Unscaled

The old strategy of beating competitors by owning scale has in many cases become a liability and burden. Procter & Gamble, with all its magnificent resources, finds itself vulnerable to a newcomer like the Dollar Shave Club, which can rent much of its capabilities, get to market quickly, target a narrow market segment, and change course easily if necessary.

Tullman was particularly interested in diabetes. For starters, it’s the fastest-growing disease in the world, and there are over 30 million people with diabetes in the United States alone. We also knew that diabetes is a manageable disease—people who are careful can live pretty normally.

Over the past four or five decades carbohydrate-heavy diets—pushed by mass-market production and mass marketing of cereals and drinks laced with high-fructose corn syrup—created an epidemic of obesity and, ultimately, diabetes.

For more than a century size mattered. Economies of scale reigned as a competitive advantage. They worked like this: if a company spent a billion dollars to develop a physical product and build a factory, the amortized cost, at the extreme, would be a billion dollars to make one unit but only one dollar for each unit if the company produced a billion of them. So scale gave a company a cost advantage over competitors.

The emergence of powerful artificial intelligence and the economic force of unscaling can trace their beginnings to 2007, when the Apple iPhone, Facebook, and Amazon Web Services—pioneering mobile, social, and cloud platforms—took wing at roughly the same time. As more of work and life moved online thanks to such platforms, the amount of data exploded. At first the explosion just seemed like more data that could inform business, and we even called it Big Data, as if that’s all there was to it. But Big Data turned out to have a higher purpose. It was the key to making AI, which previously had a long and tortured history of disappointment, into a force that will literally change the world.

The more platforms we build, the less an individual company—or lone entrepreneur—needs to do by itself in order to create, produce, market, and deliver a product.

A number of other important technology platforms emerged around 2007 and took hold in the years after. When Amazon.com, which had already moved commerce online, launched Amazon Web Services (AWS) in 2006 it gave every software developer the power to launch a cloud-based software product and become an entrepreneur. Facebook was founded in 2004, but it wasn’t until 2007 that it turned into a platform, opening up so developers could build applications on it. Added together, 2007 can be called the origin point of an AI revolution, made possible by the combination of mobile computing, cloud computing, and social networking. In 2007 a little more than 1 billion people were on the internet; by 2016 it was 3 billion. Smartphone use had grown from a tiny sliver of society in 2007 to more than 2.5 billion people in 2016.

I decided to learn at my own pace and take classes from as many departments as I could. I remember recognizing early on that being a straight-A student, although a tremendous accomplishment, wasn’t going to matter in the long run. So I would regularly skip classes, often joking with my friend Sal Khan that classes were always too fast or too slow for me. Well, at least that was my excuse for skipping the classes. Years later Sal went on to start Khan Academy, with self-paced learning as his early leverage point for transforming education.

We’re shifting from mass-produced products that can be sold to the most people possible to highly personalized products that delight small niches of passionate customers—at prices that are often lower than the mass-market products.

The unscale mindset asks: What can I build that makes each individual happy? That’s a big change from last century’s mindset of: What can I build to sell to the most people?

in the 1980s, Bob Metcalfe, who is credited with coinventing Ethernet, one of the earliest computer networking systems, described the exponential power of networks by showing that the value of a network is proportional to the square of the number of users connected to it. That exponential dynamic meant that as more than 3 billion people connected to the internet from 1995 to 2015, the internet exploded in power and value—creating a societal and economic impact far greater than just the number of people connected.

But now Moore’s Law and Metcalfe’s Law are reaching diminishing returns. The laws of physics mean that microprocessors can’t get much smaller and faster anymore, and if most of the world that will ever get connected is already connected, the benefits of Metcalfe’s Law taper off.

The cloud is essentially the meeting point of Moore’s Law and Metcalfe’s Law—where data, computing resources, and connectivity have merged.

Economist Carlota Perez describes the impact of such revolutions in her influential book Technological Revolutions and Financial Capital: “When a technological revolution irrupts in the scene, it does not just add some dynamic new industries to the previous production structure. It provides the means for modernizing all the existing industries and activities.”

Unscaling will involve transitioning away from ownership and toward accessing services.

The key to success for most people will be living an entrepreneurial life and becoming their own personal enterprises, selling services on demand through the cloud to many employers.

