The World Runs on Compute
High productivity is when entrepreneurs successfully apply compute to business problems. Concentrated power is the consequence.
Firms cluster around productivity. When choosing who to work with, highly productive firms tend to choose partners with similar efficiency. Like many things in economics, this seems like common sense. But it’s not that easy. In a recent paper, University of Pennsylvania professor Jesus Fernandez-Villaverde makes the case that firms organize around productivity. What seems obvious turns hairy when you actually try to quantify and put some tangible evidence behind the thesis. Villaverde does a good job at that. He goes even further and predicts that firms will break up their relationship if one of them drops in productivity. Take two firms, firms A and B, with similar productivity. A works with B. Now let’s assume B’s productivity drops. Villaverde predicts that this relationship will break and proves the point by running statistics on thousands of examples from the recent past.
High productivity dominates
This might be just one of those cases where Villaverde’s paper is much more profound than the authors actually thought. It’s not that productive firms "like" to work with similar firms. They "must" do so. Industries with dispersed productivity in their value chains don’t work. Investors should stay away from such sectors and focus on situations where productivity is uniform across the value chain. Eventually, if you think this through, the conclusion is that highly productive firms must vertically integrate and create their own value chain. In other words, if there is one firm in the value chain that is much more productive than the rest, this firm will eventually vertically integrate and take over all relevant parts of the value chain. As a consequence, this firm will dominate the sector. Think Amazon, Tesla, Facebook, Google, or Apple. One thing all these companies have in common is that they are by far the most productive firms in their respective value chains. That’s why they dominate their markets. We predict that this trend will continue because software and hardware are developing rapidly, enabling firms that harness the power of Moore’s Law to grow exponentially.
We will see more of that. Villaverde’s paper is a theoretical foundation for the Silicon Valley phrase "Software eats the world". Add to that Clayton Christensen’s seminal book "The Innovator’s Dilemma," and you get the blueprint for the successful enterprise of the 21st century. A firm that aggressively invests in compute and software to solve real business problems and then applies the solutions to conquer previously dispersed markets. Eventually, all business problems will be solved most productively by optimizing along the compute stack. Wether it’s launching a new fashion line, engineering cars or developing pharmaceuticals, compute will do the heavy lifting and deliver solutions faster and at lower cost.
The essay will proceed as follows: First, we tackle the question of what exactly is productivity, and why is it such a powerful force. Software and compute are the driving forces behind productivity. Second, we’ll discuss implications for industrial development and business strategy. From an entrepreneur's point of view, the key is to create a value chain that is optimized for rapid iteration. We predict that such firms will dominate markets and grow much faster than previous generations of dominant firms because compute and software are driven by Moore’s Law, which facilitates exponential growth. Furthermore, we predict that firms will grow in size and power. This brings us to the third discussion point, which is how to deal with what we call the "Libertarian Paradox". On the one hand, you want to facilitate innovation and productive growth. But this type of growth will inevitably lead to single firms becoming much bigger and more powerful. We coin these firms "GDP-sized firms". When single companies accumulate wealth and market capitalizations of tens of trillions of dollars, which is what we expect, corporate governance becomes a much bigger issue. Fourth, we offer solutions to the "Libertarian Paradox". In particular, we suggest a constitution for GDP-sized companies along the lines of the US constitution. It’s a document that governs the evolution of distributed power. Much like the Founding Fathers had to deal with the fine line of allowing for concentrated power while maintaining the pillars of liberty and human dignity, dominant tech companies of the future must find a balance between power and freedom. Fifth, we discuss implications for investors, and sixth, we offer a broader perspective for society. We strongly advocate letting such firms evolve along their chosen productivity path since those firms are going to be instrumental in driving progress and upward mobility for everybody. But we also acknowledge the risk of concentrated power. Another risk is that society puts hurdles in the way through laws and regulations.
1. Productivity
Try. Fail. Correct. Repeat.
