The New Architecture of American Innovation

From Edison's Menlo Park "Invention Factory" 150 years ago to the corporate research laboratory, the American innovation system relied on multidisciplinary teams to investigate, develop, and commercialize new ideas, while also licensing patents to industry.

That bridge between science and commerce looks very different today, write professors Ashish Arora and Sharon Belenzon of Duke University’s Fuqua School of Business in "From Edison's Invention Factory to the new architecture of innovation," a policy commentary published in Science.

The research laboratory model "has largely given way to a distributed system in which universities produce much of the science, start-up companies translate it, and incumbent firms implement and market the products," Arora and Belenzon write.

Large firms didn’t abandon corporate research labs because science lost its value, they explain, but because direct investment in scientific research became less profitable.

Science is now largely produced by universities, while startups — backed by venture capital and often absorbed by incumbent firms — commercialize their inventions. 

But today’s innovation system is increasingly science-intensive, and the new distributed model doesn’t always allocate the resources where they are needed.  While software and life science have thrived, the start-up model “is poorly suited” to deep-tech sectors, which require long, capital-intensive journeys from science to commerce. The complexity of modern supply chains makes these sectors even more dependent on “the visible hand” of large coordinating organizations, yet today's deep-tech landscape lacks firms capable of evaluating breakthrough technologies in-house.

Only “a few world-class corporate laboratories survive,” the authors note, “and policymakers should nurture these firms.” Policy, they write, should focus on sectors where the bridge from science to commerce is most fragile, experimenting with a variety of solutions, many of them public-private, on a sector-by-sector basis.

The core lesson is that the American research university is “the indispensable foundation for innovation,” they conclude. The goal of policy is to protect and strengthen the institutions that connect universities, startups, and corporate laboratories.

Ashish Arora is the Rex Adams Professor at Duke University’s Fuqua School of Business. With Sharon Belenzon, Andrea Patacconi and Jungkyu Suh, he also co-authored the book, “Inventing Prosperity.”

In the following Q&A, Arora talks about the evolution of the American innovation system and what its current configuration means for policy.

Edison’s lab produced more than 400 patents, including the phonograph and the incandescent light bulb. Yet you argue his most consequential innovation was organizational. Why? 

Because invention moved beyond an individual inventor to organized teams.  Before Edison, companies relied on outside inventors. The lab was an organization innovation because inventions increasingly required more than individual creativity. They demanded systematic trial and error, along with an understanding of scientific principles. Although Edison himself was not formally well educated, he brought together chemists, physicists, and engineers, along with machinists, carpenters, and glassblowers (who prototyped the incandescent lightbulb and were vital to the research on vacuum tubes).  Edison called it the “invention factory,” but it also supported important scientific research, including work that ultimately contributed to technologies such as the transistor. This was the first industrial research and development laboratory. 

Leading companies in both the US and Europe soon adopted the industrial R&D lab model.  German chemical firms had begun hiring academic chemists in the late 1860s, while in the US, GE started its research lab in 1900, followed by AT&T, Kodak, and DuPont.  These industrial research labs produced breakthrough innovations such as the transistor, nylon, the microprocessor, the laser, and statins for heart disease. Their researchers also made fundamental discoveries in physics and chemistry, earning Nobel prizes for corporate researchers such as GE’s Irving Langmuir and AT&T’s Clinton Davisson. It would not be wrong to say that industrial research powered America’s rise as the global technology leader.

You describe today's system as "distributed" — universities produce science, startups translate it, incumbents commercialize it. What does this model get right, and where does it fall short?

As we show in our book, “Inventing Prosperity”, the division of innovative labor between universities, startups, and larger corporations has worked well in biopharmaceuticals and, more broadly, in digital innovation. The distributed system relies on startups to find applications for university discoveries in biopharma and to apply existing computing technologies to address unmet needs of consumers and firms in digital technologies. However, in many science-based sectors, such as advanced materials, both technology and markets are uncertain. 

Managing both types of uncertainties is difficult for startups.  Manufacturing carbon-nanofiber materials at scale involves many scientific and technical unknowns, and the challenges depend on the intended application, whether for lightweight spacecraft or aircraft components. These technologies also involve long development cycles and often uncertain regulatory pathways. Also, existing financing models are better suited to biotech, where progress can be measured through clear technical milestones, or to software, where products evolve through rapid market feedback. They are less well suited to deep-tech ventures.

You write that today’s economy is even more science-intensive than Edison’s. And yet, while firms value the “golden eggs” produced by science, they have become less willing to “feed the goose” and invest in research. What changed? 

