YouTube’s secret weapon
Illustration: Christopher T. Fong/Protocol

YouTube’s secret weapon

Protocol Enterprise

Hello, and welcome to Protocol Enterprise! Today: why YouTube decided to invest millions of dollars in a custom video chip, Salesforce meets earnings expectations but cuts guidance and the fallout from the Twitter security whistleblower report continues.

Seeing the future with Argos

As much as some chip industry insiders insist that some version of Moore’s law is alive and well, Google engineering vice president Partha Ranganathan has helped make two big bets that the famous 1965 proclamation is, in fact, dead. The first was the company’s AI chips, or tensor processing units (TPUs), that the company unveiled several years ago.

The second is its little discussed — and some might say underappreciated — YouTube chip, code named Argos.

  • Argos evolved from the TPU project, after Google’s engineers figured out that they could develop more chips to attack other thorny computing problems beyond AI.
  • Video made for an obvious choice since YouTube ingests and streams a lot of it, and it accounts for the majority of internet traffic around the world.
  • “For something like transcoding, which is a very specific, high-intensity sort of workload, they can get an awful lot of bang for their buck there,” chip industry analyst Mike Feibus said.
  • After a 10-minute meeting with YouTube CEO Susan Wojcicki, the engineers under the leadership of YouTube vice president Scott Silver got the green light.

Named after the Greek mythological monster with lots of eyes, the chip helps YouTube convert videos to all the many different sizes and formats needed to play on thousands of devices.

  • Technically called an application-specific integrated circuit, or ASIC, the chip boasts impressive performance: a 20- to 33-fold increase versus the general-purpose chips YouTube had been using before.
  • The design itself is what Ranganathan called “hardware defined by software,” which meant they were able to iterate on the chip design much more quickly than a normal hardware process.
  • The chip is largely focused on transcoding video, which means taking the raw upload file and converting and compressing it into different formats so that everyone can watch it.
  • Each video on YouTube has as many as 15 separate copies in different formats, and about 500 hours of video are uploaded every minute.

YouTube’s decision to build its own video processors is part of a growing trend of tech companies deciding off-the-shelf solutions from the big chipmakers such as AMD, Intel and Nvidia are insufficient for their needs.

  • AWS has built its Graviton server chips, Microsoft is reportedly working on its own server processors, Facebook has a chip unit, just to name a few.
  • On one hand, it’s not about cost. Jay Goldberg, principal at D2D Advisory, told us that building custom silicon is about creating a strategic advantage.
  • With its YouTube chips, Google is attempting to offer a better viewing experience — by converting more videos to formats that use less data, it can get them to load faster and potentially look better.
  • On the other hand, it kind of is about the money. Silver said that the company definitely likes to save money, but what it really allows is for the company to better allocate its resources.

— Max A. Cherney (email | twitter)

Sponsored content from DataRobot

DataRobot's AI Cloud for Financial Services Unlocks the Art of the Possible: DataRobot continues to attract clients in financial services who want to de-risk their AI investments and rapidly scale AI to almost every part of their operations, resulting in improved productivity and higher customer satisfaction.

Read more from DataRobot

A little quieter in the second half

Surprise! Salesforce is still growing, but announced Wednesday that it won’t hit its previous expectations for the second half of 2022.

The SaaS giant’s revenue grew 26% year-over-over to $7.72 billion for the quarter. But the company also lowered guidance by a hair, down from previous forecasts of $31.7 billion to just $30.9 billion, as enterprise tech continues to look for signs that a consumer-spending slowdown is starting to trickle into enterprise spending.

Salesforce attributed the lowered guidance to a mixture of currency headwinds and subsequent shifts in customer buying patterns. “The foreign exchange environment is obviously unprecedented,” co-CEO Marc Benioff said during the company’s earnings call.

But Salesforce’s performance across all its clouds is still strong, and the company also announced its first-ever share repurchase of up to $10 billion.

The company’s Slack bet also seems to be paying off, with the messaging app propelling Salesforce’s platform business to more than 50% revenue growth for the quarter. Salesforce’s data business, which includes Tableau and Mulesoft, had the lowest revenue growth of all the clouds, but still hit $1 billion, meaning all of Salesforce’s clouds passed the $1 billion revenue mark for the quarter.

Salesforce is still laser-focused on growth, but is being careful with hiring and still doesn't have any plans for major M&A, although Benioff teased the idea during the earnings call. “We have found a great company … and that company is Salesforce,” he said. He has a point.

— Aisha Counts (email | twitter)

'This is the job'

On Tuesday, many in the cybersecurity community threw their support behind Peiter "Mudge" Zatko, following his whistleblower complaint that accused Twitter of ignoring an array of major cybersecurity problems (among other issues). Zatko, who'd formerly been security head for the social media platform, had pointed out numerous shortcomings from out-of-date software in Twitter data centers, to nonexistent monitoring for insider threats, to lax controls around employee access to key systems.

Today, however, some have been airing a contrary cybersecurity point of view on the subject.

"I do not know of a single working CISO who is not terrified of at least a dozen major issues that require immediate attention in the company — and who are frustrated with senior management," wrote Edward Amoroso, the former longtime chief security officer at AT&T, in a post on LinkedIn. "This is the job."

The CEO of virtual CISO service provider SideChannel, Brian Haugli, offered similar thoughts while speaking with me today. As a CISO, “your job is to unearth all of the stuff that could potentially be damaging — but then help realistically put it into expectations that [the executive team] can adjust to," he told me. "I don't think there's a single organization out there, even in the Fortune 500, that doesn't have similar problems to what Twitter is going through."

Regardless of what you think on these perspectives (and Amoroso’s take has prompted plenty of pushback, in the comments on his post and, of course, on Twitter), no one would argue that the apparent indifference of other executives to Zatko's findings is justified.

It's just that rather than Twitter being extraordinary, “this is the plight of security inside of organizations,” as Haugli put it to me. “You're constantly balancing what the business wants to be able to do, which is move fast, and what security wants."

— Kyle Alspach (email | twitter)

Around the enterprise

Nvidia’s second-quarter earnings results came in a whopping $1 billion short of revenue expectations thanks to consumer spending weakness, sending its stock plunging after hours despite solid results for its data center group.

Snowflake revenue grew 83% in its second quarter to $497 million,well above Wall Street’s expectations for the cloud data company.

The U.S. Army wants to see “much more rapid movement to the cloud” over the next 12 months, according to C4ISRNet, as the military-wide JWCC cloud contract continues to slog through the evaluation process.

Sponsored content from DataRobot

DataRobot's AI Cloud for Financial Services Unlocks the Art of the Possible: Banks need to secure a competitive advantage in an increasingly tight race to harness best-in-breed technology. Decision makers need to not just plan a future-ready strategy, but also recognize the value of AI that could boost not just their performance in-house but also their reputation among their global customers.

Read more from DataRobot

Thanks for reading — see you tomorrow!

Recent Issues