Back to the data center

Over the last several years, cloud providers have been urging their customers to adopt machine learning services to unlock new capabilities for their applications. The good news for the Big Three is that customers are really starting to show an interest in those services. The bad news? They don’t always want to run them on the cloud.

That’s the gist of a new report from Kate Kaye we published this morning , as Kate takes a well-deserved vacation from the daily news grind. Companies that started machine learning trials are increasingly pulling them back on-premises.

• “We still have a ton of customers who want to go on a cloud migration, but we're definitely now seeing — at least in the past year or so — a lot more customers who want to repatriate workloads back onto on-premise because of cost,” said Thomas Robinson, vice president of strategic partnerships and corporate development at MLOps platform company Domino Data Lab.
• The on-prem trend is growing among big box and grocery retailers that need to feed product, distribution and store-specific data into large machine learning models for inventory predictions, said Vijay Raghavendra, chief technology officer at SymphonyAI, which works with grocery chain Albertsons.
• “It’s a cost that a lot of companies are now looking at, saying, can I bring my training in-house so that I have more control on the cost of training, because if you let engineers train on a bank of GPUs in a public cloud service, it can get very expensive, very quickly,” said Danny Lange, vice president of AI and machine learning at gaming and automotive AI company Unity Technologies.

Cost isn’t the only consideration.

• Victor Thu, president at Datatron, said retailers or fast-food chains with area-specific machine learning models — used to localize delivery logistics or optimize store inventory — would rather run ML inference workloads in their own servers inside their stores, rather than passing data back and forth to run the models in the cloud.
• “Often the decision to operationalize a model on-prem or in the cloud has largely been a question of latency and security dictated by where the data is being generated or where the model results are being consumed,” Robinson said.

Cloud-based machine learning services aren’t going anywhere, but like compute services, in the future companies will probably split the difference between the cloud and their own machines more equally than once thought.

• “There's always been a bit of a pendulum effect going on,” Lange said. “Everybody goes to the cloud, then they sort of try to move back a bit. I think it's about finding the right balance.”

Read the full report here .

— Tom Krazit ( email | twitter )

SPONSORED CONTENT FROM ALIBABA

How global ecommerce benefits American workers and the U.S. economy: Alibaba — a leading global ecommerce company — is a particularly powerful engine in helping American businesses of every size sell goods to more than 1 billion consumers on its digital marketplaces in China. In 2020, U.S. companies completed more than $54 billion of sales to consumers in China through Alibaba’s online platforms.

Read more from Alibaba

14 and up

The U.S. has undertaken years-long efforts to prevent China from acquiring the tech necessary to build the most-advanced chips. Last week American officials made a further offensive move: Chip manufacturing equipment-makers will now need an export license to sell tech for 14-nanometer equipment or anything more advanced, according to executive comments on earnings calls.

For the tool-makers themselves, the policy change is a red alert moment. One of the largest equipment vendors, Lam Research, disclosed in its latest quarterly filing with the SEC that China accounted for 31% of revenue in the March quarter; KLA reported 31% of its business came from China; and Applied Materials said 34% of sales were from China .

KLA, Applied and Lam Research declined to comment.

What’s not clear is precisely how the new restrictions will affect the tool-makers. Today, chipmaking equipment operates like a car; in order to get the best performance out of the machine, owners need to get it serviced, change the tires and so on. Similarly, the tool-makers have built large services businesses around their equipment sales, and it’s not clear whether they will be able to continue to do so.

Though the nanometer label isn’t as relevant as it once was — it’s essentially just marketing terminology today, which in turn makes the export controls harder to define — restricting the sale of chipmaking tech that’s nearly a decade old marks a significant shift in U.S. policy.

With prior export restrictions, American officials attempted to choke off China’s access to the most-advanced chipmaking equipment. Now the administration appears to hurt China’s ability by curbing access to tech it had already been able to buy.

Financial corner

Acronis raised $250 million from institutional investors, including BlackRock , that valued the data security company at $3.5 billion.

Cordial raised a $50 million series C funding round to expand its personalized-messaging services for customer experience improvement projects.

No-code app platform Retool raised a $45 million round from existing investors that valued the company at $3.2 billion.

Around the enterprise

Oracle laid off several employees in its customer experience division, which has struggled in recent months according to Bloomberg .

Illuminate Education, which made news earlier this year after its educational software service was down for almost a week, just suffered a cyberattack that compromised the personal data of more than 1 million students.

SPONSORED CONTENT FROM ALIBABA

How global ecommerce benefits American workers and the U.S. economy: Using economic multipliers published by the U.S. Bureau of Economic Analysis, NDP estimates that the ripple effect of this Alibaba-fueled consumption in 2020 supported more than 256,000 U.S. jobs and $21 billion in wages. These American sales to Chinese consumers also added $39 billion to U.S. GDP.

Read more from Alibaba