Since productivity is such an important factor, it’s worth spending some time discussing it. What exactly is productivity, and how do firms gain or lose productivity? This is an open-ended question, and there is no definite answer. For the purpose of this essay, we define productivity as the process that allows firms to do more with less at scale. Productivity is driven by knowledge about how to do things better, faster, more efficiently, and at a lower cost. We assume that this type of knowledge, like any other knowledge, follows a Popperian process of conjecture, testing, and error correction. Austrian philosopher Karl Popper spent a great deal of his thinking on the subject of epistemology, which is the science of how knowledge is created. His insight is that while there might be an absolute truth, we humans can never be sure to have found it. Knowledge is created by guessing new ways to do things. These novel approaches must be tested and improved.
Firms follow the same process. In order to become more productive, they must try new approaches to doing things, test them, fix errors, and keep iterating. The faster companies iterate, the more productive they become. The key here is to iterate at low cost. In other words, you must find a way to try new things without betting on the farm. Errors will happen, and it’s precisely the moments of error-making and error-correction that are crucial to a successful entrepreneur's journey. Firms must find ways to create elastic value chains where adaptability and fixing errors are not a rare occasion but an integral part of the DNA of the company.
Compute drives productivity
In order to iterate rapidly, firms must embrace computation. It’s not just a tool. It’s the source of competitive advantage, productivity, and wealth creation. There is a saying in Silicon Valley that "every firm will become a software company." That’s only part of the story. "Every firm will become a computation company." It’s not just software; it’s the whole stack, and successful wealth creators design their companies around the computation stack and vice versa. Tesla’s Dojo supercomputer is just one example. Dojo is designed from the ground up to solve certain problems Tesla considers pivotal, such as self-driving. Dojo is designed to meet the work flow of Tesla. But this is just the beginning. Eventually, Tesla’s workflows will also be adjusted to the Dojo stack. Ganesh Venkataramanan, who helped design Dojo, talks about this. Dojo is designed to flexibly interact with new software developments. Iteration must happen across the stack. That’s because Tesla wants to be able to apply Dojo to problems as they arise, such as self-driving cars, humanoid robots, material science, and who knows what’s on the horizon. Most crucially, Tesla will adapt their workloads in the factory to what Dojo comes up with. One example is the humanoid robot, which will eventually change the physical presence of the Tesla supply chain. This is all to say that flexibility, adaptability, and iteration all operate interchangeably. Success in business will increasingly depend on how entrepreneurs integrate compute stacks into their value chain.
Compute encompasses both software and hardware. As Bill Dally, Chief Scientist at Nvidia, says in a recent talk, there are three vectors that drive compute and AI in particular. It’s algorithms, data, and hardware. Firms that find ways to optimize along those three dimensions will inevitably become more productive. Let’s look at Tesla again. In order to develop better batteries, Tesla can use computational resources, hardware, and software to iterate through the vast design space of elements that can be combined to build better batteries. Questions like "should we use more nickel, should we use more graphene, or what if we chose to use two or more layers of graphene instead of one?" Imagine a supercomputer that can run billions of iterations with different materials and layouts optimized for specific goals such as low cost, high efficiency, low power, etc. Computational engineering, which is the term used for this type of approach to material science, is already very popular in academic circles. Firms that harness these new techniques will become more productive than their peers and dominate markets.
When it comes to applying computational tools to real-world business problems, software is often the driving factor. Nvidia, for example, found ways to parallelize computation very early on, but it was the development of a software suite dubbed "Cuda", that helped propel Nvidia GPUs into the world of AI. Productivity is driven by software. Software is optimized by hardware, and data is the glue that ties both ends together.
Why is data crucial? In the world of engineering and business, most problems are complex in nature. Like the Tesla battery example, typical engineering problems have a large design space. Brute-force traditional mathematics based on differential equations doesn’t suffice. Data and software 2.0 are the solution. Reality can be represented in a distributed fashion through billions of parameters. Those parameters are fine-tuned with machine learning models. Think of data as the programmer and the model as the compiler that spits out a binary that is applied to do something useful. This is precisely the reason why AI and machine learning are all the rage in the science world. It’s because we have a new compute paradigm that allows scientists to find solutions to hitherto unsolvable problems.
