These mini-data centers would be networked together into distributed computing clusters. At first glance, the idea seems backwards. Why scatter compute across suburbs instead of concentrating it in giant centralized facilities?

Simple. Data centers have been fighting the physics. That’s a fight no technology can win.

Copper carries information by pushing electrons through resistance; fiber carries information by guiding massless photons through glass. Electrons have mass, generate heat, and interfere with one another as signaling speeds rise. Light traveling through fiber encounters far less resistance and almost no electromagnetic interference, allowing vastly more data to move over far longer distances with dramatically lower energy loss.

Electrical loss rises with current squared times resistance. The harder you push the signal, the more the copper resists. An optical signal mostly suffers slow optical fading—roughly 0.2 dB per kilometer in modern long-haul fiber. At scale the difference becomes enormous. A single modern fiber strand can carry terabits per second of data over hundreds or even thousands of kilometers with relatively modest signal regeneration. The highest-speed copper interconnects inside AI systems are often limited to distances measured in feet rather than miles, which is why fiber became the nervous system of the internet.

Even inside the data center, even within a server rack, heck, even inside a graphics processing unit (GPU) cluster, this is a problem. Electrical resistance is why our data centers threaten to glow in the dark.

Share

At the macro level—the electrical grid itself—strike “problem” and insert “crisis.”

Modern AI systems consume astonishing amounts of electricity. Training large models already requires massive, dedicated facilities. Even inference — actually using AI — consumes far more compute than traditional software workloads.

The cost of transporting the electricity to a centralized data center is both the generating cost and the strain on the grid, the threat of brownouts or worse. That’s why the grassroots movement to stop building them is having so many successes.

Today’s AI buildout depends on concentrated compute, which requires concentrated cooling, concentrated power delivery, and concentrated capital expenditure.

Nvidia and Pulte propose a flip flop. Don’t ship electrical power to the information. Ship the information to the power.

The Span idea exploits an inefficiency that Pulte deals with every time it builds a home. Because having a bit too little power is much worse than having a bit too much, millions of homes have some unused electrical capacity. A distributed network of small AI nodes might draw on that spare capacity for local processing and then be networked into a virtual data center.

This is, after all, how AI operates now, and why Nvidia is the obvious partner. As Jensen Huang rightly insists, it was not the GPU alone that made AI possible. It was the technology for linking and synchronizing first hundreds, then thousands, now (at Tesla) half a million GPUs in parallel, technology created in Israel by Mellanox which NVDA acquired in 2020.

The inside-the-data center challenge is increasingly being met by optical interconnects from server to server, rack to rack, and even chip to chip, from companies like Coherent (NYSE: COHR) and Arista Networks (NYSE: ANET).

This is leading-edge technology. But fiber optical networks birthed the Internet going on 40 years ago. Pulte and Nvidia are just catching up to what the physics now demands.

The Span idea may never catch on. But something will, whether it’s wafer scale computation with its vastly reduced power requirements, or graphene connections with 1,000 times the conductivity of copper. The massive data center buildout cannot survive its looming conflict with physical reality.

Leave a comment

Comments are huge for author morale. Don’t be shy!!!!!

Thinking of this brought to mind Amazon’s history of confronting just such problems: using firm architecture to stop mistakes before they get out of hand.

The guiding principles are familiar:

• Flatten hierarchies to push responsibility and authority down to the people closest to the work.

• Design systems that make errors visible early.

But Amazon has long been conspicuous in acting on them, largely thanks to Jeff Bezos, who did not become the world’s third richest man by accident. (As we have noted elsewhere, Elon Musk, the world’s richest man, is also a master of learning by failure.)

One of Bezos’ passions, oft cited in his shareholder letters, is the “two-pizza rule:” teams should be small enough to be fed by two pizzas. And the team should own a process end to end.

Small teams make it clear who owns what. Each team is responsible for a specific service—say, the system that handles product search, or the one that processes payments. Ideally, that same team designs the service, builds it, runs it, and fixes it. When something breaks, the people who built the system should see problem first and expect to respond.

Share

Large organizations often blur responsibility. Work is handed from one group to another: design to engineering, engineering to operations, operations to support. Problems travel upward, and then sideways, before they come back down.

Amazon is huge, with roughly 1.5 million employees, full and part time. Estimates say it employs well over 100,000 engineers. But Amazon fights hard to shorten the reporting/decision loop and give ownership to the people who know the system.

“You build it, you run it” is an Amazon catch phrase. The people who create a piece of software are also responsible for keeping it working. If the service fails at 2 a.m., the alert goes to the same team that wrote the code.

There is no waiting for a report to move through layers of management. Amazon gives that team the authority—and the incentive—to fix it quickly. They can roll back a change, shut down a malfunctioning feature, or patch the system without waiting for a chain of approvals.

The “stop” decision—to halt a process before it does more damage—lives close to the point of action.

Amazon likes small systems as much as it likes small teams. Its software, the beating heart of the company, is divided into many small, independent services. Each service performs a specific task and communicates with other subsystems through clearly defined requests. A single, complex machine becomes a network of smaller machines, each with a clear job. If one breaks, the others can keep running.

Subscribe now

In reading about this, I finally learned what an “API” actually is, I mean beyond the clear-as-mud definition: “application programming interface.” In practice, that just means one system that asks another system for a specific piece of work and gets a defined answer back.

