Texas has ordered audits for new data centre projects just as Amazon, Google and Microsoft say AI demand is outrunning new capacity.
Two developments on 4 August point to the same pressure point in digital infrastructure. In Texas, Governor Greg Abbott said all new data centre projects will need to be audited by both the Public Utility Commission of Texas and the state's grid operator ERCOT, according to TechCrunch's report. Separately, The Register reported that Amazon, Google and Microsoft are committing vast sums to expand cloud and AI infrastructure, while saying they still cannot add enough capacity to meet demand.
In plain English, the issue is this: data centres are the buildings full of servers that store data and run cloud software and AI systems. More AI use means more demand for electricity, chips, memory and physical space. The latest news suggests that even large tech companies are colliding with practical limits, while states that host many data centres are starting to ask harder questions about power use and local impacts.
Texas is moving from voluntary disclosure to mandatory audits
According to TechCrunch, Texas has been attractive to data centre developers because of relatively light regulation and what had been seen as abundant power. The state already hosts more data centres than every US state except Virginia.
That backdrop has changed as the queue of projects waiting to connect to the ERCOT grid has grown sharply. TechCrunch reported that ERCOT had 233 gigawatts of projects in its interconnection queue in January, and that figure had risen to 474 gigawatts in less than six months. Abbott's office said about 90 percent of those new connection requests are data centres.
An interconnection queue is the line of projects asking to connect to the electricity grid. It does not mean all of them will be built. As TechCrunch notes, some are only paper proposals, and many projects drop out over time. Even so, the scale matters: the report says the current queue represents more than five times ERCOT's total peak demand. That is why the state is no longer treating the buildout as routine.
The new audits are meant to gather more concrete information before projects move ahead. TechCrunch says Abbott directed the PUCT and ERCOT to collect details including proposed on-site and off-site electricity demand, water demand, noise mitigation, light controls, tax incentives and ownership information. The article also reports that Abbott had previously tried a voluntary survey, but most data centre operators did not respond.
This is a meaningful change in how Texas is handling growth. The state is not banning all development in the source report, but it is clearly adding a new review step for new projects. For developers and infrastructure planners, that may mean more scrutiny over whether a planned facility can be supported by the grid and what local effects it may have.
Cloud providers are spending more, but say capacity is still short
While Texas is tightening oversight, the largest cloud providers are signalling that demand for AI computing remains ahead of supply. According to The Register, Amazon said it now expects to spend about $220 billion in cash capital expenditure in 2026, up from an earlier estimate of about $200 billion. The report says CEO Andy Jassy told analysts that even at that level, Amazon still will not have enough capacity to meet all demand in 2026, and that he believes the same dynamic will hold in 2027.
Google also increased its 2026 capital expenditure guidance, to a range of $195 billion to $205 billion from a previous $180 billion to $190 billion, The Register reported. Chief financial officer Anat Ashkenazi said the increase was due mainly to faster delivery of capacity to meet growing demand. The same report says she also warned that the higher infrastructure investment would increase depreciation and data centre operating costs, including energy.
Microsoft's figure is more complicated. The Register says Microsoft put expected calendar 2026 capex at about $175 billion, roughly $15 billion below an earlier number, but also said this does not reflect a reduced infrastructure buildout. Instead, the reported change is tied to accounting treatment, because more future data centre leases will be classified as operating rather than finance leases. The report also says Microsoft is extending the estimated useful life of its data centres and office buildings from 15 to 25 years.
Those three figures add up to about $595 billion on paper, but The Register explicitly warns that they are not directly comparable. Amazon's number covers cash capex across several businesses, Alphabet's guidance is company-wide, and Microsoft's figure is affected by lease accounting as well as direct expenditure. So the most solid conclusion is not the precise combined total, but the broader one: the biggest cloud companies are still increasing infrastructure commitments because they say they cannot meet demand fast enough.
What is causing the bottlenecks
The source material points to several constraints at once. Power is one. In Texas, the state is responding because proposed data centre demand could strain the grid, according to TechCrunch. Cost is another. The Register says Amazon cited the higher cost of memory as one reason its 2026 capex estimate rose.
Supply constraints also extend beyond money. The Register says memory, graphics processing units or GPUs, and even hard disks are in short supply. A GPU is a type of processor widely used for AI training and inference, meaning building and running models. The article argues that shortages may be one reason some enterprises are turning to cloud providers instead of building their own systems, because hyperscalers are more likely to secure scarce components.
The report attributes that view partly to Amazon's Andy Jassy, who earlier this year said supply shortages were pushing companies with on-premises infrastructure towards the cloud because suppliers prioritise very large customers. That does not prove every organisation is making that choice for the same reason, but it does show how one major provider is framing current demand.
The Texas story adds another practical limit: information. According to TechCrunch, the governor wants details not only about electricity use but also water demand, noise, lighting, incentives and ownership. That suggests the challenge is no longer just whether a company wants to build, but whether the state believes it has enough visibility into the project's wider effects.
What these reports actually suggest
Taken together, these reports suggest that AI infrastructure growth is no longer constrained only by corporate budgets. The source material points to a wider bottleneck: cloud providers say demand still exceeds available capacity, while Texas is adding scrutiny because proposed data centre growth could strain the grid and create broader local impacts.
The strongest conclusion supported by these reports is that adding more compute now depends on more than spending. It also depends on power access, component supply and whether regulators are satisfied that new projects can be supported. What to watch next is whether Texas audits materially slow proposed projects and whether hyperscalers can narrow the capacity gap they say will persist into 2027.