Nvidia’s hometown just changed the rules for how AI data centers get plugged into the grid. A report submitted to the Santa Clara City Council on September 8, 2026 lays out a program that lets large power users connect to the electrical grid faster than under existing timelines, on one condition: they have to let the utility throttle or cut their electricity when the system is under strain. The report was first flagged by South Korean outlet Maeil Business Newspaper (매일경제) and has since been picked up by outlets covering the AI infrastructure beat.
The city runs its own electric utility, Silicon Valley Power (SVP), and the program at the center of the report is called the Flexible Load Interconnection Program, or FLIP. Under FLIP, a data center can skip years of queue time to get connected, but SVP keeps the ability to dial back or cut that customer’s supply the moment the grid gets tight. It’s a trade: speed now, control later. And it’s happening in the same city where Nvidia keeps its headquarters, a detail that hasn’t been lost on anyone covering the AI power crunch.
Santa Clara’s Report Signals a Shift in How AI Gets Powered
The September 8 filing to the Santa Clara City Council doesn’t read like a typical utility memo. It describes a supply model where electricity flows to data centers under normal conditions, but the moment the grid can’t keep up, SVP can step in and directly reduce or halt that customer’s draw. Coverage of the filing calls this an unusual, even rare, approach to powering large electricity users, and the framing makes sense once you consider where this is happening: Santa Clara, in the heart of Silicon Valley, is also where Nvidia is headquartered.
That geography matters for the story. Nvidia’s chips are the engine behind the current AI buildout, and the company’s own backyard has become a proving ground for what happens when data center demand outpaces the physical grid. SVP, as a municipally owned utility, has more room to experiment with unconventional supply arrangements than an investor-owned utility answering to shareholders and a state regulator. FLIP is that experiment.
What Silicon Valley Power’s FLIP Program Actually Does
FLIP is aimed at large electricity users, chiefly data centers, that would otherwise sit in a multi-year interconnection queue waiting for SVP to confirm it has enough capacity to serve them at full, firm power. Instead, FLIP lets those customers connect sooner, in exchange for agreeing to reduce their electricity usage whenever SVP requests it during periods of grid stress.
The arrangement isn’t purely voluntary once you’re enrolled. If a customer doesn’t cut its usage enough after SVP asks, the utility can directly limit or cut off the electricity supply to that site. That’s the part that separates FLIP from a standard demand-response rebate program, where a customer that ignores a request simply forfeits a bonus payment. Here, the utility holds the switch.
The goal, according to the council filing, is straightforward: the pace of AI data center construction has outrun the pace at which the power grid can expand. Rather than making every new data center wait for a multi-year substation upgrade, SVP is offering a faster on-ramp to power that comes with strings attached.
Connect First, Curtail Later: The Mechanics Behind FLIP
Under a conventional interconnection process, a utility studies a proposed data center’s full projected load, confirms the local grid can carry that load at all times, and only then approves a firm power connection. That study process is exactly what’s been creating multi-year backlogs across the country as AI data center proposals pile up faster than substations and transmission lines can be built.
FLIP flips that sequence. A customer gets connected against the grid’s current capacity, not its theoretical peak capacity, and accepts a ceiling that can move up or down based on real-time conditions. When there’s spare capacity on the system, the customer draws power close to normal. When demand spikes or a transmission or distribution problem crops up, SVP asks for a cut, and the customer is contractually bound to deliver one.
That requires the data center operator to have some way of actually reducing its power draw on short notice without simply shutting down. That’s where a company called Emerald AI comes in. The council filing names Emerald AI as a partner SVP has engaged to help build out FLIP, and the utility is developing a pilot with a major customer to work out the program’s exact parameters before asking the council for broader authority to expand it.
Emerald AI’s $150 Million Bet on Software-Controlled Power
Emerald AI isn’t a household name yet, but its finances got a lot more interesting three weeks before the Santa Clara filing. On August 25, 2026, the company closed a $150 million Series A round at a $1.05 billion valuation, according to SiliconANGLE. Reporting from The Next Web put the company’s total funding to date above $220 million, a number that has climbed fast for a startup that only recently emerged from stealth.