Several things.  First, the university sector has expanded. When GE started its research lab, there were very few university scientists who understood what materials would work well as filaments for incandescent bulbs, or why such bulbs were blackening. Universities were small and focused on education rather than research. Similarly, when DuPont began its research on polymers in the 1920s, the scientific understanding was rudimentary, and few universities could match the equipment and expertise of industrial laboratories such as Bell Labs (AT&T).  But as universities grew after WWII, supported by the GI Bill and federal research funding, the imperative to conduct frontier research insider the corporation diminished. Firms increasingly turned to universities to produce both new knowledge and the scientists and engineers who embody it.

Second, industrial research was greatly aided by federal support and procurement. Federal support for research and development, catalyzed by World War II and sustained throughout the Cold War, played an important role in supporting industrial research. As we document in our book — which draws on prior work, including by Dan Gross and Bhaven Sampat, and Sharon Belenzon and Larisa Cioaca — government procurement helped corporate laboratories develop technologies that later found commercial applications, from the laser to the microprocessor. With the end of the Cold War and declining interest in space exploration, government priorities shifted, and procurement support for industrial research declined in some sectors.

Third, the economic environment changed, particularly with growing international trade and global integration. Scientific discoveries “spilled over” more readily to competitors, reducing the returns to the companies that funded the research. Changing antitrust enforcement also played a role. Virtually all the companies with major industrial research efforts, from DuPont and GE, through IBM and Xerox, to Microsoft and Google, have been targeted by antitrust action.  Antitrust is a double-edged sword. It encourages dominant firms to seek new markets, but it can also reduce the profits that help sustain long-term research. Another facet is globalization, which encouraged firms to narrow their product portfolios and focus more on incremental innovation than exploration. A fourth possible facet is the changing role of financial markets. Some scholars, such as Elia Ferracuti and Rahul Vashishtha, also argue active hedge funds have pushed management away from long term research.

In the last few years, the pendulum seems to have swung back. There is growing government interest in rebuilding domestic industrial research capacity, both in the United States and elsewhere.

What have large companies lost by pulling back from directly investing in research? And does it matter for the broader economy?

For the most part, incumbents were acting rationally in pulling back. It was not in their private interest to sustain the investments in long-range research. But the broader economy has undoubtedly lost something.  As we argue in Inventing Prosperity, the university-startup ecosystem cannot always substitute for industrial research in established corporations, particularly in deep-tech sectors. 

Industrial research is large-scale, multidisciplinary, equipment-intensive, and mission-focused. Google’s machine translation, for example, required bringing together linguists, machine learning experts, software engineers and database designers, as well as hardware and chip designers to develop a new type of chip — the TPU.  It is difficult to conceive of a startup pulling this off. 

Moreover, research intensive incumbents are also essential for a well-functioning deep-tech ecosystem. Most deep-tech startups require alliances with suppliers and potential customers, and many ultimately “exit” by being acquired.  As Wes Cohen and Dan Levinthal argued, unless incumbents invest in scientific research internally, they lose the “absorptive capacity” to evaluate promising early-stage technologies. That raises the hurdle for deep-tech startups, forcing them into longer development cycles and increasing the capital required. As Roger Masclans’s research shows, it may also reduce the value startups capture when they are acquired. 

What is the role of policy to support innovation? The article calls for experimentation rather than a single fix — ARPA-style programs, public-private organizations. Where would you start, and in which sectors is the need most urgent?

The American innovation system has done well, but its triumphs have also exposed its frailties. There are no silver bullets. Still, there are several things worth trying, including encouraging industrial research by large companies, experimenting with new models for funding university-based science and early-stage-ventures, and strengthening the entrepreneurial ecosystem. 

Entrepreneurial startups are a key part of America’s innovation engine. But they need not just investment — they also need a deeper pool of industrial partners and potential acquirers. That is why protecting industrial research is a priority. We need to renew the social compact with firms that commit significant resources to advancing science. We should recognize the societal value created by businesses that push scientific frontiers. 

Universities will probably remain the main wellspring of research, funded by agencies like the NIH and NSF. Yet it is worth experimenting with new ways of funding and conducting research. Private initiatives — from the Rockefeller Foundation, the Howard Hughes Medical Institute and the Sloan Foundation to newer ones such as Schmidt Futures — can offer useful lessons. There is also much to learn from focused research organizations like BBN that bridge the gap between discovery and commercial application. Finally, much as Operation Warp Speed did for mRNA vaccines, the government can act as the “buyer of first resort," creating early demand for emerging technologies and attracting private investment.