Just as an example, we recently attended a talk by a cosmologist at UC Berkley. He spoke about the application of AI in cosmology, where vast amounts of data can be analyzed and parameters found that help build models of how the universe evolves through time. This is just one example of how AI, data, software, and hardware are driving progress at a much faster pace. What works for cosmology works for everybody else. You just have to harness the new powers. Firms that do that successfully will dominate markets.
2. Markets will be even more concentrated
When firms enter new markets, productivity determines success. This is what Clayton Christensen originally called disruption. It’s when new firms enter markets and beat incumbents. At the time Christensen wrote his seminal "The Innovator’s Dilemma", the business landscape was still fairly static. Petroleum companies and banks dominated the S&P 500 index, and West Coast technology companies were considered fringe investments on Wall Street. Fast forward to today, and the world looks dramatically different. Despite all the money printing and crony capitalism, East Coast dominance has diminished, and West Coast technology is the major contributor to wealth creation. Nvidia alone added more to its market capitalization this year than the combined market capitalization of JP Morgan and Bank of America, the two largest US money center banks. Tesla’s market cap is larger than Exxon Mobil and Chevron combined. And the list goes on.
What changed? It’s the software, stupid! Productivity is driven by software and its integration into optimized hardware. Computer architects are not only building digital edifices; they are literally building the foundation of wealth creation. People like Bill Dally at Nvidia or Ganesh Venkataramanan at Tesla are designing massive compute infrastructure, which is the foundation of wealth creation in the 21st century. Whether their companies and shareholders actually reap the benefits is not guaranteed. But what is for sure is that their combined efforts are the edifice for wealth creation. There is no doubt about that.
Starlink, a subsidiary of Space X, is solving the telecommunications problem by spreading thousands of satellites in low orbit and meshing them together with optimized network software. Space X will likely soon become a telecom provider and compete with incumbents such as AT&T or Verizon. Tesla could have chosen to develop technology and sell it to OEMs, which is what Apple, Google, and Nvidia are currently doing. We predict that all of them will eventually choose to vertically integrate and become car makers, transportation providers, and software companies in one. The main reason why these companies are not able to work with incumbents is that productivity is way too different between Silicon Valley technology companies and car manufacturers. Such dispersion is not sustainable. We predict vertical integration in the transportation market, with a few technology-focused disruptors taking a major share.
Incumbents don’t like progress
Unfortunately, there are still plenty of industries where productive market entrants struggle to gain traction. One example is radiology, where AI agents are better than humans. But instead of dominating the radiology space, lowering cost, and increasing throughput, AI-driven radiology is still struggling to replace the status quo. Healthcare is a sad example of how incumbent industries successfully hinder progress. Another example is education. In particular, higher education has so far been able to thwart new, more efficient solutions from prospering. All the top US research universities were founded more than a hundred years ago, which is a testament to how archaic the US university landscape has become. Not surprisingly, tuition expenses have skyrocketed. When I got my MBA more than two decades ago, tuition was around $60,000. Today, tuition at the same school for the full-time program is 162k, which is almost three times. Graduate compensation back then was around 125k. Today the same school estimates around 200k for freshly minted graduates. While tuition went up almost threefold, compensation didn’t even double. This is to show how inflation in academia is outpacing general inflation. Let’s hope that software driven Silicon Valley startups find a way to enter the MBA market, which would inevitably lead to lower cost and higher value for students.
3. Winner takes it all and the “Libertarian Paradox”
But there is a catch. When more productive market entrants penetrate new markets, they tend to win it all. Importantly, this is not because they engage in some obscure Rockefeller-style market manipulation. It’s because the firm with the highest productivity eventually dominates the value chain. Productivity is for business what gravity is for physics. You can’t escape it. This leads to what we call the "Libertarian Paradox". When highly productive firms enter markets, they win over the value chain and often become dominant players with lots of concentrated power. Markets don’t work well when single firms dominate markets. Society doesn’t like that, either. What can be done here?