The goal is to limit the damage when something goes wrong. If a pricing service fails, it does not automatically take down the checkout system. If a recommendation engine behaves badly, it can be turned off without stopping orders from being processed.

In Amazon-speak this is the “Bezos API mandate.” We can just think of it as minimizing the blast radius of system blow ups.

A strength Bezos and Musk have in common is treating errors as inputs, not embarrassments. That’s one reason SpaceX launches more than 100 times more often than NASA, at less than 5% of the cost per flight. Some SpaceX rockets blow up. All SpaceX rockets are data points.

Amazon Web Services (AWS) also makes the most out of failure. When AWS experiences an outage that affects customers, it publishes a detailed account of what happened: what went wrong; how the problem spread; what was done to fix it; what changes are being made to prevent a repeat.

These are not, at least not primarily, PR exercises. They are serious pieces of analysis. The tone is technical. The goal is not to assign blame, or excuse failure, but to understand the chain of events and break it, turning error into learning.

Amazon is not populated by saints. Let’s assume that Amazonians have the normal human tendencies to duck and cover. Nevertheless, this system is designed not to hide failures, but to surface them quickly, analyze them, and incorporate the lessons into future design.

Share

Amazon’s practices are not unique, in part because it is a closely studied company. Small teams, shared responsibility, modular systems, post-incident reviews have become ever more common.

Amazon’s practices were not built with modern AI in mind. Many of them predate it. But AI, more than ever, is demanding organizations that do not depend on being right the first time. Far more valuable is an organization that can recognize when it is wrong—and respond before the error spreads.

Neither Bezos nor Musk followed the natural arc of large organizations. To build Amazon and SpaceX they had to kick against the pricks. In our time the most powerful national examples of building to make error correction a feature, not a bug, are Ukraine and Israel.

For Israel wars are many, and time and resources always scarce. In their universal military service Israelis learn to accept responsibility young and in small groups. They rapidly iterate trials and errors, learning to learn.

Ukraine, with its relatively small population, up against a larger nation with little regard for the lives of even its own soldiers its rapid iteration and improvement of drone technology (now even to mount infantry assaults) was a jiu jitsu move that may yet win the war.

Many years ago, a four-year-old nephew of mine, fresh from another victory of triceratops over tyrannosaurus, proclaimed (somewhat self-servingly) “the little one always wins.” Not quite. But sometimes.

Leave a comment

Not because they are exploiting you. Because they are positioned where most learning happens—and where that learning can be captured.

In a recent note, I argued that the center of competitive advantage has shifted. Learning no longer happens primarily inside the firm. It happens across the ecosystem—through customers, through usage, through iterative failure—and flows back to the company that sits at the point where those experiences must be interpreted.

Think of Nvidia. Its power does not reside only in its chips, but in the habits of millions of developers trained on CUDA (its programming platform), in universities, in research labs, in codebases that grow far beyond the firm itself.

Or Nova Ltd. Its semiconductor fab customers generate endless trial-and-error experiments at the limits of physics. But Nova captures what those experiments mean.

Or Camtek, where every customer failure in advanced packaging becomes a data point in Camtek’s growing library of how complex systems break.

Or Mobileye, where millions of miles driven in the unruly world produce the only dataset that ultimately matters: what happens on the road.

In each case, the pattern is the same. The companies that win are not necessarily doing more of the work. They are positioned at the bottlenecks though which the hardest problems must pass—and where solving those problems generates reusable knowledge.

If this is the new source of competitive advantage, where should we look for companies that gather it?

How a country beset by, nay, formed by constraints? Hint: every one of the four companies I just mentioned springs from this tiny place. Even Nvidia would not exist today without Mellanox, the Israeli networking company it acquired in 2020.

Share

Not because Israel has more talent, or more startups, or even better technology. It’s because of the type to tech Israeli companies tend to generate because Israel’s culture has been formed by the stark necessity of transforming constraint into learning.

A small country, with limited natural resources, surrounded by uncertainty, does not have the luxury of focusing on easy problems. It must solve the hardest ones—quickly, and often under pressure. That bias shows up everywhere: in military systems designed for real-time decision-making, in cybersecurity architectures built for adversarial environments, in autonomous systems trained on worst cases

Israel operates in environments where failure is not an option—and where learning therefor must be somewhere between very fast and now, now, now!

This culture carries over into its commercial ecosystem. Israeli companies tend to emerge not by building complete, self-contained systems, but by inserting themselves into global ones—at precisely the points where those systems struggle.

Mobileye does not attempt to build the entire car. It sits at the point where perception fails.

Nova does not run the fab. It sits at the point where the fab fails because signals are obscured.

Check Point Software Technologies does not own the network. It sits where the network is attacked.

Each is placed at some control point—places where the system must pass, and where solving the problem produces knowledge that compounds over time.

The result is that many Israeli companies that look, at first, like niche providers turn out to be indispensable.

Subscribe now

The country’s small domestic market reinforces this pattern. Israeli firms cannot rely on local scale. They are forced outward—into global markets, and often into the most demanding segments of those markets.

They do not compete by being larger. They compete by being necessary. They compete by sitting at the constraint that must be conquered. What’s more, the company must not only sit at the point of difficulty but ensure that solutions flow through its systems—and that the learning from those solutions can be captured and reused.

