10 Unconventional IT Decisions That Paid Off: Lessons from Global Tech Leaders

Every IT leader eventually faces a fork in the road: do we build or buy, scale up or scale out, rent from the cloud or own the metal? These aren’t just technical choices. They are business-defining decisions with millions (sometimes billions) of dollars on the line.

The conventional wisdom says: “move to the cloud, outsource infrastructure, focus on product.” And for many, that’s right. But some companies made unconventional IT decisions that seemed counterintuitive at the time, yet delivered massive long-term advantages.

This case study dives into 10 thought-provoking case studies, from Ahrefs’ bold rejection of AWS to Zoom’s pandemic-fueled multi-cloud scramble.

Each story is less about technology itself and more about strategy, trade-offs, and outcomes.

Let’s begin with the poster child of contrarian thinking: Ahrefs.

1. Ahrefs: Saving $400M by Choosing On-Prem Over AWS

When the SEO giant Ahrefs explained why they rejected AWS and built their own on-premises infrastructure, the internet buzzed with hot takes. “Why would a modern SaaS avoid the cloud?” sceptics asked.

But Ahrefs wasn’t guessing — they had done the math.

The Challenge

Ahrefs runs some of the largest-scale crawlers and data pipelines outside of Google itself. Their crawler processes billions of pages daily, requiring petabytes of storage and high-throughput compute.

If they followed the default startup path of spinning everything on AWS, then their infrastructure bill would have exploded.

The Decision:

Instead of going “all in” on AWS, Ahrefs colocated racks of servers in Singapore and the U.S., buying and maintaining their own hardware. This allowed them to fine-tune compute and storage ratios for their specific workloads.

In a widely read post, CTO Dmitry Gerasimenko calculated that AWS would have cost approx $448M over three years, while their on-prem strategy cost them just around $39M — a massive difference of around $400M saved.

The Trade-offs:

• CapEx vs OpEx: They had to spend upfront on capital expense for hardware, whereas AWS’s pay-as-you-go model would have been an operating expense.
• Operational Overhead: Owning hardware meant hiring staff for data centre ops. That’s a recurring overhead. Add to that talent attrition, and you need the best talent to maintain those servers in-house, running 24x7.
• Flexibility: Scaling takes longer with physical servers vs instantly on AWS. When you have a huge influx of customers or want to upgrade your infrastructure, it is not just a click of a button but a series of meetings, calculations, approvals, etc.

The Outcome:

• Massive Savings: The money saved on building on-prem infrastructure was reinvested in R&D and product.
• Predictable Costs: They now know their monthly expenses for their infrastructure. There was no upscaling or downscaling.
• Full Autonomy: Ahrefs has full control of architecture and optimisations tailored to their crawler workloads. They can make the tweaks and changes as required.

Key Insight:

Cloud isn’t always cheaper. At a massive, steady-state scale, owning hardware beats renting it. Ahrefs showed that the “default cloud mindset” can cost companies hundreds of millions if not critically examined. Calculating the cost that you can incur over time can help you make a rational, data-backed, informed decision.

2. Dropbox: Building “Magic Pocket” and Saving $75M

Dropbox was born in the cloud era. From the start, it ran on AWS, a logical choice for a fast-scaling file storage service. But by 2013, the economics looked different.

The Challenge

Dropbox’s entire value proposition revolved around storage at scale. By 2013, they had hundreds of petabytes in AWS S3, and costs were climbing exponentially.

The leadership team asked: “Does it make sense to keep renting storage when storage is our core business?” And this made sense.

Dropbox itself offered storage, and yet it operated on another company’s storage. It was similar to having an office on rent elsewhere when you are building commercial buildings.

The Decision

They built Magic Pocket, an in-house object storage system. Over two years, Dropbox migrated around 500 petabytes of user files out of AWS and into its own infrastructure.

The Trade-offs

• Engineering lift: Building and testing a production-grade distributed storage system was a monumental effort. Additional manpower might have been required to deploy the system.
• Migration risk: A botched migration could corrupt user data, the company’s lifeblood. Migration had to be 100% correct and working.
• Focus diversion: Engineers spent months building the infrastructure instead of features. That’s a revenue vs cost trade-off. No new features might be seen as stagnant for the company.

The Outcome

• Savings: Savings of around $74.6M over two years.
• More Control: Increased control over latency and storage optimisations.
• Differentiation: Dropbox could now optimise infrastructure in ways AWS couldn’t.

