TrendForce puts the NOR Flash price rise at 100 to 120 percent, and says an AI server needs three to five times as much of it
The AI stack on 14 September 2026. 13 moves, 13 with a primary source.
TrendForce puts the NOR Flash price rise at 100 to 120 percent, and says an AI server needs three to five times as much of it
· L2 Silicon · 100-120% in 1H26; 3-5x more NOR Flash per AI server · Confirmed
TrendForce reported on 14 September that NOR Flash memory prices rose by a cumulative average of 100 to 120 percent in the first half of 2026 for high-capacity parts of 256 megabits and up, and forecasts a further 90 to 110 percent in the second half, which is a forecast rather than a measurement. This site already carries those two percentages from the Winbond acquisition item on 16 September; what is new here is the reason, direct from TrendForce: an AI server needs three to five times as much NOR Flash as an ordinary one. NOR Flash stores a device's startup instructions and goes into nearly every server, so a doubling in its price works through to what AI cloud companies charge and from there into the cost of the tools built on top.
Malaysia says data centre investment approvals have reached RM385.7 billion since 2021
· L3 Iron and cloud · RM385.7bn approved, 2021 to H1 2026 · Confirmed
Malaysia's investment authority, MIDA, said at a conference on 14 September that the country has recorded 385.7 billion ringgit, roughly 84 to 85 billion US dollars, of data centre related investment approvals cumulatively from 2021 through the first half of 2026. Two things to hold onto. It is a stock figure covering five and a half years, not money committed this week. And it counts approvals rather than money spent and capacity delivered, which is the same distinction Nscale's filing drew this week between contracted and active. This desk could not find a comparable prior-year total on MIDA's own site, so how fast the figure is growing is not something we can show. Malaysia is competing with Singapore and Indonesia for the same projects, and a number this size at a dedicated data centre conference is the argument that its strategy is working.
Perplexity, the AI search company, said on 14 September that it built its own database, CobbleDB, to replace the off-the-shelf system it was using, and that median read latency fell from 31.4 milliseconds to 5.60 milliseconds, with the slowest one percent of reads falling from 123 milliseconds to 24.2. It says the new system is at least 20 percent cheaper at every pricing tier than what it replaced. It also says it was built by two human engineers and hundreds of AI agents in two months, a claim about its own process that no outside party has checked. This is Perplexity's own plumbing under its own product, not something anyone can use, but faster and cheaper backend infrastructure is the sort of change that shows up later as a quicker product or a lower price without either being announced.
North Carolina's attorney general asks regulators to create a data-centre electricity rate class
· L1 Power and land · nearly $960m over two years, against a 15% rate request · Confirmed
North Carolina Attorney General Jeff Jackson asked state utility regulators on 14 September to create a separate electricity rate class for data centres and other very large power users. His office says Duke Energy Progress's pending request, a 15 percent increase, would otherwise put nearly 960 million dollars of what he calls unnecessary rate rises onto ordinary customers over two years, and that a separate Duke Energy Carolinas settlement would still lift average residential bills about 9.5 percent. Duke Energy is onboarding an unprecedented number of data centers without a clear plan on how to protect ratepayers from higher rates and unacceptable risk, Jackson said in his office's release. He also wants Duke to publish its template contracts with data centres and large users, report load forecasts every 90 days instead of every six months, and pause new gas plant construction. Nothing is decided; this is a request to a regulator. If adopted, this is the kind of rule that decides whether your bill rises because a data centre moved in, or whether the data centre pays for itself.
Bell Canada and the government of Saskatchewan announced plans to expand Bell's AI Fabric data centre campus near Regina by up to 900 megawatts on top of the 300 megawatts already running, a path to 1.2 gigawatts if every phase is built. The two sides describe it differently, and the difference matters. Bell's own release calls it a non-binding memorandum of understanding, says total capital investment including buildings, the computers inside them and new power generation could exceed 50 billion dollars, and states that development would proceed in phases as customer commitments are secured and subject to commercial agreements, permits and approvals, with no assurance the expansion happens at all. Saskatchewan's government release calls it a historic 52 billion dollar expansion and one of the largest private-sector investments in the province's history, and does not use the words non-binding. If it goes ahead the concrete change for Saskatchewan is construction and operating jobs. For everyone else nothing changes unless the plan becomes a signed, funded project.
Texas Governor Greg Abbott directed the state's water agency, the Texas Water Development Board, to start enforcing a water-use reporting requirement against large water users including AI data centres. The governor's own release says the Texas Water Code requires recipients of a survey to complete and return it, that a data centre which fails to do so commits an offence punishable as a crime, that the board must refer non-compliant facilities to the appropriate county or district attorney, and that a facility which misses the deadline becomes ineligible for state environmental permits, amendments or renewals. The board itself cannot prosecute anyone, so it must report its enforcement progress to the governor's office by 14 October 2026. According to Texas Tribune reporting, compliance has been low for years: 32 percent of 22 data centres surveyed responded in 2023, 28 percent of 67 in 2024, and roughly 30 percent of 329 in 2025. By our count this is the governor's third directive on AI data centres since August.