When that happens, the result is what we increasingly see across AI, semiconductors, and autonomous systems: the strongest companies learn from work they do not perform.

For investors in challenging arenas the question today is no longer simply: Which markets are growing? Or even which companies have the best technology?

The right question is “where are the unavoidable problems”—and whose systems do the solutions run through? Because that is where learning concentrates. And where learning concentrates, value follows.

Leave a comment

The story is that China got rich by stealing. It sounds right. It preserves our assumptions. If Communist China is succeeding, it must be cheating. Take away the cheating, and the whole thing falls apart.

The problem is—it’s not true. Yes, we have all read the rants about China stealing $200 billion, $600 billion, a trillion dollars a year in intellectual property.

It is nonsense on two counts.

First, the two aging studies everyone quotes to this end are both embarrassing fictions, as Richard demonstrated in Reason Magazine last year.

The second reason is even more important. IP theft—stealing the recipe—just does not get you very far. Blueprints can be copied. But the ability to build, to scale, to improve yields, to iterate under pressure—that comes from doing. It comes from the learning curve.

China didn’t steal its way to prosperity. It learned its way there. In the “Information Theory of Economics,” which we espouse, wealth is knowledge, information is surprise, money is time, and growth is learning. Summing up the learning are “learning curves,” in which prices drop by around 25% with every doubling of total output.

Share

That’s how enterprise works. The recipe is rarely the secret. The process is the point.

China plugged itself into the process. Under Deng Xiaoping, it opened up enough to allow capital, technology, and know-how to flow in and allowed entrepreneurs enough freedom to turn learning into wealth.

Two Chinas

Somewhat confusing is what partial liberalization created: two Chinas.

There is Party China: centralized control, surveillance, political bullying and murder, and not a small dose of paranoia. Party China put Hong Kong news tycoon Jimmy Lai in jail for 20 years for promoting democracy.

And there’s the Free China that makes things work: reasonably free markets, uniquely entrepreneurial venture capitalist mayors, local problem-solving, a gazillion engineers, and the factories and supply chains they built.

One controls. The other produces.

For years, many in the West assumed Free China would eventually transform Party China into something like a liberal democracy. Because that has not happened, the same folks now tend to take the opposite view: Party China is all powerful and fully in control of Free China.

That view is just as wrong. Sure Party China would like to utterly control Free China. Happily it cannot.

Xi’s constraint

By the time Xi Jinping came to power, China had become impossible to run from the top down. In a sense this has always been true. The closest the regime ever came to complete control was within its first 10 years—and at least 40 million peasants starved to death.

Mao shrugged off the 40 million. Xi can’t afford Mao’s indifference pattern because Free China has gotten too rich and Party China depends on that money. China cannot be a world power if it destroys the economy of Free China.

That Xi knows this became clear during Covid. China imposed sweeping lockdowns—an extraordinary display of state power. Then in the face of massive resistance, Xi did what Mao—and even Deng, who ordered the massacre in Tiananmen Square—never would have done—backed down.

Not because the Party lost political control. It didn’t. But because the wealth of Free China was imperiled. Supply chains broke, factories missed targets, food shortages hit neighborhoods, local enforcement began to look fragile.

The cost of pushing further became too high. Xi backed off. The great lesson: Party China is powerful and ruthless. But it is not unconstrained.

Subscribe now

Americans should be able to see this. We are used to sorting regimes from people. We do it with Iran. We do it with Cuba. We hate the governments, but we don’t hate the people.

Increasingly, however, Americans do hate China. We hate it for its success—all of which comes from Free China. In America, that success breeds fear, anger and, most dangerous of all, envy, the root of all evil, the most original sin.

Wars are rarely the product of dispassionate calculations. They happen when countries hate. Wars erupt when nations make rivals against whom they should strive to emulate into enemies they seek to destroy That does nicely as summation up why Israel has had to defend itself from annihilation since before the founding of the state.

Patriot Games

Leaders in both American parties make play patriot by whipping up American anger. Crucial to this anger is the pretense that there is only one China, Party China, which became rich and powerful by cheating. Refusing to acknowledge the role of freedom—and millions of Chinese entrepreneurs emulating America—U.S. political hacks feed the rage.

Party China rules. When it feels threatened, it lashes out. When Alibaba founder Jack Ma criticized the regime’s financial bureaucracy, Xi drove him into exile and nearly destroyed his company. Xi taught Ma a lesson: Free China makes the money; Party China makes the rules.

Still, Party China has learned some lessons as well. Without Free China, Party China would be left poor and weak. There is only so hard Party China can push.

For investors, there is also a lesson. China builds value. The Party decides how much of it you keep.

That doesn’t mean “never invest in China.” It does make it necessary to distinguish, when possible, enterprises Party China will likely ignore from those likely to provoke the Party’s rage.

Leave a comment

Within hours of the shoot down, one crew member was located and extracted. The second—injured, hiding in mountainous terrain, actively hunted—was found and brought out two days later in what officials described as one of the most complex rescue operations in modern U.S. military history.

The operation involved hundreds of personnel, special operations forces, and a large supporting air package operating deep inside Iran.

At one point, U.S. forces destroyed their own multi-million dollar aircraft on the ground rather than risk them falling into enemy hands.

