Amazon just announced that it would add more than 100,000 fulltime jobs in the US over the next 18 months.  That’s fantastic.

But even before that announcement, it turns out that Amazon was the fastest US company to reach 300,000 jobs. That’s based on a new paper we are releasing today, called “A Historical Perspective on Tech Job Growth.” Here’s the  beginning of the paper.

General Motors reached 300,000 employees in 1941, 32 years after its 1909 founding. American Telephone & Telegraph hit the same milestone in 1926, 27 years after its 1899 absorption of the local Bell systems. And Walmart went over 300,000 associates in its 1991 fiscal year, its 21st year as a public company.

But in 2016, Amazon became the fastest American company to reach 300,000 workers, hitting that mark in its 20th year as a public company. This figure, which does not include contractors or temporary workers, represents an average employment growth rate of roughly 30% per year. That’s an amazing growth rate.

But Amazon is not alone. In fact, tech giants such as Google, Apple, Facebook and Microsoft are adding jobs as fast or faster than the great job-producing companies of the past, like GM, AT&T, Walmart, IBM, GE, US Steel, and Bethlehem Steel.

Consider this: Twenty years after its 1892 founding, General Electric had 41,000 employees. Google beat that mark in 2012, only 8 years after its 2004 initial public offering.

Read the paper here.

Everyone has seen this chart, or something similar, recently. It tracks the production of the manufacturing sector over the past twenty years. Looks pretty good, doesn’t it? Since 1996 manufacturing industrial production has risen almost 40%, despite intense global competition.

But now let’s add another line to the chart–this time, manufacturing industrial production after we remove tech manufacturing.  That’s the blue line in the chart below.

What’s changed? It turns out that once we have taken out tech manufacturing–computers, communication equipment, and semiconductors–the output of the rest of manufacturing has only risen by 5% over the past 20 years, and is well below its 2007 peak.

Over the same period, the output of tech manufacturing has risen by a factor of 20, or 2000%, as least according to Federal Reserve Board estimates (and they have some of the best economists around).

The same divergence shows up in productivity growth. I do a simple calculation, dividing manufacturing industrial production by the number of fulltime equivalent workers in manufacturing, and then I do the same for non-tech manufacturing.  Over this twenty year period, it looks like this:

The top line–the manufacturing industrial production per worker–shows that productivity in manufacturing has almost doubled over the past twenty years. But roughly half of that gain is due to the tech sector. Once we remove  tech production from manufacturing, the 20-year productivity gain falls from 93% to 45%. And while that 45% gain in productivity is nothing to sneeze at, it amounts to only 1.9% annually.

Let’s break down manufacturing labor productivity growth  by industry. These figures come from the BLS.

We see that computer and electronic products are far and away the best performing manufacturing industry, in terms of productivity growth. Then it falls off quickly to transportation equipment (motor vehicles and aircraft), which is the main user of robots. Then productivity growth across the rest of manufacturing ranges from mediocre to downright uninspiring.

In a post-election  post we showed how American workers are being failed by the physical industries, which had dramatically underperformed the digital industries across a wide range of metrics, including productivity, compensation, and job growth.  This sharp and growing economic gap between the Digital Nation and the Physical Nation had profound political consequences, since the Physical Nation is still triple the size of the Digital Nation.

In this post we look at the economy from the consumer side.

1. Per capita real consumption of physical goods and services, outside of housing and health care,  has grown by only 10% since 2000, or an average of 0.6% annually.  By comparison, per capita real consumption of digital goods and services has skyrocketed, growing by 64%, or  3.4% annually.

2. In other words, to the extent that consumer living standards have been rising, the gains have been mainly driven by digital goods and services. For example, per capita real consumption of communication services–including internet, wireless, and cable–is up by 60% since 2000. This calculation is based only on government data, without taking into effect possible unmeasured gains in consumer surplus.

3..  By contrast, in the physical sector  per capita real purchases of motor vehicles and parts has risen by only 6% since 2000, or only 0.4% per year.  Similarly, per capita real purchases  of food and beverages for  home use has risen by only 4%, or 0.3% annually. It should be noted that food processing is a manufacturing industry which has shown no labor productivity growth in the past ten years, .

