New startup ideas · AI and software · AI infra and compute

startup idea

Swingbus

An agent that clears utility interconnection queues for new data center load.

Swingbus sells public power utilities and system operators an agent service that runs grid interconnection and load studies: it ingests the network model, runs power flow, stability and protection studies with learned surrogates instead of overnight solver batches, and produces the signed study package a regulator will accept.

AI agent as a service
Government and public sector
$1-10B market
Collapses a cost 10x
Global from day one

4/5

venture judge

similar startups, last 2 years (3 all-time)

99%

of 3 nearest real companies still alive

yes

8 matching federal grants and programs

Direction supported by government programs and grants

Test it before you build it

$1,000 · 6 weeks · 20 prospects

For $1,000 and 6 weeks, prove public power utilities will sign $4,500 purchase orders for surrogate-run screening studies and that licensed engineers will stamp the results without re-deriving them.

Riskiest assumption · Engineering directors at public power utilities have discretionary budget and a queue-clearing incentive strong enough to pay for surrogate-run screening studies even though study costs pass through to applicants, and their licensed engineers will stamp the results with spot checks instead of re-deriving everything

1Focus group: who and where

The director of engineering or system planning at a municipal utility or public power district serving 25,000-500,000 meters, sitting on pending large-load interconnection requests from data center developers, paying outside consultants $20,000-100,000 per study cycle and quoting applicants 12-18 month waits

where to find 20 · The American Public Power Association member directory of roughly 2,000 public power utilities, cross-filtered with free EIA Form 861 data to find territories with recent large commercial load growth (directory); the engineering committees of joint action agencies such as American Municipal Power and UAMPS, which put dozens of member utilities in one room (channel); IEEE Power and Energy Society local chapter meetings and the ESIG fall technical workshop, where interconnection engineers already gather (community and event); municipal staff directories are public record, so named contacts are free

2Sell first, build later

A benchmark re-run of one screening study the utility has already completed: Swingbus reproduces the power flow and stability screens with learned surrogates and delivers a side-by-side accuracy memo, reviewed by a licensed PE, within 10 business days of data handoff

the ask · $4,500 per benchmark study, on a purchase order; pilot pricing of $100 per MW screened locked for year one for benchmark customers

a real yes · A real yes is a signed purchase order with a named study and a delivery date; interest from an engineer without purchasing authority, offers to evaluate for free, or an invitation to respond to a future RFP do not count

3Small experiments

The first one attacks the riskiest assumption; each ends with a number that says whether to run the next.

1. Budget and backlog discovery calls
$100 · 14 days
Book 15 calls with engineering directors found via EIA-861 filtering and public municipal staff directories. Ask three questions with numeric answers: how many MW of large-load applications are waiting, whether spend under $5,000 clears without board or RFP approval, and whether their engineer of record would stamp a surrogate-derived screening result given a spot re-derivation protocol. The founder runs all calls in two weeks.
keep going if · 8 of 15 confirm a screening backlog over 100 MW and discretionary purchase authority of at least $5,000
2. Paid benchmark study on a PO
$900 · 21 days
Offer every qualified utility a $4,500 benchmark: they send one screening study already completed in the past year, Swingbus re-runs it with surrogate models and delivers a side-by-side accuracy memo in 10 business days, bought on a simple purchase order priced under typical municipal small-purchase thresholds so no RFP is triggered. Includes an NDA, a scope letter with a delivery date, and one in-person presentation at a joint action agency engineering committee meeting.
keep going if · 3 signed purchase orders from 15 qualified conversations
3. PE stamp protocol test
$0 · 10 days
Interview 5 licensed professional engineers at firms these utilities already retain for studies, such as Burns and McDonnell, POWER Engineers and Sargent and Lundy, plus any PE of record named on the discovery calls. Ask under what written protocol they would stamp a surrogate-derived screening result: spot re-derivation of a sample, full re-derivation, or never. This directly prices the stated regulatory risk.
keep going if · 3 of 5 PEs describe an acceptable protocol of spot re-derivation on 20% or less of cases, rather than full re-derivation of every result

4Collect a deposit up front

Tesla took $1,000 refundable reservations for the Model 3 and $100 for the Cybertruck before building either: the deposit is the measurement, not the revenue.

$4,500

per prospect, refundable

how · A signed purchase order for the $4,500 benchmark, deliberately priced under municipal small-purchase thresholds so the engineering director can sign without an RFP; a PO is how this buyer moves money before procurement, and asking a public utility for a card payment would not be credible, so the PO plus a countersigned scope letter is the commitment

what it reserves · One of five slots in the first benchmark cohort, a 10-business-day delivery date, and first position plus locked $100 per MW pricing for the screening-stage pilot

refund · The PO is cancelled and nothing is owed if the side-by-side memo does not arrive within 10 business days of the utility's data handoff

target · 3 signed purchase orders from 20 conversations within 45 days

before taking money · Interconnection studies are licensed engineering work in every state, so no Swingbus output may be used for a real interconnection decision without review and stamp by a licensed professional engineer, and the benchmark must run only on studies the utility has already completed.