That was not improvisation. It was not luck. And it was not just courage, though there was plenty of that on display.

It is something else, something much more expensive, much less visible and which very few nations in the world have, but the U.S. (and Israel) have in abundance.

Share

Edward Luttwak once drew a sharp distinction between militaries that can actually fight and those that merely “march in parades.” The difference is training.

That sounds almost trivial—until you see what it really means. Because training, in a modern military, is not a classroom exercise. It is not even a series of drills. It is a continuous, resource-intensive process of rehearsing reality as closely as possible.

Pilots fly constantly. Units train together, over and over again. Systems are tested, stressed, broken, repaired, and tested again.

It’s “peacetime” But fuel is burned. Engines wear out. Equipment degrades. Money disappears.

Consider just one line item. Flying a modern fighter like the F-16 Fighting Falcon costs roughly $20,000 to $25,000 per hour.

That is not the cost of war. It is the cost of practicing for it.

What the rescue in Iran revealed is that training works. To locate a downed airman in hostile territory, coordinate dozens of aircraft, insert and extract special operations forces under pressure, and do it fast enough to beat an adversary to the scene— is not something that can be assembled on the fly.

Share

It is the accumulated result of thousands of prior iterations.

Military power is not what you own. It is what you can execute—on demand, under stress, in real time. And execution depends on repetition.

Often when we think about military strength, we think about procurement—new aircraft, new missiles, new ships. Visible stuff Congressmen can count.

But the United States spends hundreds of billions of dollars each year on something harder to count: operations, maintenance and readiness. The category that includes training.

These are especially difficult to count because the dollars spent on training don’t produce new assets, but consume existing ones.

Training and readiness are the cost of making sure that when something goes wrong—when a pilot is down behind enemy lines, when a decision must be made in minutes rather than days—the system responds as if it has seen the situation before.

Because it has.

We tend to think of capability as a stock—something you possess. In practice, it is a flow.

It depends on how often you do something, how quickly you learn, how effectively you integrate that learning into the next iteration.

The rescue mission in Iran was a display of learning velocity—of a system that has paid, repeatedly and at great cost, to rehearse reality.

The same habits formed in elite military—repetition, responsibility under pressure, learning through doing—show up outside the formal structure of the armed forces. They shape how quickly people adapt, how systems respond, how institutions behave under stress. These are a big part of the reason veterans are so valuable.

Subscribe now

When nearly everyone in a society—as in Israel—has several years of intense military experience the capacity to act decisively becomes a defining attribute of the culture. Powerfully, in Israel much of the “training” is live in a country that has been at war for 80 years.

Parade militaries display. Real militaries rehearse—until the rehearsal becomes instinct. And that is what brings people home.

Leave a comment

It was “wires and switches.” In his book Microcosm, which I edited, he described the economics of the old Bell System. Almost since the invention of the telephone, the most expensive part of the network had been not the wires, but the switches—long dominated by human operators and then by clumsy mechanical devices.

Then, at Bell Labs, the transistor was invented, and everything flipped. Suddenly switches became the “defining abundance” of the system. Once to be conserved, now the path of wisdom was to “waste” the switches and conserve the wires. So abundant were the switches that they would become computers—infinitely more capable than the human operators they replaced.

Then another flip. Wires became the scarcity—until copper was replaced by optical fiber, vastly expanding bandwidth, which became a new defining abundance and gave us the Internet.

As the Internet and the computer together multiplied data, the new scarcity became storage. George christened it “storewidth”: not the absolute amount of data that could be stored but the readiness by which it could be accessed.

When one scarcity becomes abundance, a new scarcity emerges.

Share

When switching became cheap, we placed more calls. When compute became cheap, we wrote more software. When storage became cheap, we kept everything. When bandwidth became cheap, we filled it with video.

This is the tale of economic growth. We spend freely what is abundant—and by doing so, we expose what has become scarce.

“Wires and switches” was the price system at work. Not in equilibrium, but in motion. Technology makes one thing cheap. We use it freely. And in doing so, we discover—through rising costs, delays, and failures—what is still scarce.

The new scarcity becomes the realm of potential profit—until it too is transformed into an abundance. The map of scarcity is not fixed. It is continuously redrawn. And the key to the map is the price system.

Share

Today the map is being redrawn by artificial intelligence.

For years, compute was the obvious constraint. Training large models required enormous capital, specialized hardware, and time. That constraint is, if not vanishing, loosening under the combined force of scale, better architectures, and relentless iteration.

Now the question is: what becomes scarce because AI becomes abundant?

There is more than one possibility. As models proliferate, high-quality data becomes more valuable. As outputs multiply, verification and trust become binding constraints. As capabilities expand, integration into real workflows becomes the hard part. And as inference scales, memory bandwidth, power, and latency begin to matter as much as, or more than, raw compute.

Consider NVIDIA (NASDAQ: NVDA). It is widely seen as the prime beneficiary of AI because it sells computation. That is true as far as it goes. But as GPU performance explodes, the binding constraint becomes moving data fast enough to keep the compute busy. High-bandwidth memory, interconnects, and system architecture begin to dominate.

The value is already shifting—from chips to systems, from FLOPS to data movement, from hardware to the full stack. NVIDIA is winning not because it sells abundant computing power, but because it is positioned at the new scarcity: coordinating that abundance.

