Texas Greenlights 2GW Data Center Without Naming the Grid's Cost or Who Pays

IREN announced Tuesday that its 2GW Sweetwater Hub, two stacked data centers totaling 1.4GW and 600MW, cleared ERCOT's Batch Zero process as Base Load, a preliminary grid-connection checkpoint that sent IREN shares up 7% and Applied Digital up 6% in sympathy.^[1] The milestone matters: grid interconnection is now, by IREN's own account, 'the scarcest input in this entire industry,' with data-center operators pricing grid access near $25 million per megawatt.^[1] But what the market celebrated and what the grid actually approved are two different things, and Texas's press release reveals almost nothing about the real transaction.

Start with what is hidden. ERCOT's Batch Zero is a queue sorting, not a power purchase. It tells you IREN submitted the paperwork and passed an early technical screen, that the project fits the grid's spatial and voltage profile. It does not tell you whether ERCOT, the local transmission owner, or the retail distributor have committed to build or upgrade substations, lines, or generation specifically for Sweetwater 1 and 2. It does not tell you the term of any grid-connection or special-tariff agreement, the demand charges, the ratchet (the minimum MW IREN must pay for whether or not it draws them), the collateral posted, or the exit fees. These are the numbers that separate a rock-solid corporate anchor from a speculative pipeline entry that collapses if hyperscaler capex slows. IREN will 'include these projects in its announced development portfolio following the execution of the relevant grid connection agreements', meaning Batch Zero clearance is not the deal itself, and the real deal remains confidential.^[2]

This opacity serves a policy function. When ERCOT or the Texas Public Utility Commission review special contracts or large-load tariffs with data-center operators, those filings are routinely sealed on the grounds of competitive sensitivity. The economics, what capacity is reserved, for how long, at what minimum take, with what collateral and exit terms, are redacted. Meanwhile, the capex the utility builds to support that load (new gas plants, transmission backbone, substation upgrades) flows into the rate base and is socialized across all ratepayers. Texas allows cost recovery before the load materializes: utilities can request 'used and useful' status for facilities not yet fully energized, and they can build ahead of forecast. The result is a two-tiered system: hyperscalers and AI operators negotiate in secret and can walk away with collateral; residential customers and small business ratepayers fund the grid in public and eat the stranded cost if the load never shows up.

The Sweetwater numbers amplify the risk. At Sweetwater 1, a 1.4GW substation was energized in 2026, but only 300MW of data-center capacity (21% of the site's eventual footprint) is under construction, targeted for Q4 2027.^[3] That gap between substation capacity and data-center load is the capex funnel: transmission and generation upgrades are built on the forecast of future load that may or may not arrive. IREN has announced a '>5GW global development portfolio' of which Sweetwater is one piece.^[3] The company is also shopping projects to other utilities and grid operators; developers typically file with multiple RTO/ISOs and utilities to hedge their options, inflating apparent pipelines. Federal and state forecasts have themselves become rate-case weapons: utilities cite AI load surges to justify new gas plants and transmission buildouts, and those justifications survive even if the load materializes slower or at smaller scale than promised. Grid Strategies and others have documented forecast revisions of 3x to 5x in recent years, driven by announced data-center pipelines that have not yet firmed into signed contracts.^[2]

The protective answer exists and is buildable now. Virginia, Ohio, and Oregon have adopted or are evaluating large-load tariffs that shift cost responsibility back to the customer class. The model: a 10+ to 30-year minimum term matched to asset life; a high demand ratchet (80, 90% of contracted capacity) so unused grid infrastructure isn't socialized; collateral posted by the hyperscaler (roughly $1.5M per MW in Virginia's model); and 100% cost responsibility for dedicated network upgrades. Weak versions of these tariffs quietly drop the ratchet or collateral, leaving ratepayers exposed. Texas has not adopted a standardized large-load tariff; special contracts are negotiated one at a time and sealed. That discretion is the vulnerability. The Texas Public Utility Commission can open a docket to establish binding rules: minimum ratchets, collateral requirements, cost isolation, and additionality standards (the grid upgrade must not have been planned anyway). It can also mandate that new data-center load accept curtailable or flexible-capacity service, data centers can operate at reduced capacity for 0.25% to 1% of annual hours (roughly one to three days a year) without harming performance, and that flexibility buys the grid 76 to 126 GW of room without new firm generation.^[2] IREN and other operators would likely prefer firm, full-time capacity; the commission can price that choice honestly by assigning the full capex cost to the customer class, not ratepayers.

The clock is open now. Sweetwater 1 remains conditional; full grid-connection agreements have not been filed. If IREN or other data-center operators approach the commission or the utilities with requests for firm generation, transmission upgrades, or special tariffs, those dockets are the intervention window. Ratepayers can demand transparency (unsealing the contract), a cost-responsibility baseline (high ratchet, collateral, and term matching asset life), and an alternative (curtailable load or bring-your-own-generation). Every GW of data-center load that accepts those terms instead of firm grid service saves ratepayers roughly $700, 900 million per GW in stranded costs if the load underperforms.