How Bitcoin Miners Became
Texas's Most Valuable Grid Asset

On July 22, 2026, the ERCOT grid served a preliminary 91,308 MW. That is the most electricity Texas has ever consumed in a single moment, beating the August 2023 record by roughly 5,800 MW. ERCOT issued no emergency alerts. Reserves never dropped below 10 GW. The grid operator described conditions that afternoon as normal.

Five years earlier, the same grid killed at least 246 people and came within four minutes and thirty-seven seconds of uncontrolled collapse.

Part of what changed is more generation, dispatchable and renewable. Part of it is battery storage. And part of it, the part almost nobody outside the energy industry understands, is that Texas built the most responsive industrial demand response fleet in North America. Its anchor tenants are Bitcoin miners.

This document traces how ERCOT integrated large-scale mining operations as Controllable Load Resources, industrial consumers that can shed hundreds of megawatts on dispatch in seconds. It covers the program mechanics operators actually get paid through, the audited financial record, the peer-reviewed research, the criticisms worth taking seriously, and the regulatory shift that converted this model from a market opportunity into Texas law.

The core finding: interruptible industrial load is a grid asset with a measurable market price, and Texas is the first jurisdiction to prove it at scale and then legislate it.

Grid balancing got harder, and the reason is structural

A grid operator's job is to keep generation and consumption equal at every instant. For most of the last century that was manageable, because a small number of large plants ran steadily and demand followed predictable daily and seasonal curves.

That world is gone. Wind and solar now supply a large share of ERCOT's energy, and they produce when weather allows rather than when the grid needs it. Meanwhile the demand side has stopped being predictable too, as data centers and industrial electrification pile new load onto the system faster than transmission can be built to serve it.

Analysts at K33 Research framed the underlying shift precisely: the supply side's ability to provide system flexibility is decreasing as controllable generation gives way to weather-dependent generation. The International Energy Agency's Sustainable Development Scenario projects non-controllable sources rising from 11% to 42% of the global generation mix by 2040. Something has to replace the flexibility that retiring dispatchable capacity used to provide. The options are storage, transmission, or demand.

The duck curve, and why Texas has its own version

California's grid operator popularized the chart that made this legible. Plot net demand, meaning total demand minus solar output, across a day of high solar penetration and the curve sags through the middle of the day, then ramps violently at sunset. The shape looks like a duck.

Texas has the same problem with a wind accent. ERCOT's wind fleet frequently overproduces overnight and in the early morning, when demand is low and transmission out of West Texas is constrained. When supply exceeds what the grid can move or absorb, the operator has two choices: curtail generation and waste it, or find a load willing to consume it on short notice at whatever price clears.

ERCOT has an additional constraint that makes this sharper than anywhere else in North America. The Texas grid is islanded, with almost no interconnection to the Eastern or Western grids. It cannot import its way out of a shortfall or export its way out of a surplus. Whatever balancing happens has to happen inside Texas.

Bitcoin miners turned the second option into a functioning market.

On February 10, 2021, a polar vortex settled over Texas. Over six days the ERCOT grid suffered the worst failure in its history. Gas wellheads and pipelines froze. Wind turbines iced. Heating demand surged into a system losing generation by the hour. At the depth of the crisis roughly 34,000 MW of installed capacity, about a third of the fleet, was unavailable.

ERCOT ordered 20,000 MW of rolling blackouts, the largest manually controlled load shed in United States history. The grid came within four minutes and thirty-seven seconds of uncontrolled collapse. Had that happened, cascading equipment damage would have left much of Texas dark for weeks rather than days. At least 246 people died. Estimated economic damage reached $300 billion.

The structural failure wasn't the weather. It was that ERCOT had almost no flexible demand. When generation fell away, the operator had no way to reduce consumption quickly and precisely. The only tool available was cutting power to entire neighborhoods.

The price signal during Uri illustrates why the loads that could respond, did. K33 Research calculated that across the seven-day storm the average ERCOT power price ran $5,972 per MWh against Bitcoin mining revenue of roughly $480 per MWh. A miner who kept machines running would have spent close to ten times more on electricity than that electricity could earn. Miners powered down and sold their contracted power back to the grid. Texas conventional data centers, serving latency-sensitive customers, did not.

The conventional post-crisis response would have been to build peaker plants, gas turbines that sit idle most of the year and fire during scarcity. Planners costed that buildout at roughly $18 billion.

