The next phase of the artificial-intelligence boom may be less about building ever-larger systems and more about working out whether the economics still add up. A report from The Register says analysts at Jefferies are questioning whether large language models (LLMs) will become profitable as the price of processing AI requests continues to fall.
The report points to cheaper Chinese models and open-source LLMs as possible forces behind a major change in the AI infrastructure market. If capable models can be run at lower cost, businesses may have less reason to pay for the most expensive systems or use the largest model for every task. That could affect demand for data-centre capacity, specialist chips and the cloud services built around them.
A reset, not necessarily an AI retreat
It is important to separate pressure on prices from a collapse in interest. Lower token prices can make AI more affordable for customers, while also making it harder for providers to recover the cost of developing and running their services. The Register describes the possibility as an infrastructure “reset”, reflecting uncertainty around a market that has attracted enormous spending.
Microsoft’s plans illustrate why the investment story is still moving in the opposite direction. In a separate report, The Register says the company is preparing a multi-gigawatt rollout of its next-generation Azure Maia AI accelerator racks. The report says the new configuration is expected to deliver seven times the token throughput of its predecessor, with production deployments due in 2027.
Those plans are an expectation rather than proof that the spending will pay off. They also show the tension at the heart of the market: providers are preparing more specialised hardware and capacity just as cheaper software may reduce the amount customers are willing to pay for each AI interaction.
Cloud convenience has a reliability trade-off
The infrastructure question is not only about price or performance. The Register also reported an outage affecting Microsoft 365 users in Germany and the UK, in an article connected with the end of Office 2021 LTSC. The supplied report summary does not establish that the end of that product caused the outage, so it would be wrong to treat the two events as directly linked.
For households, schools and small organisations, the practical lesson is familiar: a move to online services can simplify access and updates, but it also makes availability dependent on a provider. Keeping important files in more than one place, knowing which applications work offline and maintaining a current recovery plan can reduce disruption when a cloud service is unavailable.
Different answers to an AI-saturated future
There is also a counterpoint to the assumption that every computer project needs AI. The Register reports that NetBSD may be finding a role as an AI-free free and open-source operating system, alongside increasing awareness, adoption and donations. That is not presented as a replacement for mainstream desktop or cloud platforms, but it is a reminder that some users value systems that remain focused, inspectable and free from unwanted AI features.
Elsewhere in consumer technology, BBC Technology reports that Bumble has been bought by an investment firm for $1.3bn. The dating app had recently dropped its “women-first” chat rule. The deal is described as a bet on the future of humanity, but its completion does not by itself tell users how the service, its features or its approach to communication will develop.
What this means for everyday technology users
The common thread is a need to look beyond headline promises. Falling AI prices could bring useful tools within reach while challenging the businesses supplying them. New accelerator hardware may improve capacity, but planned deployments are not the same as guaranteed benefits. Cloud software can be convenient, yet resilience still matters. And open-source alternatives continue to offer a different set of priorities for people who want control over their computers.
For anyone choosing a device or service, check the basics: what works without an internet connection, how data can be exported, whether an older system is still supported, and what happens if a provider changes direction. At The Tech Surgery, our practical view is that dependable technology is not always the newest or most powerful option; it is the option whose costs, limitations and recovery choices you understand.
Sources and further reading
- The Register: Cheap Chinese and open source LLMs could trigger trillion-dollar AI infrastructure “reset”
- The Register: Microsoft plans multi-gigawatt rollout of next-gen Azure Maia AI accelerator racks
- The Register: Microsoft celebrates end of Office 2021 LTSC with MS 365 outage
- The Register: NetBSD may have found a new role: The AI-free FOSS OS
- BBC Technology: Bumble bought for $1.3bn in “bet on the future of humanity”