Unscaling is disruptive. It is remaking an old economy into a new one. Whenever that has happened in history, whole categories of jobs disappear, and it will be no different this time as AI automates many new tasks.

accountability in their services. Several big projects in longevity are aimed at extending life expectancy by decades. Google’s Calico is putting $1.5 billion into discovering the basic science behind aging, the Jeff Bezos–backed Unity Biotechnology is investigating drugs to rejuvenate aged tissues, and we at General Catalyst invested in Elysium Health, a company with a stable of expert aging and bio-scientists focused on boosting cellular NAD+, a critical coenzyme that begins to decline in our twenties.

My friends Sam Altman, who runs the tech incubator Y Combinator, and Chris Hughes, a cofounder of Facebook, have kicked off two separate universal basic income (UBI) projects that explore replacing employment-derived income with unconditional stipends. Both are trying to get ahead of the impacts of the highly automated, postwork world we’re headed toward. But although UBI replaces monetary loss, it does not address something just as fundamental: purpose.

In digital industries, more than in physical-product industries, the tendency is toward winner-take-all.

We’re at the cusp of an amazing adventure. We have a chance to rewrite our world and solve some of the greatest problems we face, from climate change to cancer. As with the last technological revolution, by the time we’re finished, the planet will be almost unrecognizable.

Artificial intelligence is this century’s electricity.

In the 1880s small electric stations built on Thomas Edison’s designs were spreading to cities, but each could only power a few blocks of buildings. At the end of the 1890s New York patent attorney Charles Curtis developed the steam turbine generator, which for the first time allowed mass-market electricity to be produced inexpensively…

With electricity Guglielmo Marconi completed the first twoway wireless message—a fifty-four-word greeting…

Alexander Graham Bell invented the telephone in 1876, and it took hold in…

Other bold technologies arose. From 1900 to 1902 the Germans invented the zeppelin airship, George Eastman developed the first consumer camera, salesman King Gillette created the first safety razor, and the first electric stoves made their way into homes. People could do things they’d never done before—fly,…

Into this milieu strode Henry Ford. In the late 1800s he had worked at the Edison Illuminating Company, where he met and was inspired by Thomas Edison. By night he experimented with motorized quadri-cycles. He started two automobile companies in three years, and both failed. In 1903, just before…

The effect on popular thinking is hard to imagine. In 1903 horses were so prevalent that every day in New York 2.5 million pounds of horse manure was deposited on the city’s streets. In 1908 Ford unveiled the Model T, which blew the lid off the industry, albeit with a modest…

By 1910 one cartoonist’s depiction of the future showed grade-school students driving tiny cars to class. By 1913 Ford’s sales mushroomed to 179,199 cars, and the numbers shot up from there. And it was all about scale. The Model T famously came in one color:…

In Dayton, Ohio, the Wright brothers built on advances in engine technology and new ideas about winged flight. In their bicycle shop they and their mechanic, Charlie Taylor, built an engine and married it to a flyer with a wingspan of about forty feet to be the first, in 1903, to successfully…

Radio opened up the concept of mass-market…

To build massive amounts of physical products, ship them, sell them, and advertise them through a limited number of media outlets, companies needed to get big—and once they were big,…

The Fortune 500 list—an unabashed celebration of scale—debuted in 1955. General Motors topped the list, and it had 576,667 employees. In 2016 Walmart was number…

Governments scaled up too. The US federal government employed about 1 million people in 1900 and…

Massive brands like Budweiser, Coca-Cola, and McDonald’s served the same thing to everybody and wiped out niche competitors, while Walmart nuked local retailers by…

The Western world scaled up mass-education schools, modeling them on assembly lines—children entered kindergarten and would move through the system one step at a time, all learning mostly the same things, until…

around 2007, as we started to move more of our lives onto mobile, social, and cloud platforms that could collect enormous amounts of data, AI could finally make an impact that rivals that of electricity in the early 1900s.

Much the way we electrified the world to set the previous scaled era in motion, we are infusing AI into the world today, and that has set in motion unscaling.

Hunch, a business founded by Chris Dixon, marked the start of my involvement in AI-related companies. As an undergraduate at Columbia in the early 1990s, Dixon majored in philosophy. He went on to get his MBA and became a software developer for Arbitrade, a hedge fund that focused on high-frequency trading. Dixon then started a company called SiteAdvisor, which helped internet users avoid unwanted spam. SiteAdvisor was the first investment I ever made as a venture capitalist. In 2006 we sold the company to security software firm McAfee.