Texas went a different direction. Rather than fund idle capacity, ERCOT expanded and formalized its Controllable Load Resource program, a framework that had existed on paper since 2004 without ever being used at meaningful scale. The logic states in a sentence: if you can't reliably add supply during a crisis, reduce demand instead. The hard part is that reducing demand on command requires loads that are pre-registered, telemetry-connected, and contractually obligated to curtail. In February 2021, almost nothing on the ERCOT system met that description.

Most industrial load cannot actually be interrupted

Demand response has always run into a physical constraint: most large industrial processes cannot stop mid-cycle. A steel mill cannot pause a heat. A semiconductor fab cannot abandon a wafer run. Chemical processors managing exothermic reactions cannot cut power and walk away safely. Cold storage has thermal mass to work with, which is why it is one of the few genuinely flexible industrial verticals, but its shiftable window is measured in hours and bounded by product spoilage.

K33 Research decomposed interruptibility into four factors and scored mining against a conventional data center and a steel plant. Cost of reacting: for a miner, the only cost is hashprice not earned, which is calculable in advance. Reaction time: a mining facility runs one process, SHA-256 hashing, which can be interrupted instantly. Availability: mining load is stable and always present, unlike a steel plant whose energy-intensive stages come and go. Granularity: a mining fleet adjusts load in near-infinite increments, where a steel plant's arc furnace is effectively binary.

Luxor Technology, which builds the firmware and fleet management software much of the industry runs on, puts numbers to the speed claim. Its LuxOS firmware can drop a miner to roughly 25 watts in five seconds and restore it in ten. That is fast enough to qualify for ERCOT's most demanding product.

The CLR designation and who has earned it

ERCOT created the Controllable Load Resource designation in 2004. No electricity consumer met the requirements until 2020, when demand response technology firm Lancium qualified a Bitcoin mining load. As of K33's September 2022 analysis, crypto miners were the only load type that had earned the CLR designation at all.

Registration is not a formality. ERCOT requires demonstrated response capability, real-time telemetry into ERCOT's grid management systems, a registered Qualified Scheduling Entity, and the ability to curtail to a specified load level on dispatch. This distinction matters more than any other point in this document. A registered CLR with working telemetry and automated dispatch is a grid asset. An unregistered miner buying cheap power is a large load. The hardware is identical. The difference is infrastructure, market registration, and operating discipline.

Most coverage of this topic collapses everything into the phrase demand response. Operators experience it as several distinct revenue mechanisms with different response requirements, different compensation, and different infrastructure prerequisites. Understanding the stack is the difference between an operator capturing full value and one leaving most of it unclaimed.

The ancillary services ladder

Luxor Technology's energy research team, which operates a registered Retail Electric Provider and Level 4 Qualified Scheduling Entity in ERCOT, maps the program stack against required response speed.

The ladder is structured this way because the grid needs different coverage for different failure modes. Non-Spin and ERS-30 address longer-duration shortfalls where there is time to coordinate. Responsive Reserve Service exists for sudden generation loss, where frequency deviation starts immediately and response has to be automatic.

The practical consequence is that infrastructure investment buys program access. A fleet that curtails on a manual signal qualifies for the slower programs only. A fleet with automated dispatch qualifies across the full stack including RRS, which carries among the higher capacity payment rates in the ancillary services market. Automation is not an efficiency upgrade. It is an eligibility requirement.

Capacity payments and performance payments do different work

Two mechanisms operate simultaneously, and conflating them produces bad models.

Capacity payments accrue for enrolled availability whether or not a dispatch is ever called. They function as an ongoing offset against energy cost during normal full-load operation, which means the effective cost of power during non-event hours is lower than the nominal contract rate.

Performance payments convert curtailment into revenue. ERCOT dispatch events cluster around high-price, high-stress hours, exactly the intervals when a rational miner would curtail anyway to avoid elevated spot prices. Enrollment converts that curtailment from cost avoidance into compensated load reduction. The economic effect of participation therefore spans the entire operating period, not only the hours when events occur.

Four Coincident Peak, the mechanism outsiders miss entirely

ERCOT allocates transmission cost among large industrial customers through a mechanism called Four Coincident Peak. ERCOT records the four highest 15-minute load intervals across June, July, August, and September. A site's average load across those four windows sets its peak load contribution, which is multiplied by the following year's PUCT-set transmission rate and billed monthly for the next twelve months.

