05.16.2025

ForwardThinking 🤔

Cloud repatriation revisited

It seems like every couple of years, we hear about companies making the move from the cloud back on prem, or at least to co-location facilities where they are content to manage their own hardware and reduce their cloud bills.

There was a lot of talk about this in April 2023, which prompted me to explore the subject. My conclusion at the time: “For most companies, the cloud remains an essential part of their technology strategy, and few are going back to running their own data centers anytime soon. But for certain workloads, it might make sense to move them to managed private clouds,” I wrote.

Now, the repatriation talk is back, this time with an AI twist. A recent column by my friend David Linthicum in InfoWorld, called ‘Cloud repatriation hits its stride,’ in particular, caught my eye. Linthicum has spent years working with large companies on cloud projects, first at Deloitte and now as an independent consultant.

When someone with his experience points out a shift, it’s worth paying attention. As he put it, “2025 feels different. Repatriation—once a quiet undercurrent—has surged into the mainstream. The driving force behind this movement? Artificial intelligence.” But just how widespread is this shift?

Back to the data center

We’re definitely seeing some high profile examples of this type of transition happening. 37signals CTO David Heinemeier Hansson has been a vocal supporter of cloud repatriation in recent years. Just this week, he published a LinkedIn post touting his company’s move away from Amazon S3, shifting petabytes of storage back in-house using Pure Storage gear. The company expects to cut its annual storage costs from $1.5 million to $200,000, a hefty savings if the numbers hold up.

“The savings will rack up pretty quick from there, once we've amortized the $1.5m outlay on the combined 18 petabytes of capacity we've bought from Pure across our two data centers. After that, the yearly cost will be below $200,000. Much easier to swallow than $1.5m/year,” he wrote.

Data center with rows of stainless steel racks.

Photo by İsmail Enes Ayhan on Unsplash

He’s not alone. I recently wrote about Akamai moving almost all of its substantial infrastructure out of the cloud, saving $100 million a year along the way. CIO Kate Prouty challenged her engineering team to question every cloud workload. “And there were places that it became very obvious where we had workloads that hadn't been evaluated or looked at since they were deployed,” she said. In many cases, they were moved back. But two examples, no matter how strong, don’t necessarily make a trend.

Trend or no trend

If the shift away from the cloud is truly substantial, you would think it would start to be reflected in the revenue numbers. Yet there’s little evidence of a sharp drop in business. In fact, for the most recent quarterly earnings cycle, cloud infrastructure revenue totaled $94 billion for the quarter, up 23% year-over-year, per data from Synergy Research. That doesn’t feel like an industry going in the wrong direction.

John Dinsdale, chief analyst and research director at Synergy, says he doesn’t see much evidence to suggest a shift away from the cloud. “There will always be examples of cases where some form of repatriation makes sense, either as companies rebalance after pushing too much onto the public cloud, or just learning from experience and improving the mix of cloud versus on-prem workloads. But this is like a trickle in one direction versus a torrent in the other,” Dinsdale told FastForward.

Cloud infrastructure market size by quarter from Q1 2020 until Q1 2025, growing from around $30 billion in 2020 to over $90 billion today

Chart courtesy of Synergy Research

As for AI workloads, he sees cost being an impediment for a lot of companies. “Will AI shift that balance? In some cases, yes, but the hardware cost involved is a big step up from traditional servers and that represents an additional barrier to those considering running their own infrastructure,” he said.

Synergy also reported growth of 140-160% in GenAI-specific services last quarter in the cloud. So far, it seems that AI is driving revenue growth for the cloud providers, not diminishing it.

The on-prem argument

It’s worth noting, however, that hardware vendors have been pushing AI PCs, specifically designed to run large language models locally, but sales of those devices have been slow to take off so far. Companies like Dell, which clearly would benefit from a shift on-premises, have stated that it’s cheaper and more efficient to run AI workloads locally, but the jury is still out on the cost question. Forrester, for instance, says it depends on the task.

“For many organizations, it will be necessary to leverage different providers for each of the workloads they are tackling. For example, when it comes to data management, an organization may choose a vendor with a strong on-premises solution, but for inferencing, they may select a large cloud services provider,” Forrester’s Mike Gualtieri wrote in a post last year.

"For many organizations, it will be necessary to leverage different providers for each of the workloads they are tackling."

Although companies like Akamai and 37signals are fully committed to moving out of the cloud, Linthicum ultimately sees most organizations taking a more balanced approach. “Repatriation doesn’t signal the end of cloud, but rather the evolution toward a more pragmatic, hybrid model,” he wrote. With hybrid, you can put your workload in the place that makes most sense for your organization, based on factors like cost, security and governance.