The technology worked well. But we ran into a challenge: AI needs to learn from enormous amounts of data, so the more data, the better the AI. As an independent entity, Hunch just could not get enough data from enough users to make the AI sufficiently effective to convince even more users to join Hunch—which might’ve kept making the AI better in a virtuous cycle.

In 2011 an excellent path forward surfaced. We sold the company to eBay for $80 million. At the time, eBay had 97 million users, 200 million active listings, 2 billion daily page views, and around nine peta-bytes of data about all of that activity. “With eBay’s data behind us, expect Hunch to get much, much better,” Dixon said when announcing the deal. We finally had the data to train our AI to be really good.

“Hunch discovered that a certain class of users who were buying gold coins were also the perfect customers for a microscope that they could use to examine those goods,” eBay chief technology officer Mark Carges told the press after the deal. “That is the kind of odd association we never would have found on our own.”

Importantly, this taste graph is exactly what Facebook implemented with its “like” button—and did so much more effectively because it assembled more than a billion users who clicked “like” buttons constantly throughout each day.

So we spent a decade connecting people and moving a great deal of our activities online, where every action generates data. Then, in recent years, industry jacked up the planetwide data-generation machine by implementing the Internet of Things (IoT), which puts “things,” not just people, on the global network.

As sensors get embedded into almost everything, the technology will create a kind of quantified planet.

In the 2000s, as people in Silicon Valley liked to say, software was eating the world, moving into every nook and cranny of business and life. But in the 2010s and beyond, the world is eating software—everything is ingesting software, becoming smart and connecting to the global internet.

In industrial settings sensors throughout factories or in assets like trucks or jet engines can all feedback through a system like GE’s Predix, a cloud platform that connects things and people.

The IoT explosion will give us an astonishing flow of data, opening possibilities for AI to conduct deep analysis of how the world works.

IoT is giving us instant, real-time views, as if we are hooking up the planet to an EKG and watching its heart beat.

AI is all about discerning patterns in data, predicting behaviors, and deciding on actions.

With little data coming in, AI is like a baby’s brain—all the smarts are in place, but it has too little knowledge of the world to understand what’s going on or to know that, say, pulling the cat’s tail will get you scratched. So, like a brain, as AI gets exposed to greater amounts of data, it can see patterns and predict behaviors with greater accuracy. The more data coming in, the better AI gets.

We already encounter AI all the time. Google’s AI-driven search algorithm learns from every search and gets better. Facebook’s AI learns from your posts and likes and then populates your timeline with feeds and ads you’ll probably want to see. Netflix learns from your viewing habits, matches it with what it’s learned from millions of other viewers, makes recommendations, then uses that AI-based learning to guide its decisions about what movies or series to produce in order to appeal to the most users.

In 2016 IBM bought The Weather Company, which gathers incredible amounts of weather data from sensors all over the planet, so IBM can feed its data to IBM’s Watson AI. Now Watson can literally “learn” how weather works and make hyper-local microforecasts—for instance, predicting the wind patterns at the location of an outdoor Olympic diving event.

As I write this in 2017 Google, Tesla, General Motors, and others are developing self-driving cars. AI makes the technology possible, and the more cars run on auto-pilot, the more data those AI systems will collect, making the autonomous vehicles ever better.

AI is getting built into everything that has computing power. In another ten years anything that AI doesn’t power will seem lifeless and outmoded. It will be like an icebox after electric-powered refrigerators were invented.

Society has come to need AI. The world’s systems have gotten so complex and the flood of data so intense that the only way to handle it all will be to employ AI. If you could turn off every AI program in use today, the developed world would shut down. Networks would seize up, planes couldn’t fly, Google would freeze, spam would overrun your inbox, the Postal Service couldn’t sort mail, and on and on. As the years go by, AI will become yet more integral to keeping the planet’s systems running.

Jeff Hawkins, the CEO of brain-like software company Numenta (and the guy who invented the PalmPilot) explains, “We have made excellent progress on the science and see a clear path to creating intelligent machines, including ones that are faster and more capable in many ways than humans.” As an example, Hawkins says we can eventually make machines that are great mathematicians.

“You can build an intelligent machine that is designed for that. It actually lives in a mathematical space, and its native behaviors are mathematical behaviors. And it can run a million times faster than a human and never get tired. It can be designed to be a brilliant mathematician.”