Go: build it if

3 purchase orders totaling $13,500, 8 of 15 utilities confirming a 100+ MW backlog with discretionary budget, and 3 of 5 PEs accepting a spot-check stamping protocol - build the screening product

Kill: stop if

Zero POs after 20 conversations, or every PE interviewed insists on fully re-deriving each result - the cost pass-through and licensure objections hold and the study cycle cannot be removed; stop

5 Scripts to run itoutreach message, landing copy, deposit terms · click to open

outreach message

Your interconnection queue is holding data center applications you cannot study fast enough, and each outsourced screening cycle costs tens of thousands of dollars and months of waiting. I'm building Swingbus: we re-run power flow and stability screens with learned surrogate models and deliver a package your engineer of record can review and stamp. The first step is a $4,500 benchmark: send one screening study you've already completed, and we return a side-by-side accuracy memo in 10 business days. Do you have 20 minutes this week to walk through your backlog?

landing page

Clear your large-load screening backlog in days, not months $4,500 benchmark on a simple PO: one completed study re-run, side-by-side memo in 10 business days Request the scope letter - five benchmark slots this quarter

deposit terms

Your $4,500 purchase order covers one benchmark re-run of a screening study your utility has already completed and reserves one of five slots in the first cohort, starting the week we receive your inputs. If the side-by-side memo does not arrive within 10 business days of data handoff, the PO is cancelled and nothing is owed. Benchmark results are for comparison only and are reviewed by a licensed professional engineer before delivery.

Would you run this test?

One tap. The yes-share feeds the Demand pillar of this idea's score; nobody sees who answered.

Budgets are out-of-pocket estimates for a team of one to three, US market. Size the deposit to the deal, and check the terms before taking money in a regulated line.

Scorecard

One score that balances how trendy the idea is, the demand for it and its potential for 100x, with competition measured relative to every other idea in the catalog. Recent startup trends first, government priorities second.

Idea Score, 0-100 · raw 59.6 x 1.60

Warm

competition: more crowded than 32% of ideas · headwind x0.84

+3.7

government priorities, secondary (37 matching grants)

Trend

Is the wave forming now? 2025-26 entrants vs 2023-24, rounds since 2025, the sector's live-batch direction, the 2026 trend analyst.

Entrants 2025-26 vs 2023-24 (similar companies)90
Rounds announced 2025+ in the sector82
Sector direction (live batch)100
2026 trend analyst75

Demand

Does anyone want it? YC's current RFS, companies already paid for something similar, the operator judge, founders' yes-rate in decks, readers who would run the test.

YC asks for it (current RFS: idea / sector)30
Someone already pays (similar companies, recent / all-time)33
Operator judge: real pain75

100x potential

Can it return a fund? The venture judge (double weight), market-size and moat axes, neighbours still alive, the technologist judge.

Venture judge75
Market size axis33
Moat axis80
Neighbours still alive81
Technologist judge100

Score = 100 x cbrt(Trend x Demand x 100x) x (1 - 0.5 x crowding) + government bonus (max 5), calibrated so the 95th-percentile idea scores 90 (order never changes). A geometric mean: a weak pillar cannot be papered over. Percentiles are among the 272 ideas in the catalog; the terms matched were clears, utility, interconnection, queues, center, load, sells, public.

The idea in full

What: Swingbus sells public power utilities and system operators an agent service that runs grid interconnection and load studies: it ingests the network model, runs power flow, stability and protection studies with learned surrogates instead of overnight solver batches, and produces the signed study package a regulator will accept. Each utility gets a hardened appliance inside the substation network so the agent works on live topology data without exporting it. Utilities pay per megawatt studied, roughly a tenth of what an outsourced study cycle costs today.
Why now: AI load is the reason the queues are jammed, and the capital is chasing power directly: Atomarine (YC S26) is building floating nuclear powered data centers at sea against YC's Fall 2026 Compute at Sea request, and Callosum raised a $100M seed led by Atomico to tackle AI compute bottlenecks; Vorelios (YC F26) shows learned physics surrogates running 1000x faster, which is the technical precondition for study automation.
Wedge: first customer and entry point: One municipal utility or public power district drowning in large-load applications: automate the screening study stage only, where the answer is a yes-no and liability is lowest.
Path to 100x: Grid study and interconnection engineering is a $1-10B global spend growing with every gigawatt of AI load, and pricing per megawatt screened means revenue tracks the buildout, not the team. The moat is distribution: studies are sold through the SCADA and protection integrators utilities already trust, and once the appliance holds a utility's network model, the same agent extends into operating and reclearing the queue.
Ceiling: Utilities treat studies as a regulated cost pass-through with no incentive to cut them, which caps price per megawatt and turns the business into slow-growing licensed software.
Closest real companies, as the generator saw them: Vorelios simulates physics 1000x faster as a horizontal capability; Swingbus is the regulated grid application of that idea, with the study package, protection settings and audit trail a regulator signs off. Atomarine and Callosum add the load that creates the demand rather than clearing it.
Main risk: Regulators require a licensed engineer to re-derive every result, so the agent saves cost but never removes the study cycle.

