of servers. And a coordinated set depends on communication. The market likes processors because processors look like brains. But brains die without circulation.

Perhaps in 2014, if a company like Arista comes to the public market, analysts will try to explain it as a switch maker. That description will be technically acceptable and strategically poor. Switch is the object. The thesis is different: if large-scale data centers are redesigned for cloud, software, low latency, automation, and massive machine-to-machine traffic, a new class of networking infrastructure will be necessary. In that case, Arista would not be selling only boxes. It would be selling the ability to make data centers programmable, scalable, and less fragile.

The investor must be careful with small words. "Switch" is a small word. "Router" is small. "Colocation" is small. "Data center" sounds like rented concrete with air conditioning. These are ugly, industrial, charmless words. And for that reason, they can carry profit. The market prefers words that allow dreams. But technical words, when they sit at the right bottleneck, allow margins.

Equinix represents another piece of this thesis. What looks like a data center business may, in practice, be a market of proximity. In digital infrastructure, distance is not an abstraction. Distance becomes latency. Latency becomes experience. Experience becomes revenue, churn, arbitrage, speed, risk, or loss. Companies do not merely want to put servers somewhere. They want to be close to networks, customers, partners, cloud providers, exchanges, carriers, suppliers, and ecosystems. A good data center is not just real estate. It is an address inside an economic topology.

This changes the lens. The ordinary real estate investor asks: what is the occupancy rate? The digital infrastructure investor asks: who needs to be connected to whom, with what latency, with what redundancy, and with what switching cost? Equinix should not be analyzed only as powered square meters. It should be analyzed as an interconnection marketplace. The difference is brutal. Empty real estate is cost. Dense interconnection is network. And a good network tends to improve as more participants enter.

Digital Realty enters through a complementary path. The economy of cloud, software, media, and data will require physical scale. Someone must build, finance, operate, and expand the places where computation lives. Digital romanticism always tries to sell the idea that bits do not occupy space. It is a lie. Bits need machines. Machines need energy. Energy needs contracts. Heat needs to leave. Security needs to enter. Land needs to be chosen. Capital needs to be raised. Capacity needs to be delivered before demand explodes. When the world pretends technology is immaterial, smart money buys the assets that prove otherwise.

Broadcom is the more subterranean piece. It does not need to be loved by the public to be necessary. Connectivity semiconductors, wired infrastructure, chips for networking, communication, and enterprise equipment may seem too far from the elegant narrative. But computation at scale does not live only on central processors. It lives on movement. Every data hop has cost, limitation, delay, consumption, heat, and hardware dependency. The whole stack has to improve. The investor who looks only at the company on the cover misses the suppliers charging tolls inside the machine.

Here is the great shift: the value of the internet will migrate from the site to the system. In previous years, it was reasonable to ask who would win in search, e-commerce, advertising, social networks, and software. Those questions remain important. But from now on, a deeper question grows beneath them: who provides the common infrastructure for all the winners? Who profits if there is more video, more SaaS, more cloud, more data, more mobile, more transactions, more security, more backup, more streaming, more messages, more scientific computing, more automation, more artificial intelligence in the future?

The investor should look for businesses that gain from rising complexity, not only from one specific application. Applications die. Complexity remains.

That is the difference between buying fashion and buying plumbing. Fashion must remain interesting. Plumbing only needs to remain necessary.

Perhaps the market wakes up to part of this sometime in 2014. Perhaps the public offering of a cloud networking company serves as the visible signal of a shift already underway. Not because one specific company explains everything, but because public markets sometimes turn technical phenomena into investable categories. Before that, the thesis lives scattered: cloud growth, data center expansion, more traffic, more software delivered through the internet, more video, more mobility, more companies outsourcing infrastructure. Then the market begins to say: "there is a category here."

When the market gives a category a name, it reduces the discount of ignorance. The best return is usually before that baptism.

The reader can profit in a few ways.

The first is to stop looking only for the next internet application. That is the obsession of the investor who wants to look modern. Every generation looks for the winning app. But winning apps often become obvious too early or die too early. Infrastructure that serves many apps may be less glamorous, but it carries an advantage: it does not need to know exactly which human behavior will win. It only needs to know there will be more data, more traffic, and more computation.

The second is to build a layered reading. On top, software and consumption. In the middle, cloud, platforms, operating systems, databases, security. Below, semiconductors, networking, data centers, energy, and interconnection. When everyone discusses the top layer, look for pressure in the layer below. Margin appears where demand arrives before capacity.

The third is to understand that digital infrastructure can have characteristics of local monopoly, network, and scale. Not every data center is the same. Not every network is the same. Not every chip supplier is replaceable. Not every network software layer is a commodity. The market loves to say hardware becomes commodity. Sometimes it does. But in complex systems, the right hardware, in the right place, with the right software and the right customer, becomes dependency.

The fourth is to watch capex signals. If internet, cloud, and software companies start spending more and more on infrastructure, someone is receiving that money. The crowd usually celebrates the growth of the company spending. The colder investor asks who is on the other side of the invoice.

That question is offensive to the romantic. But useful.

The romantic wants to buy vision. The operator wants to buy budget.

The counter-thesis must also be respected. Infrastructure is capital-intensive. Data centers can suffer from oversupply. Companies can build before real demand. Leverage can punish REITs and real estate operators. Energy can become expensive. Large customers can pressure prices. Networking can face brutal competition. Semiconductors can be cyclical. Nvidia and AMD can be dragged by PC, console, inventory, and compressed-margin cycles. Arista, if it emerges as a public company, can face giants with larger channels. Broadcom can depend on equipment cycles. Equinix and Digital Realty can look safe until the cost of capital changes.

Whoever ignores risk in infrastructure has not understood infrastructure. Heavy assets amplify both demand and error. Building too early is almost as bad as building too late. The difference between vision and recklessness is utilization.

But the central point remains. After the 2008 crisis, the world will not become smaller. It will become more digital, more interconnected, more dependent on remote systems, more addicted to availability, and less tolerant of slowness. The recession may delay investment, but it does not eliminate the direction. Crises clean excess. They do not revoke structural needs.

The corporate world learned to outsource factories. Now it will begin to outsource computation. First for cost. Then for flexibility. Then for speed. Then because keeping everything internal will be an expensive form of nostalgia.

When companies deliver software through the internet, they need infrastructure. When consumers watch video, they need infrastructure. When banks process transactions in milliseconds, they need infrastructure. When social networks store photos, they need infrastructure. When governments digitize services, they need infrastructure. When stores sell online, they need infrastructure. When startups scale too quickly to plan servers calmly, they need infrastructure. When the future demands artificial intelligence, it will need infrastructure at an even more absurd scale.

The greatest irony of the digital age is that it will be sold as lightness while requiring an increasingly heavy physical base.

The investor should distrust every thesis that treats technology as magic. Good technology does not eliminate reality. It reorganizes where reality charges. If the user receives instant software, someone paid for latency. If the application seems infinite, someone paid for capacity. If the cloud seems elastic, someone built slack beforehand. If data appears anywhere, someone connected places. If the internet seems invisible, it is because a great deal of visible infrastructure was hidden from the consumer.

The next cycle will not be about "getting on the internet." That phrase is already old. The next cycle will be about making the internet capable of bearing the economic weight everyone wants to put on it.

Application companies will compete for attention. Infrastructure companies will compete for necessity.

Attention changes mood.

Necessity changes more slowly.

Leo Bentier