G.E. Makes a Sharp ‘Pivot’ on Digital

But the rethinking of GE Digital, with its future still under review by Mr. Flannery, points to the difficulty of producing modern software for industrial businesses as they adopt digital technology.

No one disputes the overarching vision of the so-called industrial internet of things — which includes low-cost sensors and a flood of data and clever software that should deliver insights to cut costs, conserve fuel and design better products, faster. But the company greatly underestimated the challenges of creating all the software needed to achieve that grand vision, said analysts and former G.E. managers.

G.E.’s technical prowess, they said, lies in designing and manufacturing big machines like power-plant turbines, jet engines and medical-imaging equipment. Its traditional software skills have been in the specialized programs that control the machines and factory operations. GE Digital was a striking departure into cloud-based internet software, data analytics and artificial intelligence tools like machine learning.

“G.E. reached too far outside its expertise and too fast,” said Steven Winoker, an analyst at UBS. “And it became a financial black hole.”

Just how much G.E. has invested in its digital initiatives is uncertain, but it has been several billion dollars. In an article last year in the Harvard Business Review, Mr. Immelt wrote that in 2016 “we put about $4 billion into developing analytics software and machine learning capabilities.”


William Ruh, who leads GE Digital, described the changes at the business as “a pivot” rather than a retreat.

Brad Swonetz for The New York Times

G.E.’s digital effort dates to 2011, when Mr. Immelt recruited William Ruh, an executive at Cisco Systems. Later that year, the company set up a center in San Ramon, Calif., east of San Francisco. At the time, Mr. Ruh announced big plans to hire as many as 400 software engineers to write code for the industrial internet.

The San Ramon work force swelled sharply to 1,400 in the summer of 2016, and to a peak of 2,000 last summer — before coming down to about 1,800, after a round of cuts earlier this year.

During the buildup, G.E. recruited veteran software managers who had worked at leading tech companies, including Google, Microsoft and Apple. With the cuts at GE Digital and its narrowed focus, some of them have departed.

In its early days, the San Ramon center concentrated on writing data analysis and modeling software for applications like predicting when a gas turbine or jet engine would need maintenance, before the machine failed. In 2013, G.E. called that software Predix.

The product plan later broadened to include a cloud-based software platform for handling all kinds of sensor and machine data, and secure communications from the factory floor to data centers. G.E. even built its own data centers.

The software platform under development was often referred to as an operating system for the industrial internet. When the product concept expanded, so did the Predix brand.

Today, according to former G.E. engineers, Predix has been pared back to mainly a set of software tools to help write applications rather than being connected to layers of code for automating data analysis.

The priority at GE Digital is now on selling products for specific industrial applications, sold as offerings in the “Predix portfolio,” and tailored for G.E.’s roster of existing industrial customers. Less emphasis is being put on all-purpose software for the wider industrial world.

In an interview, Mr. Ruh described the change at GE Digital as “a pivot” rather than a retreat. “We’re still 1,000 percent behind our Predix portfolio,” he said.

Two key products in that portfolio were acquired. In September 2016, GE Digital bought Meridium, a maker of equipment-tracking software, for $495 million. Two months later, it bought ServiceMax, whose software is used to manage industrial field service workers, for $915 million.


A mural at the San Ramon offices. The priority at GE Digital is now on selling products for specific industrial applications, with less emphasis on all-purpose software for the wider industrial world.

Damien Maloney for The New York Times

The current strategy, Mr. Ruh said, “is about industrial apps” like those. It is a measured, step-by-step approach but one, Mr. Ruh said, that capitalizes on G.E.’s strengths — its industry knowledge and deep customer relationships — and is lifting sales.

GE Digital remains a large player in an increasingly crowded field of companies offering industrial internet software of various kinds, in different market niches. The entrants include the big cloud suppliers like Amazon, Microsoft and Google; major business software companies like Oracle, SAP, IBM and SAS Institute; G.E.’s industrial peers such as Siemens, Honeywell and ABB; and start-ups like C3 IoT, Uptake and FogHorn Systems.

The opportunity for G.E., analysts said, centers on its longstanding relationships with customers and selling to them. And so far, only 8 percent of its industrial customers are using Predix portfolio products, the company said.

