The atlas · start or join
Start it, or join it.
Every quest on this board has quietly assumed you would found the company. That is only ever half the answer. The deciding variable is already in the data: crowding. A quest with no funded competitor is one to start. A quest with eight is a company to join, and joining is not the consolation prize, it is the correct move on the biggest problems.
The routing rule
N = 0–1 · open
Start it
Nobody funded is racing you. The starter pack has the name ideas, the riskiest assumption, and the first artifact.
N = 2–4 · contested
Check who the competition is
Four nonprofits is not four companies. If every competitor is academic, philanthropic, or governmental, the commercial lane is still empty.
N = 5+ · crowded
Join one of them
The thesis is proven and capitalised. Becoming the ninth entrant is worse than owning equity in one of the eight.
Longevity & aging
in-limit ceiling $3.0T
Start · found it yourself
Aging as an FDA-recognized indication
N=5Too crowded to found. See the join column.
The healthspan diagnostic
N=9Too crowded to found. See the join column.
Join · get in at one of them
Both longevity quests are crowded on this site's own sourced data, aging-as-an-indication at N=5 and healthspan-diagnostic at N=9, and the problem carries the largest in-limit cap on the board at $3T. That combination is the definition of a join: the prize is the biggest available and the field is already dense with funded, technically credible teams.
Private, consumer longitudinal diagnostics at scale
Furthest along on the actual bottleneck of the healthspan-diagnostic quest, which is not the assay but consumer distribution and repeat testing. Whoever wins that owns the measurement layer the rest of the field needs.
software way in · The product is a data and interpretation layer over commodity lab panels. Longitudinal biomarker interpretation is a modelling problem, not a bench problem.
Private, clinical-stage epigenetic reprogramming
One of the few companies actually pushing toward the regulatory frontier the aging-as-an-indication quest describes, rather than selling diagnostics around it. The regulatory unlock is the real prize on this problem and very few teams are attempting it.
software way in · Narrower than the others here, and worth saying plainly: this is a biology-led company. The honest software entry is clinical-data and biomarker-endpoint infrastructure, not a general engineering role.
Build this to get a reply
A reproducible analysis of a public aging-biomarker dataset testing whether a published clock actually tracks an intervention, with the negative result reported honestly if that is what the data shows. The field is full of clocks with weak longitudinal validation, and demonstrating you will publish a null result is itself a differentiator.
How to approach
Longevity has an unusually open public-intellectual layer (Impetus Grants, the Biomarkers of Aging Consortium, active researchers posting publicly). A rigorous public analysis reaches the people who hire far faster than an application does.
Energy abundance
in-limit ceiling $2.0T
Start · found it yourself
The fission permitting unlock
N=4Starter pack ready. Riskiest assumption: That the NRC bottleneck is genuinely a paperwork/process problem software can compress, not a political-risk-aversion problem no documentation quality fixes. If reviewers move slowly because saying yes carries career risk regardless of the safety case, better software does not move the timeline.
Geothermal everywhere via oilfield tooling
N=8Too crowded to found. See the join column.
Hundred-hour grid storage
N=5Too crowded to found. See the join column.
Underwriting the interconnection queue
N=4Starter pack ready. Riskiest assumption: That a stuck interconnection position can actually be de-risked with engineering and financial structuring fast enough to resell at a profit before the underlying network-upgrade cost estimate changes again. Grid operators re-run studies and costs move; if a position's economics can shift underneath the deal, this is speculation with worse liquidity than the assets it resembles, not underwriting.
Join · get in at one of them
Three of the four energy-abundance quests are crowded or contested by real venture-backed companies (geothermal N=8, storage N=5, interconnection N=4), and the highest-ceiling play on this problem, fusion, has no quest at all because founding a fusion company solo is not a real option. Energy abundance carries a $2T in-limit cap anchored to Saudi Aramco. If any problem on this board produces the next Aramco-scale company, it is this one, and the way in is employment, not incorporation.
$15.5B, Series G, Jun 4 2026
The only fusion company with a signed commercial power-purchase agreement with a hyperscaler (Microsoft), which converts fusion from a science project into a delivery obligation with a date attached. Highest ceiling on the board: a working reactor makes the operator an energy supermajor.
software way in · Pulsed-power fusion is a control and diagnostics problem as much as a plasma problem. Magnet timing, real-time pulse control, shot-to-shot data pipelines, and simulation tooling are all software. The most plausible non-physicist entry on this list.
~$4B total capital raised (one outlet reports $6.85B, unresolved), Jul 30 2026
The most-capitalised fusion company in the world, with the clearest physics milestone path (SPARC net energy, then ARC). Same trillion-shaped ceiling as Helion via a different technical bet, which is the argument for treating fusion as a two-horse portfolio rather than picking one.
software way in · Tokamak operation is a real-time control problem with a heavy simulation and ML component; plasma disruption prediction is an active published ML field. Magnet manufacturing at scale also needs ordinary industrial data infrastructure.
$10B, IPO May 14 2026, now public (FRVO)
The lowest-technical-risk entry here and the only one already selling firm clean power under real PPAs to data centers. Not trillion-shaped, since geothermal caps out at utility economics, but a revenue-generating company today rather than a bet on physics.
software way in · Fervo's actual edge is treating geothermal wells like oilfield assets: subsurface modelling, drilling optimisation, and fleet-level dispatch against power prices are all software problems it demonstrably runs on.