Emerald AI’s pitch is a software layer, marketed under the Conductor name, that sits between a data center’s compute workloads and its electrical draw. Instead of treating an AI data center as a fixed block of load that either runs at full tilt or goes dark, Conductor is designed to shift, delay, or redistribute less time-sensitive computing tasks in real time so the facility’s total power draw can flex up or down without interrupting the workloads that matter most. That’s the piece of technology that makes a program like FLIP workable: a utility can only ask a customer to cut load on demand if the customer has a way to do it without simply stopping business.
The company’s backers, drawn from both climate-tech and traditional venture investors, are betting that grid-responsive software becomes standard equipment at AI data centers the same way backup generators and cooling systems already are. Emerald AI’s own site (emeraldai.co) markets the platform directly at utilities and hyperscale data center operators facing exactly the kind of capacity squeeze Santa Clara is dealing with.
Why Nvidia’s Backyard Ran Out of Spare Power
FLIP didn’t appear out of nowhere. Santa Clara has spent the past two years as a visible example of a broader problem: data centers finishing construction and then sitting there, built but not fully energized, because the local grid doesn’t have the spare capacity to serve them at full load. A Yahoo Finance report described roughly 100 megawatts of data center capacity in the Santa Clara area sitting ready for servers, accelerators, and networking gear that couldn’t be switched on because the local grid hadn’t caught up.
One of the facilities at the center of that story is Stack Infrastructure’s SVY02A campus in Santa Clara, marketed at a total designed capacity of 48 megawatts. According to Stack’s own facility materials, the site initially had roughly 12 megawatts of critical capacity available through Silicon Valley Power, delivered via a dedicated on-site substation, well short of the full 48 megawatts the campus was built to handle. That gap between built capacity and available grid power is the exact problem FLIP is trying to work around, not by building more substations overnight, but by making the load itself more negotiable.
It’s worth remembering that this isn’t a Santa Clara-only phenomenon. AI data center demand has been growing faster than transmission and substation buildouts almost everywhere hyperscalers are concentrating GPU clusters, and Nvidia’s own quarterly results, including a record $96.2 billion quarter reported this year, only add to the pressure on the utilities serving the data centers buying those chips.
FLIP Program at a Glance
| Element | Detail |
|---|---|
| Program name | Flexible Load Interconnection Program (FLIP) |
| Administered by | Silicon Valley Power (SVP), the municipal utility of the City of Santa Clara |
| Report submitted to City Council | September 8, 2026 |
| Target customers | Large electricity users, primarily data centers |
| Core trade-off | Faster grid interconnection in exchange for accepting utility-directed load cuts |
| Enforcement mechanism | SVP can directly reduce or cut off supply if a customer doesn’t sufficiently lower usage on request |
| Named technology partner | Emerald AI, engaged by SVP to help pilot the program |
| Stated purpose | Address AI data center construction outpacing grid capacity expansion |
How FLIP Compares to Traditional Utility Interconnection
Under the traditional model, a data center developer applies for interconnection, the utility runs a study to confirm the grid can handle the customer’s full projected load at all times, and only then does construction on any needed upgrades begin. That process routinely stretches past two years in AI-hub regions, and it’s part of why so many data centers finish construction long before they can actually draw full power.
FLIP swaps a guarantee of always-available full power for a guarantee of faster access to partial, flexible power. A customer under FLIP effectively becomes part of the grid’s own balancing toolkit rather than a fixed load the utility has to plan entirely around. That’s a meaningfully different deal than the interruptible-rate tariffs some utilities have long offered heavy industrial users, because those older programs are largely opt-in and financially incentivized. FLIP ties the flexibility requirement directly to the interconnection agreement itself, not to a voluntary rebate.
For data center operators, the calculation comes down to a straightforward trade: is getting racks powered on years sooner worth designing a facility that can gracefully shed load on short notice? For AI training workloads that run in tight, synchronized clusters, that’s a harder engineering problem than it sounds, which is exactly the gap Emerald AI’s Conductor platform is built to close.
