Even Hyperscalers Struggle with AI Demand

Microsoft recently had to ask its biggest cloud competitor, Amazon Web Services, for extra computing help because its platform named GitHub is growing too fast.

GitHub is a popular platform where software developers write and store code together.

The platform handled about one billion code updates in 2025, but engineers expect that number to jump to 14 billion updates this year.

This massive growth is happening because new artificial intelligence tools are making it much faster for humans and automated systems to write code.

The sudden traffic spike caused dozens of major outages for users earlier this year.

Microsoft bought GitHub back in 2018 and planned to move the whole system onto its own cloud network, Azure, by the year 2027.

Because Azure does not have enough immediate space to handle the current traffic boom, Microsoft is now using a multi-cloud approach that includes its main rival.

The company plans to spend 190 billion dollars this year to build more data centers so it can fix these shortages.

Relying on a competitor shows that the massive scale of new technology is testing the limits of even the largest systems in the world.

An important organization called the RIPE Network Coordination Centre manages internet numbers for more than 20,000 members across Europe, the Middle East, and Central Asia.

The group is now stopping its old plan to move everything to big online networks run by businesses in the United States.

The group started a plan years ago to use public computer networks from companies like Amazon and Google.

The leader of the group, Hans Petter Holen, explained that the world has changed and relying too much on foreign companies brings big risks.

New European safety rules also mean the organization must have full control over its own important systems.

The group plans to spend 5 million euros over the next few years to rebuild its own local computer setups.

This money will buy new hardware to replace old parts and create backup systems that do not depend on outside companies.

Reversing this path requires a lot of hard work, but the group believes keeping independent systems is safer for the internet.

Building your own computer centers is becoming a popular choice again as global tensions make public networks feel less safe.

NeoClouds are specialized, AI-first cloud platforms built entirely around high-density GPU infrastructure. Unlike traditional hyperscalers like AWS, Microsoft Azure, or Google Cloud, which provide broad, general-purpose IT services, NeoClouds focus strictly on the compute, networking, and storage patterns required for modern AI training and inference.

Enterprises are turning to these specialized providers to escape the high costs and capacity constraints of traditional public clouds. AI applications often cost 10 to 20 times more to run than conventional systems, making infrastructure efficiency critical. NeoClouds offer faster time-to-capacity for scarce GPUs, direct bare-metal access to eliminate virtualization overhead, and optimized networking clusters at a highly competitive price point.

Amazon Web Services is currently considering plans to sell its custom artificial intelligence chips named Trainium directly to outside companies for use in their independent data centers.

The commercial cloud provider has historically restricted its proprietary processors to its own server network to drive adjacent subscriptions like data storage and security monitoring.

Peter DeSantis, a senior executive at the firm, confirmed active discussions regarding external sales, which would place the tech giant in direct competition with Nvidia.

Chief Executive Officer Andy Jassy recently stated that the internal chip division could achieve a standalone annual value of 50 billion dollars if it serves external buyers.

The main operational challenge for this business pivot is the limited manufacturing supply of current and next-generation processors.

Spokesperson Doron Aronson noted that while previous requests for direct sales were rejected, the firm remains open to changing its policy to meet massive market demand.

Fulfilling external hardware orders could reduce the computing power available to existing cloud clients unless global chip factory partners rapidly increase production.

Shifting from exclusive cloud hosting to physical hardware distribution marks a major strategic transition that could alter how businesses buy artificial intelligence infrastructure.

Google announced a 1.5 billion dollar investment to expand its data center campus located in Jackson County, Alabama.

The project will grow a facility that started in 2019 on the site of a closed coal plant managed by the Tennessee Valley Authority.

To satisfy the white house ratepayer protection pledge, the tech company will pay for all the power and extra utility infrastructure needed for this expansion.

This strategy ensures that the massive electricity demands of new computer servers do not increase monthly utility bills for nearby families.

The facility is running on temporary power capacity while waiting for long-term clean energy options to finish development.

Google expects to run these computers using advanced nuclear power by 2035 through a partnership with Kairos Power.

Until that pipeline opens, the firm is working as a flexible grid partner by decreasing its electricity usage during peak hours to keep local energy networks steady.

Taking full financial responsibility for power grid expansion is becoming a necessary cost for large tech firms scaling up their systems.

Atlas, a company building an integrated operating system for the restaurant industry, provides a dedicated database instance to every merchant on its platform to ensure strict data isolation and predictable performance.

The platform initially utilized a standard managed database edition, but engineers quickly ran into bottlenecks as the number of clients and features scaled up.

The engineering team spent excessive hours troubleshooting CPU spikes and manually managing connection pooling across hundreds of distinct environments.

Moving the entire fleet to an advanced database edition resolved these visibility and management issues by building pooling and performance insights directly into the platform.

The transition immediately eliminated whole categories of routine database maintenance and infrastructure plumbing for the technical team.

Database administrators now spend 30 percent less time on operational upkeep, allowing them to shift focus to product innovation and customer-facing tools.

New restaurant clients can now be onboarded with a fully isolated, production-ready environment in a matter of seconds.

Automating database management and performance tracking allows enterprise platforms to scale their client base smoothly without experiencing proportional operational drag.