Key Insight

For companies where infrastructure is the product, it can be cheaper (and strategically smarter) to build than to rent. Dropbox’s core offering is storage, and owning it is a flex. You are not dependent on any other company.

Dropbox bet big, and it paid off. Long-term, it has got savings and a satisfied customer base pulling in the revenue month on month, year on year.

3. Netflix: Betting the Company on AWS

While Dropbox left AWS, Netflix doubled down on it. Yes, you read it right. Netflix invested more in AWS servers, proving that opposite strategies can both be right, depending on context.

The Challenge

In 2008, Netflix was still a DVD-by-mail service with a fledgling streaming business. When a database corruption caused a multi-day outage, leadership realised their legacy data centres couldn’t scale to meet future demand.

When you can’t scale, you can’t grow your revenue. That is a problem.

The Decision

Netflix went all-in on AWS, embarking on one of the largest cloud migrations in history. Over nearly a decade, they rebuilt every system, encoding, recommendation, and content delivery just for the cloud.

By January 2016, Netflix completed its migration, shutting down the “last remaining data centre bits” for streaming.

The Trade-offs

• Vendor lock-in: Netflix became one of AWS’s biggest customers. With streaming and storage at scale (with Netflix’s huge library of titles), it became a key customer.
• Cost visibility: Cloud bills ballooned but were seen as necessary to scale. There was no sure figure for the AWS Bill, since customers would binge on-demand, anytime, anywhere.
• Operational shift: Engineers had to embrace chaos engineering, microservices, and cloud-native design patterns. They had to shift not just servers but their mindsets too.

The Outcome

• Jump in Revenue: Netflix scaled to serve 220+ million subscribers globally. That must have surely brought revenue in numbers since Netflix doesn’t offer free plans.
• New Practices: The company pioneered chaos engineering and resilience-at-scale practices.
• Leverage in Service: Partnering deeply with AWS gave Netflix leverage in defining new cloud services. They could suggest or even push for features and get them done from AWS, since they were the forerunners in streaming and a key account for AWS.

Key Insight

Sometimes the right move isn’t saving money but buying infinite scalability. Netflix couldn’t afford to be in the data-centre business; it needed global reach fast.

It invested in the cloud to give an uninterrupted hassle hassle-free service, and it resulted in gaining subscribers, driving up the revenue. They invested first and reaped the benefits later.

4. Zoom: Multi-Cloud to Survive the Pandemic Surge

If ever there was a stress test for IT infrastructure, it was March 2020. As the pandemic began, Zoom went from a fast-growing SaaS to an overnight global necessity.

The Challenge

In December 2019, Zoom averaged 10 million daily meeting participants. By March 2020, that number exploded to 200 million, and by April it hit 300 million. That’s a 30x spike in just 4 months (20x spike for the previous quarter).

No infrastructure team in the world could forecast a 30x spike in traffic in 90 days.

The Decision

Zoom scrambled to go multi-cloud. While AWS remained its primary provider, Zoom announced a major partnership with Oracle Cloud in April 2020 to handle overflow demand.

The Trade-offs

• Complexity: Running across clouds added operational challenges.
• Vendor management: Zoom had to negotiate SLAs with multiple hyperscalers.
• Short-term over-optimisation: In the rush, some architectures were built “just to survive.”

The Outcome

• Zoom stayed online during the biggest traffic surge in SaaS history.
• Oracle gained a marquee reference client.
• Zoom’s stock price and brand skyrocketed, cementing it as the pandemic’s default video tool.

Key Insight

Sometimes the smartest infrastructure decision isn’t elegant, it’s pragmatic. Multi-cloud became Zoom’s safety net when the stakes were survival.

Zoom might not have gone for the most cost-effective option, but it surely survived and thrived on the challenges to grow into a global business meeting app.

5. Basecamp (37signals): Leaving the Cloud to Save Millions

While startups are told to “never buy servers again,” Basecamp (37signals), the company behind Basecamp and HEY, decided to exit the cloud in 2022–23.

The Challenge

After years of AWS/GCP usage, Basecamp’s leadership realised cloud costs were eroding margins. CTO David Heinemeier Hansson (DHH) argued that the cloud tax made sense for startups, but not for profitable, steady-state companies.

The Decision

They migrated workloads back to colocated servers and wrote extensively about the process in their blog series Leaving the Cloud.

Basecamp projected savings of ~$7M over five years from the move.

The Trade-offs

• Upfront effort: Rebuilding infra and migrating workloads took months. They build from scratch to attain the best results.
• Cultural pushback: Many engineers resisted giving up cloud-native conveniences since systems were not broken, and it made no sense for them to work on a new system.
• Scalability: Colocation means slower scaling if demand suddenly spikes.