If intelligence becomes abundant—if generating answers, writing code, producing analysis becomes cheap—then the scarce resource will not be intelligence at all, but what intelligence cannot replace.

AI can generate recommendations. It cannot bear consequences. It cannot take responsibility. It cannot be accountable.

In a world of abundant intelligence, the scarce resource becomes the willingness to act—and to be accountable for that action.

Subscribe now

Who signs off?
Who takes the risk?
Who owns the decision?

Who accepts responsibility?

A system—or a nation--society that diffuses responsibility—into systems, committees, or machines—will struggle in a world of abundant intelligence.

Systems that reward responsibility—where decisions are made close to the problem and owned by those who make them—will have an advantage.

AI will not devalue humans. It will make those willing to accept responsibility more valuable than ever. That’s how to survive and prosper with AI.

Leave a comment

If you want to see this pattern early, watch Israel. In a country where constraints are fierce—limited manpower, constant security pressure, and the need to make systems work right now under stress—companies rarely have the luxury of going modular— assembling standardized, interchangeable components assembled into a system. They integrate because they need acceptable performance from day one.

Nvidia (NASDAQ: NVDA), the world’s most valuable company, is still often described as a chip maker because its GPUs (graphical processing units) are the indispensable tools of artificial intelligence. But NVDA’s dominance comes from integrating those GPUs with the rest of the system: high-speed interconnects (NVLink, InfiniBand) linking thousands of chips, CUDA (its parallel computing platform and programming model), and entire libraries of optimized software.

Share

It is not incidental that Nvidia’s control over high-speed interconnects—now essential to scaling AI—came through its acquisition of Mellanox, an Israeli company that solved precisely the “between-the-chips” problem that now defines the bottleneck.

Apple (NASDAQ: AAPL) did not integrate hardware and software out of aesthetic preference. It did so because power efficiency and performance per watt became binding constraints. The solution was to control the entire system—silicon, operating system, and device design together.

Google (NASDAQ: GOOG) moved early. It began as a search program running on PCs, which quickly proved inadequate. Rather than buying enterprise servers, Google assembled its own cheaper machines—substandard individually but optimized for the distributed system it was building. Over time, it pushed further: designing its own data centers and eventually its own silicon, its TPUs (tensor processing units).

At their founding, none of today’s “hyperscalers”—Microsoft (NASDAQ: MSFT), Amazon (NASDAQ: AMZN), or the rest—were hardware companies. Now they all design chips.

The great Clayton Christensen, in The Innovator’s Dilemma (1997) and The Innovator’s Solution (2002), argued that companies integrate when products underperform and then shift to modularity when performance overshoots. Early systems—automobiles, computers, consumer electronics—are built as unified wholes because the margin for error is small. Once performance improves, components standardize, interfaces stabilize, and competition shifts to individual layers. Systems become cheaper, more flexible, and easier to scale.

The companies behind them follow the same path: integrated at first, then increasingly specialized. Why, then, do today’s leaders appear to be moving backward—from modularity to integration?

Christensen Cycles

What we are seeing is not a reversal of Christensen’s insight, but its continuation at a higher level. In AI and large-scale computing, performance is once again not “good enough.”

The components are extraordinary. GPUs are vastly more powerful. Memory bandwidth keeps expanding. Networking has advanced dramatically.

Yet success has created new constraints. The bottleneck is no longer inside the chip. It is between the chips: latency between nodes, synchronization across thousands of processors, power delivery and cooling limits, and software coordination across distributed systems. These are not component problems. They are system problems.

System problems demand integration. Christensen’s rule still holds—but now it repeats. Industries go modular when performance is sufficient. They reintegrate when new constraints again make performance insufficient.

Integration gives way to modularity. Modularity succeeds so well that it creates new complexity. Interfaces become choke points. Coordination costs multiply. Latency accumulates. The bottleneck shifts upward—and integration returns.

Share

At each turn, advantage accrues to those who can collapse layers, eliminate interfaces, and optimize the system as a whole. That’s what made the Magnificent Seven magnificent.

Investors Take Note

In a modular world, value disperses. Specialized firms compete within layers. Margins compress. Ecosystems flourish.

In an integrated world, value concentrates. Boundaries between components—indeed between hardware and software—blur. What was open becomes tightly orchestrated.

Christensen described what happens when systems become easy. We are now watching what happens when they become hard again.

A few years ago, it was fashionable to complain that most of the NASDAQ’s value was concentrated in a handful of companies. Investors, we were told, were irrationally chasing headlines.

In this case, investors had it right. Warren Buffett defined a franchise as a business with no close substitutes and durable pricing power. As systems grow more complex the companies that can integrate them begin to look like franchises.

Subscribe now

Franchise stocks may seem expensive—until competitors start crashing themselves to bits against their walls.

This is why Israel keeps appearing at critical control points in global technology systems. In a world returning to integration, advantage goes to those who learn fastest under constraint and can make systems work before their components are ready for prime time.

Leave a comment

During a meeting with members of the New York City Democratic Socialists of America, she stated: “I have not once ever voted to authorize funding to Israel, and I will never… The Israeli government should be able to finance their own weapons if they seek to arm themselves.”

Pressed on whether that included defensive systems, she answered: “Yes.”
(Hat tip to Ed Morrissey at Hot Air for surfacing the exchange.)