Santa Clara AI Data Center Capacity Landscape
| Data point | Figure | Source, date |
|---|---|---|
| Stack Infrastructure SVY02A total designed capacity | 48 MW | Stack Infrastructure facility materials |
| SVY02A initial critical capacity available via SVP | ~12 MW | Stack Infrastructure brochure, late 2025 |
| Santa Clara-area data center capacity built but unenergized | ~100 MW | Yahoo Finance, Nov. 10, 2025 |
| Emerald AI Series A funding | $150 million | SiliconANGLE, Aug. 25, 2026 |
| Emerald AI post-money valuation | $1.05 billion | SiliconANGLE, Aug. 25, 2026 |
| Emerald AI total funding to date | $220+ million | The Next Web, Aug. 25, 2026 |
| FLIP report submitted to Santa Clara City Council | Sept. 8, 2026 | City of Santa Clara council filing |
Historical Context: The Grid Was Never Built for This
Municipal and regional grids across the US were largely sized decades ago around relatively flat, predictable industrial and residential demand growth. AI training clusters break that assumption twice over: they draw enormous continuous loads at a single site, and they can appear on a utility’s books faster than a multi-year transmission upgrade can be planned, permitted, and built. Santa Clara’s experience, with data centers physically complete but power-starved, has become one of the most visible examples of that mismatch nationally.
The industry has tried a handful of responses to that mismatch already. Some utilities have simply lengthened interconnection queues and let developers wait. Others have leaned on on-site generation, including gas turbines and, in a few cases, direct deals with nuclear plant operators, to bypass the grid bottleneck entirely. What makes FLIP distinct is that it doesn’t try to add new generation or new wires at all. It tries to make the existing grid do more by making the load itself adjustable, which is a cheaper and faster fix if the software behind it actually works at scale.
That approach also lines up with a broader shift already visible elsewhere in AI infrastructure spending, where operators are increasingly willing to trade some performance certainty for lower cost or faster access, a pattern also showing up in how memory supply constraints have forced data center buyers to rethink procurement timelines this year.
Market Impact: What FLIP Means for AI Infrastructure Investors
For real estate investment trusts and data center developers building in constrained markets, a program like FLIP changes the math on when a facility can start generating revenue. A campus that would otherwise sit half-lit for years waiting on a substation upgrade can instead start billing customers for compute sooner, even if that compute comes with a curtailment clause attached. That’s a meaningful shift for developers who’ve been absorbing carrying costs on built-but-idle capacity.
It also creates a new category of vendor risk and opportunity around companies like Emerald AI. If utilities elsewhere start requiring similar flexibility software as a condition of faster interconnection, grid-responsive power management becomes as standard a line item in a data center build as cooling or backup power, and the $1.05 billion valuation investors just put on Emerald AI starts to look less like a bet on a niche tool and more like a bet on required infrastructure.
The pressure driving all of this hasn’t let up on the chip side either. Nvidia has continued raising prices on AI server components this year amid a broader memory crunch, and utilities like SVP now have to plan around GPU clusters that draw more power per rack than the facilities they were originally designed to serve, driven in part by newer memory technology like HBM4 packed into current-generation accelerators.
Risks: What Happens When AI Training Gets Interrupted
The obvious risk with any curtailable power arrangement is what happens to the workload when the cut actually lands. AI training runs, particularly large distributed jobs spread across thousands of GPUs, are sensitive to interruption in ways that a typical industrial process isn’t. A poorly timed power reduction can force a checkpoint restart, waste compute hours, and in the worst case corrupt an in-progress training run.
That’s the specific problem Emerald AI’s Conductor platform is designed to manage, by shifting lower-priority or delay-tolerant tasks first and protecting the workloads that can’t tolerate interruption. Whether that distinction holds up in practice, especially during a genuine grid emergency rather than a planned test, is something regulators and customers alike will be watching closely as FLIP moves from pilot to broader rollout.
There’s also a fairness question buried in the arrangement. Customers who accept curtailable power get to skip the queue ahead of customers who are still waiting for a fully firm connection. That could create pressure on SVP, and on other municipal utilities considering similar programs, to justify why speed should be for sale in exchange for reliability risk, particularly if a curtailment event during a heat wave or wildfire-driven outage hits a facility running mission-critical workloads.