Luxor's analysis puts full-load 4CP exposure at up to roughly $50,000 per MW per year. A 25 MW site running through every peak faces roughly $1.25 million in transmission charges across the following twelve months. The charge scales linearly, so full curtailment across all four intervals brings it close to zero.

ERCOT does not announce the peaks. They are confirmed only after the season closes, which means operators are forecasting them in real time. The timing is more predictable than the dates: the 2025 peaks all landed between 16:00 and 18:00 Central, and the band rarely strays outside that range. One peak falls in each of the four months.

A single correctly forecast 15-minute interval can be worth more to a site's annual power cost than weeks of ordinary curtailment. This is the clearest illustration of why demand response in ERCOT is an operating discipline rather than a passive rate structure.

The aggregate effect is now visible in Bitcoin network data. Texas hosts roughly 17% of global hashrate as of mid-2026, and synchronized 4CP curtailment shows up as a seasonal signature. Across 2022 to 2025, difficulty adjustments during the June to September 4CP months averaged +0.53%, against +2.34% across the rest of the year. In June 2025 the network recorded two negative difficulty adjustments, down 0.45% on the 14th and down 7.48% on the 29th, the first back-to-back negative epoch since July 2024. During likely 4CP peak hours that month, block times averaged roughly 51 seconds slower than non-peak hours. Forward hashrate markets now price this seasonality directly, with June through September contracts trading at a premium to spot while the rest of the curve trades at a discount.

Riot Platforms, the documented case

Riot operates the largest Bitcoin mining complex in North America across its Corsicana and Rockdale sites in central Texas, carrying 1.7 gigawatts of fully approved firm power. That makes Riot one of the single largest load customers on the ERCOT system.

In August 2023, during a severe heat event, Riot curtailed more than 95% of its operations across peak demand periods and booked $31.7 million in power and demand response credits for the month. The figure became the reference point for CLR economics, and it exceeded the total credits Riot received across all of 2022.

What the $31.7 million was actually composed of

The headline number was widely misreported at the time as a payment from ERCOT, and the correction is worth stating because it shapes how the economics should be modeled. In a memorandum responding to press coverage, Riot disclosed that roughly $7.4 million came through the ERCOT ancillary services program. The remaining $24 million came from a curtailment agreement negotiated with its retail electric provider, TXU, under which contracted power earmarked for Riot is released back to TXU's residential customer base during curtailment. TXU buys that power back because doing so is cheaper than purchasing on the wholesale market when prices approach the cap, then set at $5,000 per MWh.

Riot also noted that its ancillary services premium amounted to less than one percent of a program that administered close to $1 billion over the same period. Riot's curtailment returned approximately 84,000 MWh to the ERCOT market that August.

Two conclusions follow. First, the revenue is real and auditable but it is not a single-source ISO payment, and an operator modeling it needs to understand both the wholesale-market and retail-contract components. Second, the demand response program Riot participated in is far larger than any single participant, which undercuts the framing of miners as uniquely subsidized.

The scaled result

August 2023 was the proof. The scaled result came later. Across full-year 2025, Riot booked $56.7 million in power curtailment credits and finished the year at a net all-in power cost of $0.037 per kWh. For context, Riot's disclosed 2023-era comparison put its power cost at roughly 2.8 cents per kWh against 7.2 cents for other Texas industrial users and 13.5 cents for Texas residents.

The monthly detail shows how lumpy this revenue is, which matters before anyone models it as an annuity. August 2025 produced $15.2 million in power credits plus $0.9 million in demand response credits, with all-in power cost falling to 2.6 cents per kWh. September 2025 produced $1.4 million total and power cost rose to 4.2 cents. Curtailment revenue tracks scarcity, and scarcity is seasonal.

The industry-wide picture

Riot is the most visible participant but not the only one. During the July 10, 2022 heat event, sixteen firms collectively curtailed roughly 1,500 MW. In July 2023, mining curtailment freed more than 50,000 MWh. Across 2023 as a whole, Bitcoin miners in ERCOT curtailed approximately 888 GWh.

The demand response fleet miners participate in is substantially larger than mining itself. ERCOT documentation cited during the 2023 heat events counted more than 600 enrolled load resources totaling roughly 7,000 MW of capacity across all industrial participants.

