Good morning 👋
AI infrastructure is splitting into two races: secure compute, and secure everything around it. Today’s signal is that power, grid access, long-haul fiber, liquid cooling and optical transport are being contracted or acquired years ahead of demand. The server order is increasingly the easy part.
🔥 The Big Boil
Cipher starts building the gas path for up to 2.5 GW of on-site power
Cipher Digital has begun developing natural-gas lateral pipelines at multiple data-center sites, a move intended to support up to 2.5 GW of new on-site generation. The company says the pipeline work could make new power available before the end of 2027 and plans to work with power providers that would develop and operate the generation.
The status matters. Cipher has started pipeline development; it has not installed 2.5 GW of generators, energized 2.5 GW, or leased an equivalent amount of data-center capacity. The figure is the maximum generation the planned gas infrastructure could support. That distinction is the whole story.
Why it matters: developers are increasingly treating power procurement as part of the data-center product rather than something the utility supplies later. Gas laterals can bypass years-long interconnection uncertainty and let a campus scale before the grid is ready. That creates opportunity for generation equipment, pipeline contractors, switchgear and microgrid suppliers — but also moves fuel-price, permitting and emissions risk onto the project.
This is also a financing story. A site with a credible path to power is easier to lease and finance than one sitting in an overloaded interconnection queue. The strategic bet is that speed-to-power will command enough rent to justify owning more of the energy problem. But “power before end-2027” still requires generation equipment, permits, gas deliverability and construction to line up on an aggressive schedule.
What to watch: whether Cipher names generation partners, converts the enabled power into signed HPC leases, and ultimately secures grid connections. Until then, 2.5 GW is enabled potential, not operating capacity.
⚡ What’s Boiling
📈 Broadcom’s custom AI silicon curve gets steeper
Broadcom reported $16.7 billion of fiscal-third-quarter AI semiconductor revenue, up 221% year over year and 54% sequentially. It expects $21.7 billion in the fourth quarter as demand for custom AI accelerators and networking remains strong.
Why it matters: hyperscaler infrastructure is broadening beyond merchant GPUs. Custom accelerators still pull on advanced packaging, HBM, substrates, high-speed switching and optics, so Broadcom’s growth is a capacity-planning signal for suppliers around the chip. Broadcom also says it has secured supply needed for its current outlook. The next question is whether packaging, memory and optical supply can scale as quickly as customer silicon programs do
📣 Texas is forcing the interconnection queue to prove it is real
Texas Governor Greg Abbott ordered ERCOT and the Public Utility Commission in August to audit data-center projects before they move forward. ERCOT was considering more than 474 GW of requests, about 90% tied to data centers; fresh reporting says political resistance is hardening as grid-cost concerns grow.
Why it matters: a queue is not a build forecast. Duplicate requests, speculative sites and unfunded projects can make demand look much larger than what will connect. Texas is becoming a test case for separating serious loads from “ghost demand.” Developers bringing generation, water plans and committed capital should have an advantage.
🧠 Boiled Down
Today’s Term: Behind-the-meter generation
In plain English: power produced on the customer’s side of the utility meter, often on or beside a data-center campus, rather than supplied entirely through the grid.
Why you’re hearing about it: Cipher’s pipeline plan and the Northern Territory proposal both use local generation to reduce dependence on slow grid connections.
Why it matters: it can shorten the path to power, but shifts fuel, equipment, permitting and emissions risk into the data-center development stack.
🤯 Did You Know? AI capacity is being won upstream — in pipelines, permits, fiber, optics and financing — before a GPU ever reaches the rack.
Till next time,
Grid Boiler Team