This is not a serious position. It’s not even a bad policy argument. It’s a category error, resting on the childish premise that U.S. “aid” to Israel is charity.

Iron Dome, commonly described as an American-funded Israeli system, it is in fact a joint program. It was developed by Israel-based Rafael Advanced Defense Systems in partnership with America’s Raytheon Technologies and produced with components and manufacturing spread across both countries.

The result? Iron Dome is not just a defense system. It is a learning machine—for the U.S. Every interception produces data—on detection, tracking, timing, coordination under stress. That data flows through a shared network of engineers, contractors, and military planners. Many of them are American.

Share

This is not aid. It is co-development under fire. And that’s crucial because modern war is not decided by who has the best system today, but who can iterate fastest. Missiles improve. Countermeasures follow. Cheap drones proliferate. Defenses adapt. The advantage goes not to the strongest player, but to the fastest learner.

Israel is not just an ally. It is a high-frequency combat laboratory—a place where systems are deployed, stressed, broken, fixed, and redeployed in cycles the United States cannot replicate at home (and should be grateful it cannot).

Cut off programs like Iron Dome, and you are not “ending aid.” You are unplugging from one of the most valuable real-world learning systems available to the United States. No simulation, war game, or test range can replace it.

We “aid” Israel because our two countries are engaged in a joint venture in survival—one that produces technology, data, and operational insight that flow directly back into American systems.

Subscribe now

The United States is now exploring its own layered missile defense architecture—branded, in typical Trump fashion, as “Golden Dome.”

From where do you suppose the concepts, the data, the operational instincts for such a system come? Not from committee rooms or defense white papers.

They come from systems that have already been built, tested, broken, and rebuilt under real conditions—mostly in Israel.

That’s the “aid” AOC and antisemites left and right now propose to cut off.

Leave a comment

The comparison to Amazon is not casual.

The platform was designed by Amazon Web Services and the Army Enterprise Cloud Management Agency. For now, interested vendors can respond to a “commercial solutions opening,” get Army approval for their product, and “start selling.” Soon, even more red tape will be stripped away—not by bureaucrats, but by the market itself.

“Once the Army approves a product and soldiers or other users put it to use, they are able to rate the product on a scale of one to five stars, just like on Amazon. That feedback gets back to the vendor almost immediately…”

Currently the marketplace offers only about 30 systems, all derived from existing Army programs. But the plan is to expand quickly including to drones developed by soldiers themselves, such as those built in the 101st Airborne Division and the 25th Infantry Division.

Hooray for efficiency and speed in procurement. But something much bigger is going on here.

The Army is attempting to address the central problem of modern warfare: How do you learn faster than your enemy?

Share

For decades, the U.S. military contracted for weapons systems that assumed long development cycles and relatively stable technologies. There were real successes. The B-52 bomber, now more than 70 years old, is still flying. The Army would love to retire the A-10 “Warthog,” which just celebrated its 50th birthday, but it remains too effective to let go.

Yet that model has been shattered, most visibly by drones, but more fundamentally by underlying shifts in technology.

Drones are not just cheap aircraft. They are cheap experiments. They can be built quickly, deployed quickly, lost cheaply—with no loss of pilots—and redesigned immediately. Materials are inexpensive. Software iterates fast. Tactics evolve in real time.

What matters is not the drone. It is the cycle: Build. Deploy. Learn. Adapt. Repeat.

The advantage does not go to the side with the best drone, today. It goes to the side that can run the most cycles.

The Army’s version of Amazon is not a catalog, but, like Amazon itself, a platform for iterative learning.

Amazon may look like an exercise in supercharged logistics: vast automated warehouses, overnight and increasingly same-day delivery. Impressive. But Amazon’s biggest impact was not on delivery but on the universe of sellers.

Millions of small businesses were born—and tens of thousands grew large—because Amazon made it easy for them to sell—to iind buyers and learn from them.

Many advocates for free markets explain their advantages in terms of competition and efficiency. But competition does not drive innovation. What drives innovation is the ease of buyer meeting seller.

Subscribe now

Markets do not merely allocate resources. They create possibilities. Create possibilities, and they will be met. Identify buyers, and soon sellers—bearing new products--will appear. Create new products and buyers will appear—if the creators can reach them.

By simplifying storage and accelerating delivery, Amazon increased the population of entrepreneurs—and thereby accelerated innovation. The Pentagon’s own Amazon could do the same.

If that happens then, once again, our military will be catching up to Israel.

Israel’s military has never had the luxury of slow development cycles. Threats evolve quickly. Constraints are constant. Resources—especially early on—were limited. The result is a system in which small units are empowered, feedback loops are tight, and lessons move quickly from field to design to deployment. The boundary between builder and user is thin.

Israel does not really have a “procurement system.” It has a learning system.

Markets are not primarily arenas of competition. They are arenas of learning.

And in war, Israel is the fastest learner—and teacher—of all.

Leave a comment

A guest post from Dr. Robert Castellano, my colleague at George Gilder's Technology Report

Israel has developed one of the world’s most advanced missile technology industries for both missile defense and precision strikes. A nation under attack for almost 80 years has yielded an ecosystem in which missile systems are designed, tested, and refined under real operational conditions.