Of course, high-profile cloud repatriation examples make for good headlines, but overall the cloud continues to play a central role for most organizations, as ongoing cloud revenue growth attests.

~Ron

What's new on the blog 📰

For Salesforce’s head of security, it’s all about trust

Salesforce chief trust officer Brad Arkin not only does all of the tasks that such a role demands at a large software company like Salesforce, he also uses agents to help him and his staff work more efficiently.

That includes asking an agent for status updates instead of interrupting his busy security analysts. The agent reviews the case notes, builds timelines and summarizes progress.

“It’s like talking to somebody who’s sitting in the SOC (security operations center), who's working on the case, except me interacting with the agent isn’t distracting my staff from something they should be doing."

Read the full article >>

Photo by Ron Miller

IBM sees an AI future with smaller, more focused models in the enterprise

In April 2023, while still at TechCrunch, I had a notion that smaller language models might be better suited to the enterprise. Two years later that idea is actually beginning to gain popularity, as companies look for ways to take advantage of their own data and control ballooning AI costs. It was also a key point that IBM CEO Arvind Krishna articulated at the IBM Think keynote in Boston earlier this month.

Read the full article >>

Palo Alto Networks tries to stay agile by maintaining a startup mentality

I recently interviewed Palo Alto Networks chief product officer, Lee Klarich, who has worked at the company since its earliest days. I was particularly interested in the 22 acquisitions Palo Alto has made over the years, and how the company decides when to build and when to buy. Often it comes down to buying a startup when it has a unique expertise that the company lacks in-house.

"If you look across the different acquisitions, you'll see we've done more acquisitions in new areas like cloud security, than we have in, say, network security, which I think makes sense given our roots,” he said.

Coincidentally, I got a perfect buy example at the end of last month when his firm announced its intent to acquire boldstart portfolio company, Protect AI.

Read the full article >>

Palo Alto Networks chief product officer Lee Klarich, speaking at an event during RSA on April 28, 2025 in front of a sign bearing the company's name.

Photo by Ron Miller

Enterprises continue to struggle to find AI success

While the AI hype continues unabated, the data doesn't lie, and study after study is showing that enterprises continue to struggle to implement AI projects at scale successfully. The number has been stuck stubbornly at around 25% success rate for months. Two more studies, this time from AWS and IBM, continue to bear this out. I pulled out what I consider to be the key reasons.

Read the full article >>

Laptop computer partially open with keyboard showing bathed in blue light.

Photo by Philipp Katzenberger on Unsplash

News of the Week 📣

Databricks makes its third billion dollar acquisition with Neon database

Databricks made its third billion dollar acquisition in less than two years when it announced it was acquiring database startup Neon for a cool $1 billion this week. It followed the purchase of Mosaic in 2023 and Tabular last year in the billion dollar club.

The company can apparently afford it. At the end of last year it raised $10 billion on a $62 billion valuation. When you have that kind of cash combined with that level of value, it leaves room to make strategic acquisitions like this one to fill in holes in the product catalogue -- and Neon fills in a big hole.

It gives them a fast, serverless, Postgres, OLTP (online transaction processing) database. Now, rather than having to partner with other companies to provide database services, Neon provides Databricks with a powerful offering to call its own, says Sanjeev Mojan, a former Gartner analyst who now runs his own shop, SanjMo and has covered database technologies for many years.

Photo courtesy of Databricks

"It's an integrated end-to-end system from the replication and ETL all the way to analytical operations," Mojhan told FastFastforward. "So that entire market is what Databricks is going after."

What's more, as we enter an agentic world, it's a tool built for agentic workloads. As Neon noted in a blog post announcing the deal, "AI agents need state — they need to store context, embeddings, user interactions, intermediate plans, and more. But traditional Postgres setups weren’t built for this scale or dynamism. Neon was."

Now, Databricks has a much more complete solution to offer customers looking for ways to build modern applications, including AI agents, on top of all the data stored inside Databricks. We have been waiting for years for this company to finally go public, but it is instead taking advantage of its high value to patiently put all of the pieces into place first. One these days though, it's going to happen and it will be a more complete company when it does.

Klarna plans to hire human customer service agents after all

Last year, Klarna CEO Sebastian Siemiatkowski caused quite a stir when he reportedly claimed he was giving up Salesforce and other enterprise SaaS products and replacing them with AI agents. While the story gained momentum, it turns out it was harder to pull off than he initially thought.

In March on X, Siemiatkowski provided some further clarification, and maybe the company wasn't actually giving up on SaaS after all. "So no, we did not replace SaaS with an LLM, and storing CRM data in an LLM would have its limitations. But we developed an internal tech stack, using Neo4j and other things, to start bringing data and knowledge together," he wrote.