4. The Digital Nation is mainly characterized by falling or flat consumer prices, while the Physical  Nation is mainly characterized by rising consumer prices.  In the digital sector, consumers spend less to get more, while in the physical sector, consumers spend more to get less. The price of communication services is basically flat since 2000, allowing even low and middle-income households to participate in the App Economy.

5.On the other hand, low- and middle-class households have battled with the rising price of food,  housing, healthcare, and transportation. The nominal incomes of middle class households (middle quintile) have risen 34% since 2000. But food prices have risen 41%; the price of housing, fuels and utilities (except phone and cable) is up 50%; and the price of food services and accommodations is up 50%. And while the price of new motor vehicles has been relatively flat since 2000, the price of gasoline is up 65%; the price of auto maintenance and repair is up 53%, and the price of taxicabs is up a stunning 74%.

6. The rate of inflation has actually accelerated in the Physical Nation. As the chart above shows, the price of physical goods and services, outside of health and housing, rose at a 1.9% rate in the period 2000-2015, compared to a 1.3% rate in the 1990s.

7. The digital sector has performed extremely well for consumers over the past 15 years. From the policy and political perspective, our focus should be on why prices rose so fast in the physical sector since 2000 compared to the previous decade. The price increases for physical goods and services helps explain why low- and middle-class households have been struggling, and why they are so angry.

Google’s decision to pause its roll-out of Google Fiber to new cities is an important data point in our understanding of both the economics of telecom and the economics of innovation. The announcement said that:

In terms of our existing footprint, in the cities where we’ve launched or are under construction, our work will continue. For most of our “potential Fiber cities” — those where we’ve been in exploratory discussions — we’re going to pause our operations and offices while we refine our approaches.

Google should be applauded for its willingness to take a chance on laying fiber to the home—such big bets are the only way that innovation happens. Too many companies around the world choose to play it safe and avoid putting money into “moonshots.”

However, Google’s move emphasizes that telecom infrastructure investments cost big money and are risky. A company with one of the deepest pockets in the world—and a long-established willingness to innovate—has been forced to reassess its strategy. Advocates such as Susan Crawford have claimed in the past that the roll-out of Google Fiber proves that it is “cost-effective to install and sell fiber connectivity.” But by the same token, Google’s pause confirms that gigabit networks to the home are expensive, as incumbent providers have been saying, and not yet a winning proposition for most consumers.

Regulators need to remember that telecom infrastructure investments are not only large, but carry significant technological and business risk. Major telecom providers such as AT&T, Verizon, and Comcast have been at or near the top of our “Investment Heroes” list year after year precisely because they are willing to take those risks. And the risks will only get larger. No one is sure yet the best way to connect fiber backbones to homes and businesses, and what the business model will look like. For example, will fixed 5G networks be the best way to bridge the “last mile” to the home? It’s tough to say, given that the technical standards have not yet been set. A wrong bet could lead to billions of dollars in losses.

But there’s a broader point as well about size and innovation. Economists have long debated whether large companies or small companies contribute more to innovation. As a general rule, the answer is “it depends.” As we wrote in our policy memo, “Scale and Innovation in Today’s Economy”:

the current U.S. economy is dealing with a particular set of conditions that will make scale a positive influence on innovation. First, economic and job growth today are increasingly driven by large-scale innovation ecosystems, ….These ecosystems require management by a core company or companies with the resources and scale to provide leadership and technological direction. This task typically cannot be handled by a small company or startup.

Second, globalization puts more of a premium on size than ever before. A company that looks large in the context of the domestic economy may be relatively small in the context of the global economy. In order to capture the fruits of innovation, U.S. companies have to have the resources to stand against foreign competition, much of which may be state supported.

Finally, the U.S. faces a set of enormous challenges in reforming large-scale integrated systems such as health, energy, and education. Conventional venture-backed startups don’t have the resources to tackle these mammoth problems. Only large firms have the staying power and the scale to potentially implement systemic innovations in these industries.

These considerations raise certain questions about how regulators—in the US, Europe, and Japan—should think about the role of scale. Scale by itself should not be viewed an impediment to innovation. Small firms are essential to innovation, but big firms have an essential role to play as well.