Daniel Batten's BEEST model, which reconstructs Bitcoin network energy use from the bottom up rather than relying on aggregated pool data, produced an independent read on ERCOT mining scale. Contacting operators directly, Batten identified ten major mining companies on the ERCOT grid drawing 3.258 GW of contracted capacity. Critically for this analysis, he applied a 94% uptime assumption to ERCOT miners specifically, against 98% for miners elsewhere, on the grounds that ERCOT participants curtail materially more because of demand response participation. Average actual draw came to 3.06 GW. That uptime gap is demand response showing up in the data as a measurable, distinguishable behavior.

ERCOT's own mid-2026 figures show 5.9 GW of observed energized large loads with another 3.2 GW approved but not yet operational. Of the 8,926 MW holding approval to energize, ERCOT observed a non-simultaneous monthly peak of 3,966 MW in June 2026. Total demand response capacity available for summer 2026 came to roughly 3,500 MW, about 4% of normal peak demand.

Pablo Vegas, ERCOT's chief executive, has said publicly that large flexible loads are necessary to grid reliability and that he wants more CLR-registered load connected. Brad Jones, ERCOT's interim chief executive during the program's formative period, told CNBC in March 2022 that crypto miners help ERCOT keep renewable energy operational. These are operational judgments from the people accountable for keeping Texas lit.

Grid stability effects

Academic work has independently confirmed what ERCOT operators observed. Menati, Lee, and Xie at Texas A&M, publishing in IEEE Transactions on Energy Markets, Policy and Regulation in 2023, built a synthetic Texas grid model calibrated to ERCOT data and found that cryptocurrency mining loads stabilize grid frequency, reduce reserve requirements, and dampen wholesale prices during stress events. The same work found that price-responsive mining lowers the total cost of ancillary service procurement across the market, not only for the miners themselves.

Curtailment is not a binary, and this is the important finding

The most consequential recent research on flexible compute loads comes from Dr. Veronica Margot Paez, whose doctoral work at Georgia Institute of Technology was defended in April 2026 under advisor Dr. John E. Taylor in the School of Civil and Environmental Engineering. Her committee included Dr. Neda Mohammadi and Dr. Joe F. Bozeman III of Georgia Tech, and Dr. Troy Cross of Reed College. Paez and Cross are both Fellows at the Bitcoin Policy Institute. The dissertation builds a multi-scale framework for measuring the energy and emissions impact of flexible high-performance computing and cryptocurrency data center loads, and it complicates the industry's own talking points in ways worth taking seriously.

The empirical core uses hourly electricity consumption data from 21 North American cryptocurrency data centers, matched against real-time locational marginal emissions data. That pairing is what separates the work from what came before. Most emissions analysis treats curtailment as a binary, assuming any load reduction is a carbon reduction. Grid carbon intensity moves hour to hour depending on which generator is setting the margin, so the assumption fails in practice.

Paez develops engineered metrics for curtailment dynamics and emissions alignment that test whether a facility's downscaling coincides with high marginal emissions periods, when coal and gas peakers are on the margin, or with low-emissions periods when renewables are in oversupply. Curtailing during the second is close to worthless from a carbon standpoint even though it looks identical on a load chart.

The policy implication runs against how most demand response programs are built. Programs that reward flat capacity reserves pay for availability without regard to whether the curtailment does environmental work. Programs that want emissions outcomes have to price timing and depth alongside megawatts.

Behavioral signatures without proprietary data

A second study in the dissertation, presented at the ASCE i3CE 2026 conference, introduces a data-light heuristic that separates facilities into statistically distinct operational types using only time-series energy data. The method applies the Kneedle algorithm to identify curvature knee-points in power consumption, producing what Paez calls a behavioral signature.

The value here is regulatory. Grid operators, utilities, and analysts can classify and benchmark a facility's flexibility without access to confidential hardware specifications or proprietary workload logs. The approach works on both cryptocurrency data centers and general high-performance computing facilities, which matters now that AI load is entering the same markets under the same rules.

Load as an emergent property

The third study builds a block-level cryptocurrency network simulator that endogenizes miner economics, hardware efficiency, electricity costs, difficulty adjustment, and halving dynamics. It can hindcast and forecast hashrate, revenue, and total energy draw under alternative price and policy scenarios.

For anyone operating in this market, that sentence is the strategic thesis stated academically. A mining facility is not inherently good or bad for a grid. What it is depends on how it's registered, how it's dispatched, and how well its operating behavior aligns with actual grid conditions. The same building can be a liability or an asset depending on the infrastructure behind it.