Industry Reaction and Signs of Wider Adoption
SVP has described FLIP as a first-of-its-kind program among US utilities, a framing that positions Santa Clara as a reference case other grid-constrained regions could study. That framing matters commercially for Emerald AI, whose fresh capital gives it room to pitch the same Conductor platform to utilities well beyond Santa Clara that are facing an identical version of the same problem: AI data center demand that arrived faster than the wires and substations meant to serve it.
Trade publications covering the utility sector, including Utility Dive, have tracked a wider pattern of utilities experimenting with flexible and curtailable interconnection terms as data center queues grow nationwide, and Santa Clara’s filing fits squarely inside that trend. Reaction from the data center industry has so far centered on the tradeoff rather than the concept itself. Nobody covering the story disputes that grid capacity is the binding constraint on AI buildout right now, a dynamic that’s also shaped how hardware vendors are pricing and shipping everything from GPUs to server platforms built around Nvidia’s newest CPU designs. The open question is whether operators running the most demanding, least interruption-tolerant AI workloads will accept curtailable power at all, or whether FLIP ends up serving a narrower slice of the market, like inference clusters and less time-sensitive batch workloads, rather than frontier model training runs.
What Comes Next: Five Predictions
- SVP will formally expand FLIP beyond its current pilot stage once the Emerald AI collaboration produces enough operating data to satisfy the City Council, likely within the next several council cycles.
- Other municipally owned utilities in grid-constrained tech hubs will study Santa Clara’s approach closely, since city-owned utilities have more flexibility to test unconventional interconnection terms than investor-owned utilities under state regulatory review.
- Emerald AI’s fresh $150 million round won’t be its last raise. Expect the company to pursue additional utility partnerships and, eventually, a larger growth round as demand for grid-responsive software scales with the number of AI data centers stuck in interconnection queues.
- Curtailable power deals like FLIP will increasingly get split by workload type, with inference and batch processing accepting flexibility terms while the most latency- and continuity-sensitive training clusters continue paying a premium for firm, uninterrupted power.
- Expect scrutiny to grow around what happens during an actual grid emergency, not a planned test, as the first real-world curtailment events under programs like FLIP become public and get compared against the reliability data centers thought they were signing up for.
Frequently Asked Questions
What is Silicon Valley Power’s FLIP program?
FLIP, short for Flexible Load Interconnection Program, is a Santa Clara utility program that lets large power users, mainly data centers, connect to the electrical grid faster in exchange for agreeing to reduce their electricity usage when the utility asks for it during periods of grid stress.
Can Silicon Valley Power actually cut off a data center’s electricity?
Yes. Under FLIP, if a customer doesn’t sufficiently reduce usage after SVP requests a cut, the utility can directly limit or cut off that customer’s electricity supply.
Why is this happening in Santa Clara specifically?
Santa Clara is home to Nvidia’s headquarters and has become one of the clearer examples of AI data center construction outpacing local grid capacity, with facilities finishing construction while still waiting for the utility to deliver full power.
What role does Emerald AI play in the FLIP program?
Silicon Valley Power has engaged Emerald AI, a data center power-management startup, as a partner to help pilot FLIP. Emerald AI’s Conductor platform is designed to shift and adjust computing workloads in real time so a facility can reduce its power draw without stopping its most important tasks.
When was the FLIP program formally reported to the city?
A report describing FLIP and the broader power-supply approach was submitted to the Santa Clara City Council on September 8, 2026.
How much funding has Emerald AI raised?
Emerald AI closed a $150 million Series A round at a $1.05 billion valuation on August 25, 2026, bringing its total funding to more than $220 million, according to SiliconANGLE and The Next Web.
Is FLIP a voluntary program for data centers?
Participation is opt-in in the sense that customers choose to enroll to get faster interconnection, but once enrolled, the requirement to reduce usage on request, and the utility’s ability to enforce that reduction directly, is a binding part of the agreement rather than an optional rebate program.
Could programs like FLIP spread to other AI data center hubs?
Coverage of FLIP describes it as a first-of-its-kind program in the US, and the underlying pressure, data center demand outrunning grid capacity, exists in nearly every region attracting large AI infrastructure investment, which makes similar programs plausible elsewhere over time.