Bitdeer, a company that started out mining bitcoin and has been shifting into renting out AI computing power, signed a ten year data centre services agreement for A202, a 65.1 megawatt facility on its existing campus in Johor Bahru, Malaysia. The company says this takes its secured AI cloud capacity, meaning capacity it owns or has under an executed agreement, to approximately 206.5 megawatts, about 59 percent of its target of up to 350 megawatts to be delivered by the first quarter of 2028, and that its active pipeline for AI cloud capacity is now estimated to exceed 7 billion dollars. A pipeline is potential business, not signed business. For most people nothing changes; this is infrastructure being built in Southeast Asia to rent computing power to AI companies.
Meta's data centre campus in Eagle Mountain, Utah has now drawn more than 3 billion dollars of the company's capital investment in Utah in total, the city government says. That is a cumulative figure built up since Meta arrived, not new money announced this week. Alongside it, a separate company, Lydian Energy, owns and operates a new 150 megawatt battery storage facility called Faraday that supports the site and the local grid. The city says Meta has contributed more than 6.2 million dollars to local schools and organisations to date. For Eagle Mountain residents this means continued local jobs and community funding tied to the campus. For everyone else, nothing changes.
NextEra and Dominion filed new paperwork with Virginia's utility regulator on 14 September, offering residential customers four years of 10 dollar monthly bill credits instead of two. The companies say the extension is funded mainly by redirecting credits that would have gone to large data centres, plus about 90 million dollars from shareholders, a figure reported by the Virginia outlet Cardinal News rather than stated in the filing. The package also raises Dominion's low income assistance programme by 100 million dollars through 2038, adds a 100 million dollar workforce fund, a Virginia supplier programme of up to 1 billion dollars a year for five years, and 1,000 direct jobs. Virginia regulators hold hearings in November and are expected to rule in January 2027, with the 67 billion dollar merger still targeted to close in the second half of 2027. If you are a Dominion Virginia customer this means a longer bill credit if it is approved. Anywhere else, it changes nothing for you.
Crusoe opens a second Tulsa factory making the electrical gear AI data centres run on
· L1 Power and land · 400,000 sq ft added, 1M sq ft total · Confirmed
Crusoe, a company that builds AI computing facilities, opened a second factory in Tulsa, Oklahoma that makes electrical equipment such as switchboards, adding 400,000 square feet and bringing its national manufacturing footprint to roughly 1 million square feet. This grows Crusoe's own capacity to build components for data centres; it is not new electricity or computing capacity coming online. The company says it has hired more than 300 local craft workers across its two Tulsa sites, and that its factories have supplied more than 2,500 switchboards to its Stargate Project site in Abilene, Texas. For an ordinary person the concrete change is local: manufacturing jobs in Tulsa, not a change to anyone's electricity bill.
South Korea's Ministry of Trade, Industry and Energy said the country's information and communications technology exports reached 59.98 billion dollars in August, a record and more than double the 22.84 billion exported a year earlier, driven largely by semiconductors. Imports rose to 18.61 billion dollars, leaving a trade surplus of 41.37 billion, the first time Korea's technology trade surplus has topped 40 billion dollars in a single month, according to the ministry's own release. This is a government tally of a whole export sector, not one company's results. Korea's chip exports have been one of the clearest ways the AI boom shows up in a national economy, and a surplus this large can support the country's currency and public finances, though an ordinary person sees no direct change from the figure itself.
A study for Pennsylvania's utility regulator finds the PJM grid could see more than 13 days a year of shortfalls by 2030 in a bad case
· L1 Power and land · LOLE 0.59 (reference) to 13.20 (worst case) by 2030, vs 0.1 target · Confirmed
A study commissioned by Pennsylvania's Public Utility Commission and written by Synapse Energy Economics with two other firms models how reliable PJM, the grid operator serving thirteen states and the District of Columbia, will be in 2030. It measures loss of load expectation, the expected number of days a year when supply falls short, against PJM's planning target of 0.1, or one day in ten years. On PJM's own 2026 load forecast the study gets 0.59, nearly six times worse. In its worst case, with higher demand and slower new supply, it gets 13.20, which it describes as more than 13 days a year with loss of load events. It says most of the load growth through 2040 in Pennsylvania and PJM comes from data centres. For people in the region this is a warning from a model, not an outage; whether it comes true depends on how much gets built and how fast new power follows.
Abridge expands its AI tool to check hospital bills for errors before they are sent out
· L6 Applications and agents · Pre-bill review launched; WashU/BJC access grew from 450 clinicians to ~4,000 · Confirmed
Abridge, whose AI writes clinical notes from doctor-patient conversations at more than 300 large US health systems, has launched a product that reviews hospital claims before they are submitted for payment. It compares the diagnoses and billing codes on a claim with what the clinical notes actually support, so coding teams can fix mismatches first. As an example of its growth, the company says WashU Medicine and BJC Health went from 450 clinicians in 2025 to about 4,000 now able to use its tools. For patients, catching coding mistakes before a claim goes out could mean fewer wrong bills and fewer insurance denials, but nobody has yet measured how much it actually helps.