Measuring the network honestly

Daniel Batten, co-founder of CH4 Capital and the most-cited independent analyst working on Bitcoin energy data, has spent several years on the measurement problem underneath all of these debates. His BEEST model, the Bitcoin Energy and Emissions Sustainability Tracker, was built to reconcile a wide gap between the Cambridge Centre for Alternative Finance estimate of the network's zero-emission energy share and the industry-reported figure.

The reconciliation is methodological rather than rhetorical, and Batten is explicit that CCAF acknowledges the relevant limitations. The Cambridge model excludes off-grid mining, flared and vented gas mining, and waste heat recovery, and its mining map dataset for the United States drew only on a single pool that many large Texas miners do not use. Batten replicated the CCAF result almost exactly, at 37.1% against their 37.6%, then step-added the excluded categories through direct operator contact. Off-grid sustainable mining alone accounted for a 10.8 percentage point difference. His model published a lower-bound figure of 52.6% zero-emission energy, and the raw data file is public.

The ERCOT-specific finding matters most for this case study. Batten's bottom-up count put ERCOT miners at 20.81% of global hashrate and 56.22% of United States hashrate, well above what the Cambridge mining map implied, because the pool-based dataset underrepresented Texas. He also estimated the ERCOT grid at 46.81% sustainable sources as of early 2023, against 40.3% for the United States grid overall. Miners concentrated in ERCOT are therefore operating on a cleaner-than-average grid and curtailing on it more often than miners elsewhere.

Batten's other line of work is peer-reviewed. Rudd, Jones, Sechrest, Batten, and Porter published an integrated landfill-gas-to-energy and Bitcoin mining model in the Journal of Cleaner Production in 2024, using Monte Carlo simulation to test the economics of pairing mining with methane capture. The paper estimates that a 1.14 MW facility could mitigate 2,187 metric tons of methane, valued at $7.6 million, and the reproduction code is published openly. The relevance to ERCOT is indirect but structural: it is the same argument that flexible compute monetizes energy that would otherwise be wasted or vented, applied to a different waste stream.

Miners add load and push up wholesale prices

Rice University's Daniel Cohan and others have argued that cryptocurrency operations, which consumed 14.7 million MWh in Texas in 2024, put upward pressure on wholesale electricity prices by tightening supply and demand balance during normal operating hours.

This is partly correct and deserves a straight answer. Adding any large consumer to a grid tightens the balance and tends to lift prices in normal conditions. The Texas A&M research adds geographic nuance, finding the price impact of mining load highly non-uniform, concentrated near facilities and modest system-wide. The honest framing is that the question isn't whether miners add load. They do. The question is whether their curtailment participation returns enough value to justify it. The 888 GWh curtailed in 2023, the $56.7 million Riot booked in 2025, and a record-demand day in July 2026 that passed without an emergency alert sit on the other side of that ledger.

Growing miner load creates future grid risk

This is the strongest objection in the debate and it has gotten stronger as the interconnection queue has grown. If a large share of proposed load connects without curtailment capability, ERCOT ends up with the demand growth and none of the flexibility. ERCOT board members and grid analysts have raised exactly this concern, and they are right to.

It is also the objection Texas has now legislated an answer to, which the next section covers.

Miners don't actually help the grid

This one doesn't survive the operating record. ERCOT's chief executive has publicly asked for more CLR-registered load. Two successive ERCOT chief executives have said on the record that miners support grid operations. The grid has absorbed multiple major stress events without residential blackouts since CLR participation scaled. The peer-reviewed literature supports the frequency and reserve effects.

Texas legislators tested the proposition directly. In 2023 a bill came before the Senate Committee on Business and Commerce that would have prohibited flexible load agreements between miners and energy providers, effectively banning the arrangements this case study describes. Pierre Rochard, then Riot's VP of Research, testified that the industry was a net positive for the state, noting among other things that Riot had become the largest employer and largest taxpayer in Rockdale. The bill did not become the law of Texas. Senate Bill 6, which did, moved in the opposite direction by making curtailment capability mandatory rather than prohibited.

There is a real debate about price impacts, and a real debate about whether curtailment timing delivers the emissions benefits the industry claims, which is precisely what the Paez research is designed to measure. Whether flexible load helps grid reliability is not part of it.

Between June 2025 and July 2026, the ERCOT demand response market stopped being a voluntary opportunity and became a regulatory requirement. Four developments did it.