$7.9B market cap, public (OKLO), Aug 19 2026
Publicly traded, so the join decision comes with fully disclosed financials, a genuinely different risk profile from the private fusion bets. Regulatory-gated rather than physics-gated, which makes the timeline risk legible in a way fusion is not.
software way in · The licensing bottleneck this company lives inside is the same one the fission-permitting-unlock quest targets. Regulatory-data and safety-case tooling is software, and building it inside an applicant is the fastest way to learn whether the standalone version is a real business.
~$3B estimated, Series G Aug 12 2026 (valuation undisclosed)
The leader of the hundred-hour-storage quest (N=5 crowded), included with an honest flag: iron-air storage is component economics selling into utility procurement. A good company, structurally not a $1T one.
software way in · Storage dispatch optimisation and cell-level manufacturing analytics. Real software work at a company whose ceiling is real but bounded.
Build this to get a reply
A public, working simulation or control artifact against real published data: a plasma-disruption predictor on a public tokamak dataset, or a battery-dispatch optimiser on real ERCOT price series. This repo's own kin project megawatt.fun is already the second of those. A working artifact against real data outperforms a resume at every company here, because all five are engineering-led and none can evaluate a generalist claim.
How to approach
Skip the careers page as a first move. Find the specific engineer who published the paper or gave the talk on the subsystem your artifact touches, and send it to them with one honest question about a real limitation you hit. This is a technical-credibility industry with almost no cold-inbound pipeline, which cuts both ways: no warm introduction available, but almost no competition on a genuinely good artifact.
Scientific productivity
in-limit ceiling $1.0T
Start · found it yourself
The autonomous lab
N=9Too crowded to found. See the join column.
The replication layer
N=4Startable, no starter pack written yet.
Fast grants as a standing product
N=3Starter pack ready. Riskiest assumption: Not whether researchers want fast money — they obviously do — but whether funders will delegate allocation. The customer is the philanthropist, and the product is trustworthy judgment, not software.
Join · get in at one of them
A split verdict, and the only problem on the board where start and join are both live. autonomous-lab is N=9 crowded and should be joined, not founded. But replication-layer (N=4) and fast-grants-as-a-product (N=3) are contested only by nonprofits, academic consortia, and government programs, with no venture-backed company racing for either. Those two remain genuine starts; see starter-packs.ts for the founding side.
Public (DNA), the largest operating autonomous-lab platform
The most built-out version of the autonomous-lab quest that exists. Joining is how you learn whether the automated-science thesis holds at scale, using someone else's capital to run the experiment.
software way in · A cloud lab is structurally a software company with wet-lab peripherals: scheduling, protocol compilation, and experiment-data infrastructure are the core product surface.
Private, remote-operated cloud laboratory
The purest expression of the thesis, a lab operated entirely through an API, and therefore the fastest place to find out whether researchers actually change behaviour when the friction is removed. That is the assumption the whole quest rests on.
software way in · The entire customer-facing product is a programming interface to physical experiments. Almost all of the work is software.
Build this to get a reply
Take one published protocol from a real paper, write the machine-executable version of it, and document precisely where the paper is too ambiguous to execute. That gap list is the actual product insight behind both the autonomous-lab and replication-layer quests, and almost nobody outside these companies has written it down.
How to approach
This field publishes heavily and argues in public. The gap list above is a blog post, and the post is the outreach. Metascience and lab-automation people respond to a specific correct technical observation far more reliably than to an application.
Climate change
in-limit ceiling $1.0T
Start · found it yourself
Making natural hydrogen actually flow
N=6Too crowded to found. See the join column.
Methane capture for the open dump, not the landfill
N=1Starter pack ready. Riskiest assumption: That a modular, low-capex retrofit can generate carbon credits a buyer will actually pay for at a price that clears the cost of installing and operating gas capture on a dump nobody engineered to be captured from, before a full landfill rebuild — and that a methodology exists (or can be approved) to verify avoided methane on an open, unmanaged site rather than only an engineered one.
Join · get in at one of them
A genuine split. geologic-hydrogen-flow-engineering is N=6 crowded and should be joined. dumpsite-methane-capture is N=1 and remains the single most open quest on the entire board; see starter-packs.ts, not this file, for that one.
Private, Breakthrough Energy Ventures and Amazon Climate Pledge Fund backed
The best-capitalised company in geologic hydrogen, the newest genuinely large energy category to appear in decades. Early enough that joining still carries meaningful equity, funded enough that it will survive long enough to find out whether the resource is commercial.
software way in · Finding natural hydrogen is a subsurface-inference problem: geological data modelling and exploration targeting, the same skill set Fervo applies on the geothermal side.
Build this to get a reply
A subsurface-prospectivity model built on public geological survey data, published with its own uncertainty stated honestly. Exploration companies hire on demonstrated inference judgement, and public geological data is genuinely available.
How to approach
Geologic hydrogen is small enough that the serious people all know each other and read the same handful of papers. A credible public model is a direct route to a conversation with any of them.
Financial infrastructure
in-limit ceiling $1.0T
Start · found it yourself
Join · get in at one of them
No join path written for this problem yet.
note: every valuation above is real, dated, and sourced. The “software way in” names a role category each company demonstrably needs given what it publicly builds, not a claim that a specific opening exists today. Check live roles at each company’s own careers page. Crowding counts come from src/data/quest-crowding.ts and are re-sourced periodically; the datacenter-power-smoothing count was corrected from N=0 to N=5 on 2026-08-22 after an adversarial re-search.