Five judges

Each judge scores every idea in the catalog with a named rubric; the venture judge decides whether a card is shown at all (4-5 is venture-grade).

Venture investor
4/5
Priced per megawatt studied, revenue tracks the AI buildout rather than headcount, with the appliance holding the utility's network model.
Bootstrapper
1/5
Substation appliances sold to regulated public utilities means hardware capex plus a procurement cycle measured in budget years.
Operator
4/5
Utilities already pay outsourced study cycles at ten times the price and the queue is visibly jammed; the screening-only wedge fits the existing regulatory step.
Technologist
5/5
Learned surrogates replacing overnight power flow and stability solvers, running on a substation appliance against live topology, is deep domain engineering.
Risk
4/5
Sells to many separate public utilities behind their own substation firewalls, with regulator-accepted study packages, though licensed-engineer sign-off rules could blunt the savings.
trends
4/5
Interconnection queues jammed specifically by 2025-2026 AI load growth, the same scarcity behind the Compute at Sea RFS, make per-megawatt study automation newly sellable.

Similar startups in the directory

Companies whose pitch matches most of the idea's terms (clears, utility, interconnection, queues, center, load, sells, public): 3 all-time, 2 from the last two years. Same matching as Idea Check.

Libra Roboticsyc S26 · 2026 · Robotics and physical worldalive
Robotic crews for large-scale infrastructure construction
Clear Currentplugandplay PnP 2025 · 2025 · Climate and energyalive
Our platform helps you optimize efficiency, reduce costs, and make informed decisions—all from one centralized platform
iMUplugandplay · Healthcare and bioalive
Visualizing the total load on the knee joint during walking to recommend optimal exercise levels.

Run this as an Idea Check →

The generator's reference companies

Real companies the model named as closest when it wrote the card, with their fate. A check mark is a company the radar could verify in its directory.

Public money in this direction

US federal grants, SBIR/STTR awards and open opportunities from the radar's public-money feed, matched to the idea's terms; the sector totals give the context.

grants and programs matching the idea

237

startup-relevant grants in AI infra and compute

$337M

awarded in the sector, tracked

opportunities open now in the sector

Transforming Trauma Care: Innovative Solutions for Enhancing Behavioral Health Outcomesawardhigh relevance
NIH / NIMH · STTR phase I · $456K · posted 2026-09-01
I-Corps: Translation Potential of Wearable Biomechanics Assessment for Mobility Evaluationawardhigh relevance
National Science Foundation · I-Corps · $50K · posted 2026-08-20
I-Corps: Translation Potential of Edge-Artificial Intelligence (AI) Personalized Risk Prediction for International Travelersawardhigh relevance
National Science Foundation · I-Corps · $50K · posted 2026-08-19
I-Corps: Translation potential of a monthly oral drug delivery platformawardhigh relevance
National Science Foundation · I-Corps · $50K · posted 2026-08-17
I-Corps: Translation potential of a clinical decision support system for evidence-based reasoning for disease management and preventionawardhigh relevance
National Science Foundation · I-Corps · $50K · posted 2026-08-17
Novel Immunoregulatory Therapeutics for Bacterial Gastritisawardhigh relevance
NIH / NIAID · SBIR phase I · $314K · posted 2026-08-13
SBIR Phase I: Intelligent Wireless Charging and Device Management System for High-Volume Environmentsawardhigh relevance
National Science Foundation · SBIR Phase I · $305K · posted 2026-08-11
Culturally Adapted Digital Behavioral Therapy for Overactive Bladder in Black Women: An Innovative AI-Assisted Approachawardhigh relevance
NIH / NIMHD · SBIR phase I · $328K · posted 2026-08-05

All public money by sector →

Market signal

What the radar sees in AI infra and compute: new companies by cohort year, the forming YC batch, and outcomes since the February snapshot.

AI infra and compute · 11 → 34 → 59 → 59 → 56 new companies 2022 → 2026 · 97% aliveYC F26 live: 4 in this cluster, 11% of the batch (was 8% in S26)Since February, of 117 YC companies here: 3 acquired, 2 shut down, 25 rewrote their pitch

AI infra and compute: companies, trend and grants →

Design attributes

The card is one cell of a designed set: every axis below was chosen before the text was written, and the text had to realize it.

Buyer: Government and public sector
Business model: AI agent as a service
Path to 100x: Collapses a cost 10x
Market size: $1-10B market
Capital intensity: Capital-heavy (hardware, bio, infra)
Speed to revenue: Revenue in 1-3 years
Technical depth: Deep tech: ML, hardware, bio
Go-to-market: Partners and channels
Moat: Distribution moat
Geography: Global from day one
Regulation: Heavily regulated
Vibe: Boring business