Senate Bill 6, signed June 21, 2025

Governor Greg Abbott signed SB6 into law on June 21, 2025. It is the most significant regulatory change in the history of ERCOT's demand response market, and it converts what miners demonstrated as best practice into a legal obligation.

1. Every new large load interconnecting in ERCOT after December 31, 2025 must have curtailment protocols in place, allowing curtailment during firm load shed events.
2. ERCOT can issue emergency notices directing large loads with backup generation to reduce load or self-generate, after market-based options are exhausted.
3. Large loads with backup generation capable of serving 50% or more of on-site demand must disclose those facilities to ERCOT and the interconnecting utility.
4. The PUCT must direct ERCOT to build a competitive reliability service procuring demand reductions from large loads above 75 MW, with at least 24 hours notice before deployment.
5. Initial transmission screening study fees are set at a minimum of $100,000, with additional fees at higher requested capacity.

Curtailment capability is now the price of admission to the ERCOT grid. Operators who built that capability voluntarily, over years, hold a structural advantage over new entrants assembling it under a compliance deadline. The PUCT must complete its related rule review by December 31, 2026.

RTC+B went live December 5, 2025

ERCOT replaced the Operating Reserve Demand Curve with Ancillary Service Demand Curves and moved to Real-Time Co-optimization plus Batteries. Security-Constrained Economic Dispatch now co-optimizes energy and ancillary services together, every five minutes.

For flexible load operators the practical consequence is that manual price-watching no longer works. The old ORDC adder that lifted energy prices during tight reserve conditions has been replaced by adders on real-time ancillary service prices, so the revenue accrues to resources registered in the right products and able to respond automatically. Batteries now compete in the same co-optimized dispatch. The market got more sophisticated, and the advantage shifted further toward operators with real telemetry and automated response.

The interconnection queue went vertical

ERCOT entered 2026 with more than 233 GW of large-load interconnection requests, up 269% year over year, roughly 77% of it data centers. By March 26, 2026, ERCOT's hearing materials put the figure near 410 GW, about 87% data centers. Q1 2026 alone brought 198 new large load applications, the highest single quarter on record, as developers rushed to file ahead of tightening rules and financial requirements.

Nobody should read 410 GW as a forecast. ERCOT's preliminary Long-Term Load Forecast released in April 2026 projects roughly 367,790 MW of demand by 2032, and its summer 2026 peak range of 90,500 to 98,000 MW landed against an actual of 91,308 MW. The queue is the top of a conversion funnel, not a build plan.

What the queue does establish is that firm power capacity with secured interconnection has become the scarcest asset in Texas. Riot's 1.7 GW illustrates it. In 2025 Riot signed a lease with AMD worth $311 million over a ten-year base term, delivering roughly $25 million in average annual net operating income. That transaction converts infrastructure originally built for Bitcoin mining into hyperscale AI data center capacity, and it prices what the mining industry's power portfolios are worth.

Provisional CLRs extend the model to AI

In November 2025 ERCOT filed PGRR134, creating a new participant category called the Provisional Controllable Load Resource. PCLRs are a connect-and-manage pathway: a large load can energize ahead of full transmission buildout in exchange for accepting curtailment obligations that let ERCOT manage the grid impact.

Structurally, a PCLR is the CLR model applied to AI data centers. Participation is voluntary but the curtailment commitment is firm. A locational pricing formula ties the resource to its actual grid impact, keeping large new loads from destabilizing local prices. PCLRs are dispatched after generators and before firm load shed. And they carry responsibility for managing their own consumption differential through batteries, backup generation, or workload shifting.

The broader Batch Zero framework governing large load interconnection reached ERCOT board approval under PGRR145 and NPRR1325, effective July 10, 2026. Projects of 75 MW and above are now studied together in tranches rather than sequentially.

A July heat wave pushed large parts of Texas past 100°F with heat index readings up to 108°F. Demand broke the all-time record twice in 48 hours. Tuesday, July 21 hit 87,403 MW, beating the three-year-old August 2023 record of 85,508 MW. Wednesday, July 22 reached a preliminary 91,308 MW at 5 p.m.

ERCOT issued no Energy Emergency Alerts. Reserves held above 10 GW through the peak, against alert thresholds of 2,300 MW for Level 1 and 1,430 MW for Level 3. Available generation read 113,543 MW against 90,027 MW of demand at 3 p.m. that afternoon. Earlier in the summer, demand response curtailed roughly 2 to 3 GW to hold peaks under 80 GW.