AI allows the profitable customization of everything—because AI can automate customization.

Kosslyn went on to work at Google and YouTube, while Thompson bounced through a few startups. But they kept talking VR and watching it improve. And then, in mid-2014, Facebook bought Oculus Rift for $2 billion. “That just put everything into high gear,” Kosslyn says. Facebook’s move prompted Google to pump money into VR research.

A 3D printer is a catch-all term for a robotic device that can take some kind of raw material—plastic powder, stainless steel—and shape it into a physical item based on digital blueprints.

an on-demand 3D printing center can manufacture one item for nearly the same cost as making one of one hundred thousand or 1 million. It can make those items as they’re ordered—no need to predict demand and make and ship thousands of a product that may not sell. This, Friefeld states, will turn economies of scale upside down. “We’re trying to reverse two hundred years of evolution in manufacturing,” he says. Eventually, he believes, most companies that make physical goods will rent manufacturing as they need it, just as today they can rent cloud computing power as they need it.

As blockchain develops, instead of having an internet that puts information and content online, we’ll get a system that essentially automates trust and verification—the kind of stuff we now rely on accountants, lawyers, banks, and governments to do. You’ll be able to know that anything on a blockchain (money, a deed, a person’s identification information) is authentic.

Everledger, for instance, is putting diamonds on the blockchain. First, Everledger’s software creates a digital fingerprint of a cut diamond by measuring forty points on the stone. No two diamonds are exactly alike, and this creates a unique digital fingerprint. From that point on, the blockchain has an unalterable record of a diamond’s path.

Another blockchain company, Abra, changes how cash gets sent to individuals around the world. On one side are people who sign up, in an Uber-like way, to be virtual bank tellers. On the other side are users—like an immigrant in the United States who wants to send money to his mother in the Philippines. The user pulls up a map-like app to find the nearest teller, and the two agree to meet. The user gives the teller money, and the teller uses his or her account to put that amount of money into Abra’s blockchain-based system. In the Philippines the user’s mom similarly locates a teller, who translates the money into local cash to hand to Mom. The whole process cuts out banks, costs a fraction of the fees banks charge for such transfers, and can happen in an instant instead of ten business days.

In 2016 IBM started offering blockchain technology for supply chains. As more networked sensors get embedded everywhere, these devices will be able to communicate to blockchain-based ledgers to update or validate smart contracts. This would allow all parties to know whether the terms of a contract are met.

In February 2001 the Human Genome Project and Craig Venter’s Celera Genomics published the results of their human genome sequencing within a day of each other. The results were a 90 percent complete sequence of all 3 billion base pairs in the human genome. Venter later was quoted, saying his project took twenty thousand hours of processor time on a supercomputer. Getting that first sequence proved as daunting a project as putting the first man in space.

Color Genomics, is offering a $249 genetic test that can sequence most of the pertinent genes in the human body. The goal is to make genetic sequencing so cheap and easy that every baby born will have it done and the data will inform his or her healthcare for life.

AI and data will change medicine from prescriptive to predictive: doctors will be able to treat diseases such as cancer before they even manifest.

Imagine how this could unscale the pharmaceutical industry. Over the past fifty years the industry’s goal was scale. Every company sought a “blockbuster” drug—a drug that would have some impact on the most people possible. Humira for arthritis, Crestor for cholesterol, and Viagra for erectile dysfunction were classic blockbusters.

AI-driven medicine will allow doctors to tailor care to every patient and focus on preventive and predictive medicine, which will drastically reduce the number of people who need to be in hospitals or even to see a doctor.

Economies of scale address mass markets. Economies of unscale prevail when entrepreneurs can address micro-markets.

This is the year 1900, the dawn of the twentieth century, times ten. We are reinventing our planet and ourselves. AI plus genomics will mean that precision health beats mass-market population health. AI plus 3D printing will help focused, niche production beat mass production. AI plus robotics will upend today’s transportation system. AI plus VR and AR will recreate media and personal interactions. All these technologies will come together to revolutionize industry after industry, reversing a century of scale and driving unscale.

“Humanity is now entering a period of radical transformation in which technology has the potential to significantly raise the basic standards of living for every man, woman and child on the planet,” write Peter Diamandis and Steven Kotler in Abundance: The Future Is Better Than You Think.

