‘Raise More Money Than You Think’: Eye on A.I.

John Hennessy, chairman of Google-parent company Alphabet, has words of wisdom for startups creating computer chips for turbocharging artificial intelligence: “Raise more money than you think.” 

Hennessy, one of Silicon Valley’s top technologists, should know. He was formerly Stanford University’s president and the co-founder of semiconductor company Mips Computer Systems, which went public in 1989 before it was eventually acquired.

Hennessy’s advice is targeted at the tsunami of startups trying to become the next Intel or Nvidia by developing technology that can reduce the energy and time required to train A.I. systems or more efficiently perform A.I.-related tasks on smartphones. One or more of these companies could end up creating the foundation for the next generation of corporate A.I. projects. 

But it won’t be easy—or cheap. Creating new chips involves a lot of research, fabrication, and trial and error.

“Hardware always takes longer than you think,” Hennessy said last week at a conference about A.I and semiconductors in Mountain View, Calif.

Karl Freund, with analyst firm Moor Insights & Strategy, said “We got about 130 companies now developing A.I. silicon.” Among those will only be a few winners, he predicted.

“How many do you think will survive?” Freund asked the audience of techies. “I don’t know, but it won’t be 130.”

Creating A.I. chips is just the first step. Selling them to the cloud-computing giants like Amazon, Microsoft, and Google—the obvious customers for the technology—is its own challenge because they’re developing their own A.I. chips.

Freund speculated that one reason these cloud companies are doing so is to give them leverage when negotiating deals with other companies. As he explained, the giants are showing the A.I. startups the capabilities of their own internal chip projects, and then asking the startups if they can do better. 

“If you can’t, you are done,” Freund said. 

Lingjie Xu, the director of applied A.I. architecture at Chinese e-commerce and cloud giant Alibaba, said during a panel that his business has its own set of performance benchmarks for testing other company’s A.I. chips before buying. The decision isn’t based on hype, but rather hard numbers.

“We want to see something real,” Xu said. 

Jonathan Vanian 

Subscribe to Eye on A.I. here.


Google gets healthier. Google officially absorbed the health division of its A.I. research subsidiary DeepMind ten months after announcing it would do so. Dr. Dominic King, Google Health’s U.K. site lead, explained the ten-month wait in a blog post, writing that “Health data is sensitive, and we gave proper time and care to make sure that we had the full consent and cooperation of our partners.” The BBC noted “When the restructuring was announced last year, DeepMind was accused of breaking a promise not to hand over any NHS data to the American search giant,” referring to data from the U.K.’s National Health Service.

China’s not alone. Although China is the world’s number one user of facial recognition and other surveillance-related technologies, the Carnegie Endowment for International Peace said that other countries like the U.S., Brazil, and Singapore are increasingly using the tech, The Wall Street Journal reported. The report said that Huawei is the dominant supplier of surveillance technologies and has sold products in 50 countries. “No other company comes close,” the report added.

Facial-recognition’s cosmetic-surgery problem. The South China Morning Post reported on a woman in eastern China who claimed that recent cosmetic surgery to her nose caused facial-recognition software to fail at identifying her. The report, which cited a local television news story, said that the woman “discovered she had been logged out of the online shopping and payment gateways she used because the secure identification process, backed by facial recognition technology, simply did not know who she was.”

The robotic boom. The International Federation of Robotics said that the global market for industrial robots reached a record $16.5 billion in 2018. The biggest buyers of robots are China, Japan, the Republic of Korea, the U.S., and Germany, the report said. Additionally, the report said “Robot installations in the United States increased for the eighth year in a row to a new peak in 2018 and reached about 40,300 units.”


Wired examined the difficulty of A.I. researchers reproducing other researchers’ work, and how some researchers subtly tweak their A.I. systems to validate the results of others—a laborious and expensive process. Anna Rogers, a University of Massachusetts machine-learning researcher asks, “Is that even research anymore?” She tells Wired: “It’s not clear if you’re demonstrating the superiority of your model or your budget.”


Goldman Sachs hired Marco Argenti to be the financial firm’s partner and co-chief information officer. Argenti was previously the vice president of technology for Amazon Web Services.

Charter Communications picked Stephanie Mitchko to be the company’s executive vice president and chief technology officer. Mitchko was previously the CTO and chief operating officer of Cadent.


Deep Learning to track the progression of blindness. Researchers from Roche and its Genentech subsidiarpublished a paper in Nature Digital Medicine about using deep learning to track the progression of diabetic retinopathy—a diabetic eye disease that can lead to blindness—in patients. The researchers write that their A.I. system “would enable early identification of patients at highest risk of vision loss, allowing timely referral to retina specialists and potential initiation of treatment before irreversible vision loss occurs.”

Deep Learning to crunch tabular data. Researchers from Russian search giant Yandex published a paper about creating a deep learning system that can analyze tabular data, or data that’s typically stored in spreadsheets and Excel files that contain tables. The researchers said that while deep learning has proven effective in crunching data that’s either images or text, “Overall, at the moment, there is no dominant deep learning solution for tabular data problems, and we aim to reduce this gap by our paper.”


Venture Firm Data Collective Raises $725 Million to Invest in ‘Deep Tech’– By Polina Marinova

50 Trillion Calculations per Second in the Palm of Your Hand– By Aaron Pressman

ImageNet Roulette Highlights Bias in A.I. See For Yourself– By Alyssa Newcomb


Google’s quantum achievement. Google researchers claimed that they achieved “quantum supremacy,” a milestone in which the team said in a research paper that they used a quantum computer to calculate a problem that would have taken traditional computers thousands of years to do, the Financial Times first reported.Fortune’s Robert Hackett obtained a copy of the research paper, and wrote that “researchers estimate that performing the same experiment on a Google Cloud server would take 50 trillion hours—too long to be feasible. On the quantum processor, it took only 30 seconds, they said.” The research is significant for the field of quantum computing, which could one day lead to more capable A.I. systems that function on tremendous amounts of computing power. But as Fortune’sJeremy Khan explained, “’quantum supremacy’ does not mean quantum computers have yet arrived in the sense that they will soon replace the conventional computers that power our lives.”

Subscribe to Well Adjusted, our newsletter full of simple strategies to work smarter and live better, from the Fortune Well team. Sign up today.

Read More

CEO DailyCFO DailyBroadsheetData SheetTerm Sheet