Supplying Israel’s needs is a highly concentrated and technologically sophisticated supply chain. A small number of major defense companies integrate complete missile systems, while a broader network of electronics, radar, sensor, and propulsion suppliers provides the underlying technologies that enable those systems.

Many of these technologies have evolved into export products. Israeli companies now supply missile systems, radar technologies, and defense electronics to countries across Europe, Asia, and the Middle East. Israel’s missile industry has become not only a national security asset, but also a globally competitive, high-technology sector.

Israel’s missile ecosystem is dominated by three major defense companies that serve as the system integrators, designing missile systems and coordinating the integration of propulsion, radar, electronics, and structural components produced by suppliers throughout the defense industrial base. (See Table 1).

Israel Aerospace Industries and Rafael Advanced Defense Systems are the country’s principal missile developers, producing interceptor systems, naval missiles, and precision strike weapons. Elbit Systems plays a complementary role by supplying advanced electronics, sensors, and avionics technologies used within missile guidance and targeting systems.

In the United States, defense programs are politically distributed across many contractors. And we end up with $13 billion aircraft carriers like the USS Gerald Ford, delivered years late, and pulled from active duty off Iran because of fires in the laundry room, backed up sewage systems, and non-functional launch catapults.

In Israel, the concentration of expertise within a few companies allows radar engineers, electronics specialists, and missile designers to work closely together, moving from concept to operational deployment relatively quickly.

Advanced electronics and electro-optical sensors from Elbit Systems (NASDAQ: ESLT) are at the foundation of Israel’s missile systems, detecting targets, tracking their movement, and guiding the weapon to its objective. These systems capture real-time imagery of a target, and use onboard processors to maintain a lock even when the target changes direction or attempts to evade detection. These systems require advanced semiconductor devices, high-speed signal processing electronics, and precision optical assemblies, technologies that overlap with the broader semiconductor and electronics industries.

Radar technology is also critical to detect incoming threats, calculate their trajectories, and provide targeting information that enables interceptor missiles to engage those threats. A major supplier of these technologies is Elta Systems, a subsidiary of Israel Aerospace Industries, that develops advanced radar and electronic warfare systems used in air defense networks. Elta’s EL/M-2084 radar system, used in Iron Dome, detects incoming rockets and missiles, calculates their flight trajectories, and determines whether the projectile will strike a populated area. If interception is required, the radar transmits targeting data to the interceptor missile system.

These radar systems must process large volumes of data in real time, while tracking numerous airborne objects simultaneously. That requires advanced RF electronics, powerful signal-processing hardware, and sophisticated tracking algorithms.

Rafael Advanced Defense Systems
One of the most influential companies in Israel’s missile ecosystem is Rafael Advanced Defense Systems, which produces several of the country’s most widely deployed missile systems. Iron Dome was designed to intercept short-range rockets and artillery shells, while David’s Sling provides protection against longer-range ballistic threats. Both systems combine advanced radar tracking with interceptor missiles capable of engaging high-speed targets.

Rafael also produces the Spike family of anti-tank missiles, deployed on ground vehicles, helicopters, and naval platforms. Spike has become one of the most widely exported precision missile systems in the world, used in most NATO countries and being evaluated by the United States.

Israel Aerospace Industries
Israel Aerospace Industries is central to Israel’s long-range missile defense and naval missile programs. The Arrow interceptor system was developed jointly with the United States and is designed to intercept ballistic missiles at high altitude. The Barak missile system provides naval air defense for ships and coastal installations, while the Gabriel missile serves as an anti-ship strike weapon. These systems integrate propulsion technologies, radar tracking, and advanced guidance electronics to engage high-speed threats at long distances.

Israel’s missile industry differs from larger defense ecosystems in several important ways. The industry is highly concentrated, with a small number of companies responsible for developing complete missile systems while working closely with specialized suppliers that provide sensors, electronics, and radar technologies.

Israeli missile technologies benefit from continuous operational testing. Iron Dome and David’s Sling have been deployed repeatedly under real combat conditions, allowing engineers to refine performance.

Share

Israel has developed strong export markets for many of its missile systems, providing Israeli companies with international revenue streams that support ongoing research and development.

Investor takeaway
The concentration of Israel’s missile industry, dominated by a small number of companies, many of which remain government owned, limits direct equity opportunities in Israel’s missile industry. Among the companies discussed in this article, Elbit Systems is the primary publicly traded company. Israel Aerospace Industries and Rafael Advanced Defense Systems remain government owned. Elta Systems is a subsidiary of IAI.

Still, the broader Israeli missile ecosystem demonstrates how innovation in radar systems, precision guidance, and missile interception technologies can generate export demand across global defense markets. Israeli missile systems such as Iron Dome, David’s Sling, and the Spike missile family have been adopted by numerous countries, expanding the international footprint of Israel’s defense technology sector.

The Israeli missile ecosystem also illustrates how modern defense systems increasingly depend on advanced electronics and semiconductor technologies. As missile systems become more precise and capable of operating in electronically contested environments, the importance of these technologies continues to grow.

Leave a comment

The system is designed to intercept aerial threats using directed energy, potentially at a fraction of the cost per shot of traditional interceptors.

Company President and CEO Bezhalel Machlis disclosed the deal, inked in late 2025 on Tuesday during discussion of the company’s 2025 year-end results

Machlis described the laser solution as economical in an era of asymmetric warfare, meaning that rather than spending huge amounts on missile air defenses to stop cheap drones, one can spend ‘cents of electricity’ to down threats.