The Outcome

• Lower, more predictable costs.
• A marketing boost since their contrarian stance drew attention and sparked industry debate.
• Renewed engineering focus on efficiency.

Key Insight

Cloud isn’t “forever the right answer.” For stable businesses with predictable loads, owning infrastructure can restore profitability and control.

Basecamp thought something similar to Ahrefs and made their own infrastructure to save millions in the long term and give better services to their customers.

6. Spotify: Cloud Migration for Agility, but Hybrid by Necessity

While some companies save by leaving the cloud, Spotify did the opposite, betting on Google Cloud to drive agility.

The Challenge

By 2015, Spotify was managing its own data centres. That worked when it was a scrappy music startup, but with millions of users streaming concurrently and global expansion looming, its infra team faced mounting complexity. Scaling clusters, managing Kafka pipelines, and maintaining high uptime across multiple regions was overwhelming.

The Decision

In 2016, Spotify announced a strategic partnership with Google Cloud Platform (GCP). They migrated data analytics, backend services, and machine learning workloads to Google Cloud.

The Trade-offs

• Migration effort: Years of moving services, rewriting dependencies, and retraining engineers.
• Vendor dependence: Becoming heavily tied to Google’s ecosystem.
• Hybrid reality: Some components (like parts of music delivery infra) remained in Spotify-run data centres for years.

The Outcome

• Spotify engineers gained agility with GCP tools like BigQuery and Dataflow, speeding up experimentation in recommendations and ads.
• Global user growth continued past the 500M monthly active users mark by 2023
• The company modernised its data platform without a catastrophic outage.

Key Insight

For businesses with rapid product iteration cycles, the speed and tooling of cloud providers can outweigh pure cost savings. Spotify chose agility over cost savings. Thus, fueling their competitive edge.

7. Cloudflare: Owning the Edge Network

While most SaaS companies rely on hyperscalers, Cloudflare built its own edge network from scratch and turned it into its core differentiator.

The Challenge

Cloudflare’s mission was to “help build a better Internet”, and it meant operating at the network edge. Their services (DDoS protection, CDN, DNS, Workers) needed to run close to users everywhere, not in a few centralised cloud regions.

The Decision

Instead of leaning on AWS or GCP, Cloudflare built a global anycast network spanning hundreds of cities. By 2023, their infra was present in 300+ cities worldwide.

The Trade-offs

• Heavy upfront investment: PoPs in hundreds of cities, hardware supply chains, and global ops teams.
• Operational complexity: Running your own backbone network is not for the faint of heart.
• Delayed profitability: Cloudflare ran losses for years before scale economics kicked in.

The Outcome

• Cloudflare became synonymous with Internet performance and reliability.
• Edge computing platform Workers emerged as a strategic differentiator.
• By 2023, Cloudflare's revenue topped $1B annually, proving the network bet paid off.

Key Insight

Cloudflare shows that sometimes the best strategy is to own your moat. Instead of building on someone else’s infrastructure, they built infrastructure as their product and turned it into an enduring advantage.

8. Shopify: Cloud Migration for Global Retail Scale

Shopify is another instructive case, one that demonstrates why infrastructure decisions can’t be judged in isolation.

The Challenge

Shopify powers millions of merchants, each with spiky, unpredictable traffic patterns (think Black Friday). Running everything in Shopify-managed data centres created scaling headaches.

The Decision

In 2018, Shopify partnered with Google Cloud to migrate its core infrastructure. The move gave them access to elastic scaling, global availability, and advanced data tools.

The Trade-offs

• Cost exposure: Cloud bills grew, but predictable elasticity reduced the risk of downtime.
• Migration lift: Shopify had to re-architect monolithic parts of its Ruby on Rails stack.
• Dependency risk: Strategic reliance on Google’s infrastructure.

The Outcome

• Shopify handled record-breaking Black Friday / Cyber Monday traffic year after year without major outages.
• Developers could focus on features (Shopify Plus, payments, Shop app) rather than firefighting servers.
• The migration positioned Shopify as a resilient platform through the pandemic-era ecommerce boom.

Key Insight

Shopify has a variety of customers who run their e-commerce stores on its platform. In our experience, many even shut down the business and hence do not renew their subscriptions, which is a revenue loss for Shopify.

In a situation where customer continuity and revenue can’t be predicted with certainty, having an elastic option makes sense.