“A society that had established countless routines and habits, norms and regulations to fit the conditions of the previous revolution does not find it easy to assimilate the new one,” writes Carlota Perez. “So a process of institutional creative destruction will take place.” As always, creative destruction is kind to the creators but brutal on those whose companies, careers, and finances get destroyed in the process.

Inventions like gene-editing technology CRISPR allow us to alter genes and thus alter people. We’re close to being in control of our own evolution.

In 1998 Patel helped found Sycamore Networks, which made optical switches and software that helped move data around fiber-optic communications networks, and he worked as its chief technology officer. The company was red-hot during the late-1990s public infatuation with the first generation of internet companies, rocketing to a $45 billion valuation in 2000.

“I saw the beginning of customer-driven change,” Patel says. “One-service-fits-all is no longer going to be sufficient.” With that observation he founded Gridco, a company that would make internet-style switches and software for the electric grid. The internet can transmit information in two directions—from any provider located anywhere to users and then back—because of a system of routers and the software inside those routers.

Energy and transportation go hand in hand. Energy made the transportation complex possible, and transportation created massive demand for energy. Just as the energy sector sought to operate at scale, so did transportation.

The scaled approach built inefficiency into the system. The electric grid and highways were overbuilt to serve the most people at peak times, even if those resources went to waste most of the rest of the time.

The car is perhaps the most tangible example of this wanton inefficiency. We waste a vast amount of energy building and maintaining a car, often two or more cars, for almost every person—yet most cars stay parked and idle about 90 percent of the time.

If one result of unscaling is creating focused, innovative companies around car sharing and on-demand transportation, we’ll expend less energy to build fewer cars that will serve more people, greatly improving energy efficiency while remaking transportation into a customizable service that’s actually more practical than owning a car. After all, most people just want to get from point A to point B.

Energy can’t unscale without transportation unscaling, and vice versa.

Tesla, founded in 2003 in California, started out by bringing a high-performance electric sports car to market and has since expanded into electric sedans, home battery systems, and solar power. Musk gets it right when he thinks about his pioneering company as an integrated energy and transportation entity.

Musk in 2016 put forth his “Master Plan, Part Deux” (a sequel to his first Master Plan, which he published in 2006). He wrote that Tesla’s ultimate goal was never to produce hot electric cars (even though Tesla unveiled the fastest-accelerating car on earth). Tesla built hot electric cars as an entry point for ending dependence on oil. “The point of all this was, and remains, accelerating the advent of sustainable energy, so that we can imagine far into the future and life is still good,” Musk wrote.

A Finnish company, backed by Toyota, is promoting what it calls “mobility as a service,” or MaaS. The company, MaaS Global, developed a transportation subscription service called Whim. In 2017 early users paid from around $100 to $400, depending on the level of service, for Whim. A user could select a destination on a map, and the app would list possible ways to get there, including taxis, public transportation, rental cars, and bikes—choose the best one, and the subscription fee covers the cost.

Founder Samp Hietanen started the company based on a research paper he wrote while working for a Finnish smart-transportation think tank, ITS Finland, and Hietanen wants to take the concept global. “What if you had ‘unlimited Europe’—ground and air transport included—through one app that would be your companion wherever you go? Then you could truly be a global citizen,” Hietanen told reporters.

A ten-year unscaled vision for energy and transportation goes like this: An increasing number of homes and buildings will have cheap and super-efficient solar panels on their roofs and high-powered batteries in their basements or garages. The batteries will store power generated when the sun shines for use when it doesn’t. The electric grid will operate more like the internet, allowing anyone to sell excess energy or buy needed energy in an eBay-style marketplace. Energy customers will have more choice about where their power comes from, much the way they can now choose many different ways to make a phone call (landline, cell phone, Skype, and so on).

Home solar panels and batteries will charge a family’s electric car so the home will supply much of the clean power a family needs. Some families might not need to buy any electricity from far-off coal-burning electric plants or ever again fill up at a gas station.

As transportation moves to the electric grid and off carbon, oil will become a shrinking piece of the energy pie—much like the decline we see today in coal. Some people will drive gas-powered cars for fun, just as some people still ride horses. One by one, gas stations will go out of business or convert to electric car charging stations. As demand for oil drops and prices plunge further, drilling new wells will become bad business. Before long the amount of carbon we’re chugging into the air should drop dramatically.