People working in customer service center with head sets on.

Photo by Charanjeet Dhiman on Unsplash

Just like many companies, Klarna is trying to figure out what's possible with being an AI-first organization, and got a little ahead of its skis. This week, the CEO admitted that he might have gone too far with his AI-first approach and the company will be hiring humans to take care of any customer who wants to talk to one. He also admitted that going strictly with AI led to lower-quality interactions.

Siemiatkowski tried to go all in, and he ran up against some of the current limitations of the technology. He admitted he was more than a bit embarrassed by the attention he got for it, but sometimes it takes bold statements and big swings to make a substantive change, and if he was guilty of anything, it was probably being too early before this technology is fully baked.

Microsoft Lays off 6000 employees, many in coding roles

Microsoft announced this week that it was laying of 3% of its workforce, which doesn't sound like much, but actually in a company the size of Microsoft with over 225,000 employees, it means that over 6000 people lost their jobs this week.

It's worth noting that CEO Satya Nadella got a 63% salary increase last year to just under $80 million per year, and the company made $245 billion in 2024, up 16%, yet the company still found it necessary to downsize. That said, every company, especially one the size of Microsoft, likes to cut its workforce from time to time. This one can probably be attributed to normal culling, but what's surprising is not the layoff themselves, so much as how the cuts broke down this time.

Photo by Ron Miller

Bloomberg reports that 40% of those cuts in Washington State were software developers, a group that tends to be safe when it comes to these periodic layoffs, but as AI becomes a capable coding partner, perhaps it's opening the door for developer cutbacks becoming more common place.

Adventures in AI 🤖

Photo by Andy Kelly on Unsplash

This is a new section this week on how I'm using AI to produce the FastForward blog and newsletter. Hope you like it.

I was using ChatGPT to help with some research for today’s newsletter. I asked it to find independent voices—not vendors—and to include links to the sources. Pretty clear ask, right? Yet it came back with three examples… all from vendors. Exactly what I said not to do.

So I clarified my prompt with a bit more guidance, like “think Accenture or IDC” as examples of the more independent perspective I was looking for.

It then did something interesting: It gave me two responses and asked me which one I liked better. One was clearly more aligned with what I wanted and I chose that, but I kind of liked the door number one or door number two strategy to hone in on exactly what I was looking for.

And to its credit, the second answer really helped providing me with a solid source that saved me some time and digging.

How are you using AI in your job? I would love to hear from you at ron@boldstart.vc.

What we're reading 📚

Person reading book, but all you can see is their hands holding the book. They are sitting on a hardwood floor steeped in shadow, creating an interesting look.

LivPhoto by Blaz Photo on Unsplash

The Verge

Why Workday’s new product head hopes he can make you like Workday

Nilay Patel

FastCompany

Coinbase cyberattack: What users need to know about stolen customer data, password security, and more in $400 million incident

Jennifer Mattson

ZDNet

The fastest-growing jobs for new grads and how to land one, according to LinkedIn

Sabrina Ortiz

Platformer

The myth of the AI-native startup

Casey Newton

Look who's talking 👄

"So you’re like, “wait a minute, this doesn’t make any sense. These 10,000 companies just got funded, and only 100 of them are good.” But you have to fund them all to get to the 100."

- Salesforce CEO Marc Benioff in interview with the Financial Times this week on the difficulty of finding the generational startups.

Please subscribe to the newsletter and feel free to share.

3350 Virginia St, Suite 262, Miami, FL 33133

ForwardThinking 🤔

Cloud repatriation revisited

Back to the data center

Trend or no trend

The on-prem argument

What's new on the blog 📰

For Salesforce’s head of security, it’s all about trust

IBM sees an AI future with smaller, more focused models in the enterprise

Palo Alto Networks tries to stay agile by maintaining a startup mentality

Enterprises continue to struggle to find AI success

News of the Week 📣

Databricks makes its third billion dollar acquisition with Neon database

Klarna plans to hire human customer service agents after all

Microsoft Lays off 6000 employees, many in coding roles

Adventures in AI 🤖

What we're reading 📚

The Verge

Why Workday’s new product head hopes he can make you like Workday

FastCompany

Coinbase cyberattack: What users need to know about stolen customer data, password security, and more in $400 million incident

ZDNet

The fastest-growing jobs for new grads and how to land one, according to LinkedIn

Platformer

The myth of the AI-native startup

Look who's talking 👄

Subscribe

FastForward #22: Cloud repatriation revisited

You may also like...