‘The advantage of the aerial laser is that it is less affected by humidity, rain, dust, atmospheric conditions the higher you go,’ he said. It can operate above the clouds, for instance. It can also strike threats before they arrive because it can see them from the air.

I’ve been following laser weaponry since the 1980s. It’s hard to exaggerate the difficulty of doing what Elbit proposes. The biggest problem is probably finding and staying on a target from a moving vibrating platform. It was the challenge of targeting that made the prospect of laser weapons seem so daunting during the Reagan administration. Add to that the challenges of powering up an energy weapon in the confined space and weight of an airplane and this development becomes almost miraculous.

As Machlis pointed out “You need to miniaturize the elements,” Machlis said. “While moving, you need to lock yourself on a target and in a very precise way.”

Amazingly he said Elbit has been able to “overcome all these challenges,” and after “advancing with large investment, they will be operational with the air force…”

God bless Israel, again.

George conceived it as a test of personal character, of whether one was inclined toward envy or gratitude. It is still that, but events have made it into something more, a test of whether you still believe the principles that made the modern West possible. Israel is a test above all of whether you believe in America.

Zionism is the belief in the Jewish story. And to believe in that story is to believe in the possibility of America.

The Haggadah is the Exodus story as retold every Passover. As Robert Goldberg points out in his new book, The Haggadah, it is not a story of mere memory, a lament of the victims. It is an imperative to action.

Zionism is the Exodus made political—the ancient demand for freedom translated into sovereignty in the modern world. A new chapter to that story was opened in 1948 and is still being written. For it is one thing to tell a story about liberation. It is another to build a society that sustains it. Freedom, once achieved, must be governed, defended, made productive. It must survive contact with reality.

Subscribe now

Israel is one of the very few places in the modern world where this entire chain—from memory to sovereignty to survival—has been attempted, continuously, and under conditions of extreme constraint.

Consider what Israel is, stripped of rhetoric. It is a country with no margin for error. No strategic depth. No ability to lose a war and recover at leisure. It cannot afford the luxuries of confusion that larger nations indulge. In Israel, the distance between decision and consequence is short; reality cannot be avoided.

Under those conditions, self-pity is unaffordable. Dependency is fatal. Failure must be converted into learning, quickly. “Responsibility” is not a slogan; it means no one else is coming, and excuses do not change outcomes.

Israelis do not whine. They fight. They improvise. They adapt.

This is how Israel became America’s most effective ally.

It builds missile defenses that work. It develops intelligence that delivers. It produces technologies that embed themselves deep inside the systems of the global economy, as we have called them “platform” technologies—often unseen, but indispensable.

This tiny desert nation can do these things because of a particular relationship to reality: a willingness to learn, to iterate, and to own outcomes.

Share

Success of this kind does not go unnoticed. It provokes a reaction.

Some see in Israel a model: a small nation, under pressure, that creates more than it consumes, that converts adversity into innovation, that sustains freedom not as an abstraction but as a discipline. We admire it. We study it. We ask what we can learn.

Others see manipulation, conspiracy, and unfair advantage—never the discipline that produced the result. They explain success not as creation but as exploitation. They do not ask how it works; they ask how it cheats. Achievement is recast as illegitimacy. Contribution becomes control.

The enviers fail not only the Israel Test but also the America Test.

The United States was built on a similar premise. We have our own Haggadah, our own imperative, still unfolding story. It says that free people, accepting responsibility for their own fate, can create a society that is both prosperous and just. That excellence is not a threat but a source of general enrichment. That truth, however uncomfortable, is preferable to comforting illusion.

With these premises now challenged by cynics and cowards, ingrates and traitors, Israel is not merely an ally. It is a mirror. If Israel is right—if a small, exposed nation can sustain freedom through responsibility, learning, and creation—then the Western project remains viable.

Subscribe now

Israel has become the most divisive issue in international life because it forces a choice.

It asks, relentlessly, do you still believe in the things that made your own society possible? Or have you come to prefer explanations that excuse failure, resentment that replaces aspiration, and narratives that dissolve reality?

That was the Israel Test. Today it is the America Test.

Leave a comment

Unlik…

Read more

For the past several years, commentators have marveled at a new asymmetry in warfare.

A drone costing a few thousand dollars can destroy a tank worth millions of dollars. A loitering munition assembled from commercial parts can disable equipment worth a thousand times its cost.

For centuries, military advantage flowed from industrial m…

Read more

71 days.

That’s how long it took Venom, a powerful, new, U.S. autonomous strike aircraft (aka a drone) to move from concept to flying prototype.

71 days. In traditional aerospace terms, that is almost indecent.

For decades, the path from concept to first flight has stretched across years. Requirements proliferated with each committee’s …

Read more

The conventional story portrays the Inquisition as a religious crusade against insincere converts.…

Read more

The “Dirty Jew” Super Bowl ad badly missed the mark by analogizing antisemitism to anti-Black racism. The ad, well intentioned, was fatally confused.

Antisemitism is different from racism. It is spread and maintained primarily via factual but false allegations. It blames Jew not for who they are, but what they allegedly do.

Yes, antisemitism h…

Read more

Read more

Read more