First, we don’t want to prevent more productive market entrants from bringing their superior business practices into markets. Companies like Tesla, Space X, Nvidia, or Amazon create new, lucrative jobs, foster research and development, and inspire young people to study hitherto unpopular subjects such as manufacturing, computer architecture, or logistics. These companies are good for society. But their market power isn’t. So the question is not how to stop them, but how to mitigate their market power.
4. Constitution for corporates
We suggest a constitution for corporations. A document that governs the impact of dominant tech companies along the lines of the US constitution This is a much bigger topic and should be discussed thoroughly in a separate essay. The idea is to do what the Founding Fathers did when they wrote the US constitution. Their task was to balance the need for executive power with the overarching goal of liberty, freedom, and the pursuit of happiness. The US constitution is not a business plan to create the most powerful country in the world. Nor is it a blueprint for wealth creation. It’s main purpose is to safeguard the pillars of liberty while navigating the necessities of political realities where concentrated power is sometimes required.
The "Libertarian Paradox" is a similar problem. On the one hand, we must encourage innovation. On the other hand, we don’t want too much concentrated power. To solve this problem, we need a constitution for corporations. Something along the lines of the US constitution We will describe our ideas in more detail in an upcoming essay.
5. Compute generates outsized returns
At Orange Capital Partners, we concern ourselves with wealth creation. Our purpose is to invest in fundamentally new concepts and engineering practices with large impact. Investing alongside entrepreneurs with long-term vision, high productivity, and exceptional business acumen is what we do. Of course, this process is hard because it’s complicated. But there are some basic factors to consider. The most important driver of productivity is compute. Harnessing the power of compute and applying it to business problems is what generates wealth.
Take a shipping company in the mid-19th century. How should they route their carriages from New York to Chicago? One way to find out is by trial and error. A better way is to use computational resources and calculate the most efficient trajectory. Even in the 1850s, computation was available to entrepreneurs. Moore’s Law has been operating even before the invention of the transistor. Computation is at the heart of science, technology, and entrepreneurship. Today, it's at the center of wealth creation because we have much more computational resources at our disposal. And the fundamental concept remains the same. "Use computational methods to solve problems". Whether it is to schedule rides (Uber) or to find more efficient ways to develop batteries (Tesla), Computation is key. That’s because computation, in essence, is a tool for rapid iteration. Productivity is a function of rapid iteration. In order to increase the speed of iteration, entrepreneurs must find ways to lower the cost of errors and error correction. In other words, iteration can only happen at a rapid pace if it can be done in such a way that errors don’t cost too much. Fixing those errors is key, but you must make sure you’re not on the brink of bankruptcy every time you iterate.
The best way to iterate rapidly is to do it in virtual space. That’s what computation is. It’s iteration in virtual space. Take the example of material science. When firms like Tesla try to find better battery chemistries, they can either do it the old-fashioned way, which is try and error in the lab. A much better way to do this is to use computational methods and iterate over a multitude of elements and combinations thereof. The same method can be applied when trying to find better alloys for the car or more suitable materials for fast charging. Some of this research is done in-house, and some of it is sent out to universities. Firms don’t have to do all the work. But they must find ways to coordinate the value chain and apply computation to crucial business problems.
The key insight here is that iteration can be accelerated and costs substantially lowered if firms find ways to iterate in virtual space and then take the learnings from there to build the most optimal solutions in physical space. AI is at the core of this endeavor. We predict massive advancements in applying AI to engineering problems, which will increase the pace of innovation and propel firms that harness this power to market dominance.
One of the most obvious computational problems is self-driving cars. Computation is at the heart of the solution to this problem. That’s why Tesla has chosen to develop its own training and inference hardware. But this is just the beginning. In a recent talk, Ganesh Venkataramanan said something very interesting. He describes the virtual circle of algorithms and hardware. It’s not just that hardware has to adapt to the latest innovations in software; software can also be developed based on innovations in hardware. Software and hardware can be optimized in tandem. Computer architects such as Ganesh are going back to the roots of computer architecture by designing their systems from the ground up with the full stack in mind. This will unleash a massive wave of innovation.
