Twenty gigawatts of AI compute finance, and ERCOT is tightening the gate before the load gets energized. You're listening to The Data Center Daily. Today: somebody's looking to finance an enormous pile of compute—while the grid starts asking for receipts. Plus, a long-delayed Western transmission line finally moves from proposal to ground. Start with Comcast. From Business Wire:
Comcast Business today announced a collaboration with Equinix, the world’s digital infrastructure company®, that will allow enterprises to order Comcast Business last-mile connectivity through standards-based APIs directly within Equinix Fabric®, Equinix’s software-defined interconnection service that enables customers to digitally connect their distributed infrastructure.
Comcast is opening the last mile inside Equinix Fabric, starting with Ethernet at eligible sites. Great—ordering a circuit may take days instead of weeks. Building it is still physical work, folks. The key detail is the API path: Comcast's orchestration layer plugs into Equinix Fabric under the Mplify lifecycle framework. For enterprise buyers spread across facilities, that makes Equinix a more consequential place to order service, not just a cross-connect address. And Comcast already reaches more than 700 data centers nationwide, so this isn't some cute pilot in one metro. But “eligible locations” is the catch—availability still decides whether that dashboard promise becomes an installed circuit. Business Wire wants us to picture optical wavelengths, cloud connectivity, and cybersecurity next. Fine. The first deliverable is last-mile Ethernet, and the weeks-to-days claim now has a clear operational test. From Farid Ishak at EPE:
PGRR144 is now expected to move to the ERCOT Board of Directors for the next stage of the approval process. The changes will introduce additional requirements and checkpoints for dynamic modeling throughout the large load interconnection process, with particular requirements for Large Computational Loads (LCLs).
ERCOT is finally asking giant computational loads to act like real grid participants. PGRR144 calls for ride-through testing, dynamic models, and validation against actual hardware before a facility energizes. Good. And it goes beyond the servers. The proposed model covers the whole facility—cooling, protection, controls—the gear that changes how a campus responds when the system takes a hit. TAC recommended it; the ERCOT Board is next. Bring ERCOT a model in an approved tool—PSSE, PSCAD, or TSAT—that resembles the building you actually plan to operate. No more billion-dollar load behaving like a tidy little spreadsheet box. ERCOT even gives Large Computational Loads their own category. The grid has noticed that a data center full of converters isn't interchangeable with a cement plant. Buried in Broadcom's latest 10-Q is a disclosure that changes how the company sees its AI accelerator business—not just as a chip supplier, but as a platform architect. Listen to how they describe what they've built.
We established the AI XPV platform with certain sophisticated financial partners to enable more than 20 gigawatts in compute capacity using our custom AI accelerators or XPUs and networking solutions customized for the leading frontier AI labs through 2028.
So Broadcom isn't waiting for hyperscalers to bring their own capital to the table. It's pre-assembled the financing stack, which puts some of the deployment risk and deal cadence on Broadcom and its financial partners, not just the customer. That's a very different go-to-market from simply selling silicon. Keep an eye on how the $35 billion initial tranche gets deployed, and whether this financial-partner model attracts other frontier labs. If it scales, Broadcom could have a new way to capitalize AI compute build-outs outside the traditional hyperscaler procurement cycle. From Latitude Media:
Building interregional transmission lines in every part of the U.S. is slow. But things are worse in the West. Projects can take two or three times as long as they do in other parts of the country, due in part to the region’s vast tracts of federal land that trigger permitting reviews by an alphabet soup of government agencies, as well as the fact the region has no central grid planner.
The final Idaho leg of a 575-mile line proposed 30 years ago breaks ground this month. And we wonder why a data-center campus can find servers faster than it can find deliverable power. SWIP's first segment entered service in 2014; Latitude says the northern leg is slated for 2028. Federal right-of-way and NEPA cleared early, then county fights, rate disputes, and cost allocation ate up the calendar. The West has one ISO—California—and no central grid planner for the region. So every utility can agree transmission is urgent, right up until somebody has to pay for wires crossing their county. Breaking ground matters, but it doesn't solve the problem. If SWIP reaches service in 2028, it will be a real north-south bridge—just on a timeline that should make anyone forecasting new Western load a little less casual. If your team needs this kind of briefing for your own industry, Lantern makes private daily podcasts covering your competitors, market, or beat, delivered to your whole team's private feed. Learn more at lantern podcasts dot com slash briefings, with a 14-day free trial.
We're watching PGRR144 as it moves to the ERCOT Board of Directors, and SWIP's northern leg as it heads toward a planned 2028 launch, assuming the project stays on schedule.
Links to every story are in the show notes—take a look at the ones you want to explore further. Thanks for listening, and we'll be back with the next episode on Monday. That's it for The Data Center Daily today. This is a Lantern Podcast.