The grid, as noted above, is evolving into a platform. And as a platform it can become a key to unscaling: a rentable resource, like cloud computing, on which small, product-focused companies can innovate and address smaller niche markets. Call it the “power cloud”—an energy platform that will enable an explosion of unscaled energy businesses.

Solar will be a major driver of new unscaled energy business. The technology of rooftop solar is on a predictable trajectory—echoing Moore’s Law, which explained why computers got twice as powerful for the same price every eighteen months for decades, though solar is not improving at that same breakneck pace. Still, the cost of solar has dropped 95 percent since the 1980s while efficiency has rocketed.

By some calculations the amount of solar energy that hits the earth is more than five thousand times the amount of energy all of humanity uses. The challenge has been harvesting it. Solve that problem, and mankind would never need to burn another atom of carbon.

Google, which in 2016 consumed as much energy as the whole city of San Francisco, said that by the end of 2017 all its data centers globally will run entirely on renewable energy.

Solar has reached its PC moment. It’s now economically viable for individuals to install panels and build their own power plant—and I expect it will soon get easier for any entrepreneur to start a solar power–generating business and connect it to the grid.

Wind power has a different dynamic. In renewable energy wind power is actually not unscalable. Wind has economics similar to that of power plants. Wind is designed for scale. The enormous windmills you see on hillsides generate enough power to make them worthwhile. But you can’t put a small windmill on your roof and get enough power to make a difference; the physics don’t work that way. If energy is trending toward unscale, wind is likely to play a relatively minor role in the future of energy.

Over the past fifty years the car companies convinced everyone that we had to own at least one car, if not more. Car sharing is breaking that cycle. It’s teaching us how to get around without owning a car. One car can serve many people now instead of one person owning many cars.

Uber is offering rides in autonomous cars in Pittsburgh, and major car companies—Ford, Volvo, BMW—are predicting that they will be selling driverless cars by 2021. General Motors and Lyft are collaborating on self-driving cars and say they’ll be ready by 2021, though Lyft CEO John Zimmer says the vehicles will serve limited geographic areas where top speed is twenty-five miles an hour. Tesla cars are already completely capable of driving themselves in many situations, though a driver still needs to be behind the wheel, ready to take control.

In Pittsburgh Aquion Energy has been developing what it calls saltwater batteries and has installed them on a solar farm in Puerto Rico.

In the United Kingdom Dyson, the engineering company that makes the popular vacuum cleaner, is working on home batteries.

In Germany Mercedes makes a home battery that it intends to roll out to global markets.

John Goodenough announced in early 2017 that he and his team at the University of Texas at Austin had invented a glass-based battery that blows away the performance of every previous kind of battery, including lithium-ion batteries—which were invented in the 1980s by … him. Goodenough’s new battery can store three times more energy than a comparable lithium-ion battery, according to the Institute of Electrical and Electronic Engineers.

On the consumer side Nest popularized the smart thermostat that can learn the patterns of a home’s residents and use that information to more efficiently control heat and air conditioning. Smart lights from Philips and GE could gather data about lighting use and automatically turn on and off. In businesses General Electric introduced the “industrial internet,” and tech giants such as Cisco and IBM created IoT offerings that could put sensors into almost anything that ran on electric power.

All this IoT activity is generating enormous amounts of data about energy usage—information no one ever had before. These insights will play an important role in unscaling.

As is often said, in the twenty-first century data is the new oil—the raw material that makes everything go. In energy, data is more than just a metaphorical new oil—it’s critical to how we’ll actually replace oil.

Time after time we’re seeing small companies win in established sectors by building on existing platforms and finding a new market. That’s what needs to happen in energy.

A company called Bastille is developing IoT devices and software that can monitor an electric grid and detect threats from hackers, storms, or anything that might disrupt service.

“By the time your latte hits the end of the counter, your car’s probably charged. That’s what we want to get this to,” CEO Pasquale Romano told Forbes. “It can’t be any harder than going to a gas station.”

THE STORAGE CHALLENGE: Battery technology remains one of the toughest problems to solve in moving power off carbon and over to solar or wind generation. Oil and natural gas can be stored in a tank for use later. The only way to store sunshine or wind is to hold the power it generates in batteries. But batteries aren’t yet good enough or cheap enough to universally solve this problem. We’ll need batteries that can store enough solar-generated power to last through a week of stormy days or power a car for a full day of driving. Getting there will take breakthroughs in materials science. The company that gets there first will change the world.