For a platform with volatile, spiky traffic, elastic cloud infrastructure is often the lesser evil. Shopify bought resilience at scale, even at a premium.

9. Apple: Building Data Centres for Privacy and Control

Apple stands apart: it’s one of the few trillion-dollar tech companies that still invests heavily in its own data centres, even while using cloud partners tactically.

The Challenge

Apple’s brand promise revolves around privacy and control. iCloud, iMessage, and Siri all rely on sensitive user data. Outsourcing 100% of that to AWS or GCP would not align with Apple’s narrative of being the guardian of user privacy.

The Decision

Apple built and expanded its own data centres across the U.S. (North Carolina, Nevada, Iowa, etc.), while still leasing capacity from AWS and Google for specific workloads. Apple announced multi-billion dollar U.S. data centre investments, including a $4.7B renewable energy-powered facility expansion in Iowa.

The Trade-offs

• CapEx burden: Data centres cost billions to build and operate.
• Time to scale: Building infra isn’t as fast as spinning up cloud services.
• Hybrid complexity: Apple still juggles workloads across its own infra and cloud partners.

The Outcome

• Apple retained tight control over sensitive workloads, reinforcing its privacy-first reputation.
• Significant economic impact: thousands of U.S. jobs tied to Apple’s infrastructure expansion.
• The company maintained leverage over cloud partners by not being fully dependent.

Key Insight

For Apple, infrastructure isn’t just cost; it’s brand.Owning the stack signals trust in a way outsourcing never could.

Apple doesn’t just sell iPhones and Macs; it sells privacy, which is built on trust. People trust Apple since its priority is privacy. And owning its infrastructure stack reiterates the same.

10. Airbnb: Riding the Cloud from Day One

Finally, Airbnb illustrates the opposite of Dropbox or Basecamp. They leaned on the cloud early, and it enabled survival.

The Challenge

In 2008, Airbnb was a struggling startup renting air mattresses. Infrastructure wasn’t their core competency. They needed to scale fast, cheaply, and globally without hiring ops teams.

The Decision

Airbnb built directly on AWS (read AWS’s official case study), embracing managed services to keep engineering focused on the product.

The Trade-offs

• Vendor lock-in: Airbnb became deeply tied to AWS.
• Cloud cost exposure: Bills scaled alongside bookings.
• Potential migration headaches: Switching infra later would be costly.

The Outcome

• Airbnb scaled to millions of listings across 220+ countries.
• AWS enabled global reach long before Airbnb could have built its own infrastructure.
• During COVID, when demand collapsed, Airbnb avoided stranded infra costs thanks to AWS’s elasticity.

Key Insight

For startups, cloud is survival, not a luxury. Airbnb could never have afforded global infrastructure on its own in 2009. Renting allowed them to disrupt an entire industry.

When operations were hit during the pandemic, they did not bear an overhead since the server bills also dropped. This saved them millions.

With AWS, Airbnb has infrastructure as a variable operational expense, not a fixed cost that it has to pay every month. And it makes sense for them.

Conclusion

When we look closely at these ten case studies, some clear patterns emerge. Let’s try to understand them.

• Scale determines economics: At a small-to-mid scale organisation, cloud is cheaper and faster (Airbnb, Spotify). At a massive, steady-state scale, such as Ahrefs, on-prem can save hundreds of millions.
• Agility vs cost is a trade-off: Netflix, Zoom, and Shopify chose agility and scalability over raw savings. Their reward was resilience during hypergrowth and crises.
• Infra can be strategic branding: Apple’s data centres aren’t just about compute; they reinforce its privacy-first brand. Cloudflare’s network isn’t just infrastructure; it is their competitive moat.
• Hybrid is the norm, not the exception: Very few companies are 100% cloud or 100% on-prem. Most mix strategies for resilience, compliance, or cost savings.
• Timing matters: Startups often should rent first, own later. Mature companies can revisit infra decisions to claw back margin (as Basecamp & Dropbox did).

Each case reveals a different calculus. Every organisation needs to assess its position, requirements, trade-offs and benefits to weigh the options.

Cloud, On-Prem, Mixed Strategy — there is no correct answer. It all depends on the factors mentioned above and what the organisation wants to achieve. Also, risk appetite matters.

Making such decisions requires years of experience and expertise. It is not only about the calculations made on a spreadsheet, but also about factoring in future expectations and past experience.

We at Pleximus Inc. have the experience and expertise to guide you in making such decisions. We have migrated hundreds of servers, managed databases and optimised technical operations for clients spread across industries.

Our team is adept at helping you make the right decision and save costs while growing your business.