Follow the compute
We suggest a new mantra for investing, which is "Follow the compute". Entrepreneurs who run their businesses with compute in mind are well positioned to become the wealth creators of their generations. Jeff Bezos, Elon Musk, and Jensen Huang are contemporary examples of this new breed of entrepreneurs. Wealth creators still must excel at traditional entrepreneurial skills such as business acumen, successful team building, and political skills. But computation is at the center of wealth creation. Those who get this will dominate markets. Harnessing compute to solve business problems is at the heart of wealth creation and the source of outsized returns.
6. Productivity is the key to prosperity
It’s hard to make this case without sounding like a libertarian parrot. But I’ll do it anyway. Disruption is good for society. It’s good for everybody because it transcends hierarchies and democratizes access. Any piece of advanced technology is close to magic, and magic is what makes people better off. One of the most striking examples of the magic of technology comes from the Viking age. When Vikings were sailing from Norway to Greenland, they often found covered skies. How could they navigate without seeing the sun? They used something called a sunstone, which is a crystal that serves as a measurement for light polarization. Thanks to the polarization readout, Viking navigators were able to detect the location of the sun and the direction they were sailing in without seeing the sun directly. This must have felt like magic to the Vikings, who didn’t have a clue about the actual physics behind it. But they used the technology anyway and successfully navigated the northern waters for centuries. Science and technology are the keys to magic, and magic makes people better off. Whether they use their powers for good or bad is up to them, but technology offers options, and options are good for everybody.
David Deutsch defines wealth as the "set of possible transformations," which is to say, the more options you have to do something, the wealthier you are. These options are a function of knowledge. The more you know about the world, the more you can transform it. Knowledge is the key to wealth creation, and according to Deutsch and Popper, knowledge is created by iteration.
As discussed in this essay, iteration is fundamentally driven by computation. We therefore postulate that wealth and market dominance are functions of computation and iteration. Firms that harness the power of compute for their business problems will be more successful and consolidate their respective value chains. By doing so, they will increase the wealth of society. In order to be socially acceptable, firms must generate more wealth for society than for themselves. Let’s call ws the wealth generated for society and wf the wealth captured by the firm.
wf + ws = total wealth
wf < ws
total wealth > 10 * wealth prior to disruption
In order to be socially acceptable, disruptors should generate at least 10 times more value than incumbents. While some of that value must accrue to investors, most of it should stay with society. Dominant firms will only be tolerated if they satisfy these conditions. Crate value, make money, but leave most of the value to society. If we are right and firms become massive productivity machines, they will be able to do this. Tesla, for example, is already generating massive amounts of wealth for society by reducing emissions and inspiring new studies in academic fields such as material science, battery technology, and solar power. Some of that wealth is captured by the firm and reflected in Tesla's market capitalization. But most of it dissipates into society. A fine balance between those two factors is necessary to maintain social acceptance. It’s important to note that disruptors are good for society. We want more of them. We also don’t want society to put regulatory hurdles in their way. Yes, there are losers in this game, such as the workers in incumbent industries. While they are losing wealth and status, many more are gaining. That’s why disruptors must be protected. We suggest a constitution to govern this dynamic of productivity-driven market dominance.
Conclusion
Productive firms choose to work with other similarly productive firms. That’s why productivity levels converge across value chains. If differences in productivity persist, firms either consolidate or vertically integrate. We expect software-driven technology firms to enter new markets and reach dominant positions because they reach higher levels of productivity than incumbents. "Software Eats the World" meets "The Innovator’s Dilemma". Disruption is a matter of applying compute to real business problems. Due to higher levels of productivity, firms will dominate markets and achieve GDP-size market capitalizations. High market power and concentration can be mitigated by creating a constitution for corporates modeled after the US constitution.