Jay Whitacre, of Carnegie Mellon and Aquion Energy, is pushing ahead with his “salt water” battery. A Chinese company, Contemporary Amperex Technology Ltd (CATL), is competing against Tesla to become a giant manufacturer of lithium-ion batteries, the type of batteries that now power electric cars and serve as home batteries.

Because of the hard science and big manufacturing involved, this will be a tougher sector for startups. Yet venture investment is flowing to companies.

Current nuclear-energy plants work on fission, which is difficult to control and releases radioactive material. Fusion is the way the sun generates energy—by fusing atoms together under enormous pressure.

Whereas scientists have been able to set off fusion reactions, it’s always taken more energy to create the reaction than the fusion generates—not a great equation for supplying the world with energy.

investor Peter Thiel funded a fusion startup called Helion Energy. The International Thermonuclear Experimental Reactor (ITER) Fusion project is a joint effort by thirty-five countries working to prove that fusion is feasible. The budget for this project is $20 billion, and it aims to power up the reactor in 2025 for the first time.

Tri Alpha Energy, funded by Microsoft cofounder Paul Allen and other investors, has built a fusion machine that forms a ball of super-heated gas—at about 10 million degrees Celsius—and holds it steady for five milliseconds without decaying away; five milliseconds is far longer than other efforts have managed.

The world invests almost $2 trillion in energy every year, but just hundreds of millions of dollars go into fusion research and development, according to Tom Jarboe, an adjunct physics professor at the University of Washington who studies controlled fusion. If we are dedicated to getting mankind off of carbon-based energy, pumping more money into fusion research would be a good investment.

Most every week for a few years after joining Twitter, Othman Laraki and Elad Gil would get lunch, climb out onto the roof at the company’s San Francisco headquarters, and eat and talk for an hour about technology and what they might do next. Laraki was the big-systems software guy with an MBA from MIT. Gil got his PhD in biology at MIT and was naturally interested in genetics. In the early 2000s they both worked at Google, and in 2007 the two cofounded Mixer Labs, which developed software that can help a cloud-based application learn more about its users’ location. They sold Mixer to Twitter in 2009, hence the rooftop meetings.

During one of the meetings, in 2011, Gil brought along a hard drive. He’d paid about $5,000 to have his genome sequenced. A decade earlier that same process would have cost $1 billion. Now Gil’s genome was right there on a simple drive. It was just data. This fascinated Laraki, whose family had a history of the BRCA mutation—the genetic predisposition for cancer made famous when Angelina Jolie in 2013 discovered she had it and opted to proactively get a double mastectomy to head off breast cancer. So Laraki asked whether he could borrow the drive and play with the data to see what he could find. “I told Elad that I’d use it to find all the bugs in him,” Laraki jokes now.

Once Laraki dove in he found that the available software tools for analyzing genetic data were, to put it politely, poor. Or, as he more colorfully described them, “they stunk.” “We were in the prebrowser days of genetics,” he says—in other words, doing anything with genetic data proved as frustrating and balky as trying to use the internet before the arrival of the web browser in the mid-1990s.

In 2013 the duo helped found Color Genomics, based on Laraki and Gil’s insight about a broad swath of people being able to affordably obtain information about their genetic data.

Instead of treating diseases such as diabetes or high blood pressure based on what’s worked for the broad population, you can treat your medical issues based on what will work for you. Drugs can be prescribed or even created based on what will be effective for you, even if they won’t work for anyone else.

Doctors will work hand in hand with artificial intelligence systems that plumb patients’ data to understand a great deal about each patient’s body. We won’t need to build more mega-hospitals because unscaled businesses and services will be able to handle more and more of patients’ needs.

It’s instructive to understand why healthcare scaled so we can see how it can unscale.

UnitedHealth became the largest healthcare company in the world, treating nearly 40 million people a year in the United States and 5 million in Brazil.

Express Scripts grew to be the largest US pharmacy benefits manager, processing more than 1.3 billion claims a year.

McKesson, a huge pharmaceutical distributor, now brings in more than $100 billion in annual revenue.

Two companies, Labcorp and Quest Diagnostics, dominate the testing business.

Johnson & Johnson, Pfizer, and a handful of other corporations have become goliaths of the global pharmaceutical business.