The New York Power Authority, the nation’s largest state-owned utility, is one of them. The utility is working with GE Digital to build apps that improve the efficiency of its power generation and distribution network. In pilot projects, the partnership has saved or avoided $3 million in costs, said Gil Quiniones, chief executive of the state power authority. The goal is $500 million in savings over the next decade.

But for a technology project to help its customers reduce energy consumption, the state authority went with a start-up, C3 IoT.

Brian Hurst, chief analytics officer at Exelon Utilities, a large electric utility corporation based in Chicago, is also working on early-stage projects with GE Digital. Mr. Hurst said the progress was encouraging and he saw no evidence that the cuts at GE Digital had affected their joint work.

But it is something he is watching closely. “The technology is moving so fast,” Mr. Hurst said. “Today’s players may not be tomorrow’s players.”

GE Digital, Mr. Ruh insisted, is building a business for the long term. Revenue from the Predix portfolio products reached $550 million last year, he said, and sales are growing rapidly. GE Digital’s total revenue of $4 billion also includes all of G.E.’s traditional industrial software, and the unit’s global work force is more than 4,000.

“Our approach is heads-down,” Mr. Ruh said. “We’re going to show you success.”

For his part, Mr. Flannery has consistently declared his support for G.E.’s digital-industrial vision — but that does not guarantee the future for GE Digital. Mr. Flannery is selling off businesses to streamline the company. The lighting and railway locomotive divisions are up for sale, and others could follow.

A smaller G.E. might not need a stand-alone digital unit. Some analysts recommend putting all the software in the industrial divisions, like power and aviation, closer to customers.

“San Ramon shouldn’t exist,” said Scott Davis, chief executive of Melius Research, an independent financial analysis firm.

Continue reading the main story

Computer Chip Visionaries Win Turing Award

Today, more than 99 percent of all new chips use the RISC architecture, according to the association.

“This is the one fundamental idea that has been sustained over the last several decades of chip design,” said Dave Ditzel, a chip industry veteran who studied with Mr. Patterson at Berkeley. Mr. Ditzel helped popularize many of the same ideas and is now building a new RISC chip at a start-up called Esperanto.


Google has developed new chips that were specifically designed for artificial intelligence applications.


Mr. Patterson and Mr. Hennessy were interested in simpler chips because they ran faster, consumed less power, made life easier for chip designers and allowed machines to evolve at a faster rate. In the mid-1980s, new RISC chips emerged from two Silicon Valley start-ups, Sun Microsystems and MIPS Technologies, becoming the standard for the computer workstations and servers that underpinned big corporate operations.

Those processors were eventually eclipsed by chips from Intel, which put its considerable muscle behind a competing design. But as computing expanded into smartphones, tablets, and other small devices — where power and space are at a premium — more and more chips used designs from a British company called ARM, short for Advanced RISC Machine.

As a book written by the two researchers in 1989, Computer Architecture: A Quantitative Approach, became the standard text for chip design, even Intel took a partial step toward the RISC idea. Its chips continued to use their own complex way of talking to a computer’s software, but started to use some aspects of RISC.

Intel chips still drive the data centers that power the internet. But as these chips approach their physical limits, internet giants like Google, Facebook, and Amazon are pushing tasks onto a wide range of simpler processors that consume much less power, sparking a renaissance in chip design.

“Complexity is even more of an enemy than it was before,” Mr. Ditzel said. “We have to design differently.”

Mr. Patterson and Mr. Hennessy are at the heart of this change. Their book is now in its sixth edition. Mr. Hennessy is on the board of directors of Google’s parent company, Alphabet, after serving as the president of Stanford for 16 years. And Mr. Patterson works in the Google research lab that is designing low-powered chips specifically for artificial intelligence.

As time goes on, the world could move even more toward the RISC way of doing things, thanks to an organization called the RISC-V Foundation, which has published a chip architecture that anyone can use for free, said industry veteran Dennis Allison. The organization was founded by Dave Patterson and others from Berkeley.

“I expect this to play a vital role in the future,” Mr. Patterson said. “And the architecture is not that different from what John and I described back in 1980.”

Continue reading the main story