Compare that to August 2023, when a lower absolute record came with far tighter reserves and a grid flirting with emergency conditions. The 2026 event was larger and calmer. That difference is what five years of generation buildout, storage deployment, and flexible load integration bought.

The counterweight belongs in the same paragraph. ERCOT's own Capacity, Demand and Reserves report from May 2025 projects summer planning reserve margins declining through 2030, crossing into negative territory for peak net-load hours as early as 2026 or 2027. The EIA has warned that data-center-driven demand could push ERCOT North Hub wholesale prices as much as 79% above baseline under a high-demand scenario. A grid that served 91 GW comfortably is sitting on a demand curve that could run several multiples higher within the decade.

Both things are true. The system handled a record without stress, and the margin it handled it with is forecast to shrink every year through 2030. That gap is precisely where flexible load earns its money.

ERCOT is not a Texas-specific anomaly. The model applies wherever three conditions hold together: excess or stranded generation that can't reach load centers, a grid operator under pressure to manage peaks without building generation, and a regulatory framework that lets large loads register as controllable resources.

Those conditions exist across SPP, MISO, NYISO, ISO-NE, and PJM, in different forms and with different market mechanics. The ancillary service products differ. The payment structures differ. The underlying economics, that perfectly interruptible load carries a market-clearing value which can exceed the cost of curtailing, transfers cleanly. The IEA has estimated that meeting Paris Agreement targets requires roughly 500 GW of demand response capacity globally by 2030, against a 2020 baseline roughly a tenth that size.

For an operator holding stranded wind in the SPP wind belt, underutilized capacity at a legacy thermal asset in MISO, or a new interconnection facing PJM capacity prices, ERCOT supplies the financial model and the regulatory template.

What replication actually requires

The Paez research is a useful corrective for anyone assuming this is a hardware question. It isn't. Facilities running identical machines produce measurably different grid and emissions outcomes depending on when, how often, and how deeply they curtail, and those differences are detectable from operating data alone.

The gap between a CLR-registered operation booking Riot-scale curtailment revenue and an unregistered miner buying cheap power comes down to four things: telemetry giving the ISO real-time visibility, automated dispatch response inside the required window, market registration across the right ancillary service products, and an energy management function treating curtailment as a revenue line rather than an emergency measure.

The returns to getting the last of those right are measurable. Luxor's backtesting of intelligent curtailment strategies in ERCOT found 8 to 14% higher profitability against legacy on-off approaches. In a documented deployment across Soluna's Texas fleet, faster curtailment recovery and improved fleet-level efficiency produced roughly $937,000 in annualized revenue uplift on identical hardware. Same machines, same power contract, different operating layer.

That list is the actual barrier to entry, and it is why the SB6 compliance deadline creates a real opening. Most of the interconnection queue has never built or run a CLR-compliant system. The PUCT deadline is December 31, 2026. The expertise required to meet it is scarcer than the capital.

1. The balancing problem that flexible load solves is not specific to Texas. It's the defining operational challenge for every grid integrating significant renewable generation, and AI-driven load growth is making it harder faster than planning models anticipated.
2. Bitcoin miners are a documented, financially material solution. Riot's $56.7 million in 2025 curtailment credits is an audited figure in SEC filings, not a marketing number. The grid performance record across multiple stress events is public. The academic literature supports the frequency and reserve effects, and the independent measurement work supports the claim that ERCOT miners curtail materially more than miners elsewhere.
3. The model transfers to any market with stranded power and a flexible load registration framework, which currently describes SPP, MISO, NYISO, ISO-NE, and PJM.
4. The technical bar is rising. RTC+B, the SB6 flexibility mandate, and the Batch Zero framework all raise the cost of participation. Operators who invested in telemetry, automation, and market sophistication before the mandates arrived are the ones positioned to clear it.
5. And the development that changes the strategic picture: the flexibility model miners built as a competitive edge is now the entry requirement for the largest wave of new electricity load in Texas history. Every AI data center seeking ERCOT interconnection under Batch Zero has to demonstrate controllable load capability. The operators who learned to do that for Bitcoin mining are now the infrastructure providers, advisors, and landlords to the AI buildout.

Texas didn't set out to build a demand response market. It set out to avoid a repeat of February 2021. What it got was proof that a grid can be balanced from the demand side, and a legal framework that makes that the default. On July 22, 2026, that framework served 91,308 MW without anyone noticing.