In 1960 average life expectancy in the United States was 69.8 years. Thirty years later, in 1990, the average person lived 75.2 years. Today lifespans are around 78.8 years, according to the Centers for Disease Control and Prevention (CDC)

The economics of twentieth-century healthcare were built primarily on the concept of treating people after they got sick. After all, doctors and hospitals made more money in the fee-for-service model when people came in for more care. In more recent years the industry has pushed toward a fee-for-results model that incentivizes doctors and hospitals to keep people well.

Of course, treating sick people is a lot more expensive than keeping people well, just as it’s a lot more expensive to fix your car after it breaks down compared to the cost of regularly doing preventive maintenance.

The most financially successful drugs have been those that work for the greatest number of sick people and for the most common conditions.

Health insurance has worked best when it can spread the risks among the largest possible pool of customers.

Drug development has been overtaken by what’s called Eroom’s Law—that’s Moore’s Law spelled backward because it is the inverse of the Moore’s Law concept, which dictates that computing power constantly gets cheaper and better. Conversely, drugs constantly get more expensive and less impactful.

The pharma companies are challenged by what’s known as the “better than the Beatles” problem: because most human conditions that can be treated with drugs are already being treated with drugs, any new drug needs to be much better than the old drug in order to reach mass-market success and bring in a return on the investment.

Genomics will exert enormous force on healthcare, and the upside is almost impossible to describe today. A parallel is the time during the 1970s when information was just beginning to be digitized. At that time it would have been hard to envision today’s databases and digital media and data analytics. We’re at such an early point in genomics; not even .01 percent of the population has had its genome sequenced, according to UBS Securities. The cost of getting genetic information is falling faster than Moore’s Law would dictate, diving from a project only a government could fund to, well, Color’s $249 test.

The Broad Institute, an MIT and Harvard genomics research center, predicts that the boom in sequencing will collect a zettabyte (1 sextillion bytes) of data per year by 2025. A zettabyte is equivalent to all the internet traffic on the planet in 2016.

To all this new data we’ll add one more important information stream: our personal health records. Electronic health records (EHRs) are finally taking off after years of discussion. In 2009 16 percent of US hospitals were using EHRs; by 2013 that had rocketed to 80 percent, according to Becker’s Hospital Review.

As our health records become digital and searchable, patients can get access and take more control of their healthcare. The data from IoT devices, genetic testing, lab tests, and doctors’ notes can all combine to generate deep knowledge about each individual’s body and health, teeing up the ability to treat each patient as a market of one instead of as simply a part of a mass market.

a startup called Spruce—founded by Ray Bradford, who’d been an executive at AWS—is a step toward allowing you to “see” a doctor through your smartphone. Spruce started with dermatology. Download the app, and it offers a choice of conditions: acne, eczema, bug bite, and so on. Based on the condition, it asks a series of diagnostic questions (“How would you describe your skin? Normal? Oily? Dry?”). Then it tells you what kinds of pictures to take of your condition. Finally, you can pick a participating doctor to send everything to or just choose “first available.”

Over time AI systems such as Watson will be able to get to know patients while constantly reading and learning more about medicine, leading to better and better answers for the doctors who consult with it.

“As a computer science guy, I see that the way doctors operate today is like expert-systems software,” Color’s Laraki says. “They rely on a decision tree that we made simple enough to fit into an average smart person’s brain. We load it up in medical school. But later its decisions are based on a limited set of inputs and complexity.” The opportunity and upside now is that the doctor, working hand in hand with AI-driven systems, will be able to access a level of data and medical knowledge that would never fit inside one person’s head. “The doctor becomes less like someone using a rare superpower and more like a data practitioner,” Laraki says.

The da Vinci Surgical Systems robot, approved by the US Food and Drug Administration (FDA) in 2000 to assist surgeons in minimally invasive procedures, is so precise that it can peel a grape. Yet some of these robots will take time to perfect and become accepted.

Sedasys is a robot developed by Johnson & Johnson that can deliver and monitor anesthesia for short surgeries without an anesthesiologist present. The FDA approved it in 2013 after it underwent extensive safety trials. And it could drastically lower costs—the charge for using Sedasys for anesthesia was about $200, compared to $2,000 for a human anesthesiologist.

the human anesthesiologists complained vociferously. Hospitals then stopped buying the robots. In 2016 J&J reluctantly stopped making them.