Unlock the option
Architectures, corpora, model types and infrastructure sit behind research. Cash helps. It does not erase prerequisite chains or calendar time.
An AI company tycoon / in development
Scaling Laws
January 2022. Twelve million dollars. No product. Train models, fund research and commit real money to compute while the frontier moves whether you are ready or not. The expensive decision is rarely just what to buy. It is when.
Current build
Unity / current development build
HCK Labs
Campaign begins
01 Jan 2022
Starting capital
$12M
Released products
0
The frontier does not wait for your roadmap. Neither does the hardware market.
Starting small
It starts in a house, not a glass tower.
The first company view is deliberately modest: a desk, a bed, a living room, a garage and a cheap car. Later company growth is meant to become visible as well as numerical, with larger offices, staff and owned infrastructure replacing the starter setup.
Visual progression is still in development. The simulation already exists underneath it; character movement, staff population and later building tiers are part of the next physical-world pass.
The loop
Research. Train. Release. Survive the bill.
Scaling Laws is built around decisions that consume both money and calendar. A better number is useful only if the company can afford the time it took to get there.
Commit compute
Choose the model, scale, data and compute budget. The creator shows projected capability, training time and the cash the run is expected to burn.
Choose the day
A finished run can sit on the shelf. Waiting may dodge a rival launch or buy time for an upgrade, but the market keeps moving while you wait.
Make it pay
Set pricing, manage free access, build brand and fund the next cycle. A company that ships one model and coasts is designed to fall behind.
Systems in the current build
The numbers are connected on purpose.
The UI is still moving. The underlying systems already share one simulation: the same capability, market, research and company state drive the screens instead of separate decorative scores.
SYSTEM 01
Playable
Build the model one decision at a time
The staged creator separates foundation, scale, data, compute and review so each choice has room to explain what it changed. Capability, training time and cash burn reprice as the blueprint changes.
SYSTEM 02
Playable
Research costs calendar
The technology tree gates what the company can use. Money can fund a programme, but it cannot buy back months already lost.
SYSTEM 03
Playable
Design a house architecture
Spread a research programme across sparsity, throughput, quality, serving cost and reasoning. A cheap rushed programme carries more outcome variance, and chasing every direction at once dilutes depth.
SYSTEM 04
Playable
Capability is not a business model
Price, free access, brand and model age all affect demand. Several models can be live at once, so a specialist or lower-priced product can coexist with the flagship.
SYSTEM 05
Simulation
Rivals can wait on purpose
Competitors are agents rather than a static leaderboard. Some delay a release when better silicon is close, some rush when the player gets too far ahead, and the field keeps improving between launches.
WAIT / RUSH / DRIFT
ONE SHARED FRONTIER
ONE SHARED FRONTIER
SYSTEM 06
UI next
Compute is a commitment
Renting is flexible and expensive. Owning can be cheaper per FLOP, but the hardware ages, sits on the balance sheet and depends on CPUs, memory and fabric feeding it properly.
RENTED CLOUD → COLOCATION → OWN DATACENTER
SYSTEM 07
Simulation
Paid intelligence can still be wrong
Intel can warn about hardware launches, price collapses, supply squeezes or a rival holding back. The cheaper desk sounds more certain than its real accuracy deserves.
INFORMATION HAS A PRICE
CERTAINTY DOES NOT
CERTAINTY DOES NOT
Under the interface
Real rules, coarse enough to play.
Scaling Laws uses published scaling-law behaviour and public hardware data as reference points, then turns them into a game economy. Estimates and future projections are meant to stay labelled as estimates.
~20×
Tokens per parameter
Compute-optimal training sits around twenty tokens per parameter in the model used by the game. Undertraining or overtraining makes the same compute budget buy a worse result.
10× → +10
Compute treadmill
Ten times the effective compute buys about ten capability points on the game's relative scale. The frontier therefore keeps moving even when the player's last model felt expensive.
2 ways
Hardware loses value
Owned hardware depreciates with time and again when meaningful successor generations arrive. Buying too early can leave capital trapped in a fleet the market has already passed.
Engineering notes and source are public in the Scaling Laws repository.
Current build
The foundation is playable. The world around it is catching up.
The simulation, first creators and major management screens exist today. Visual progression and infrastructure interaction are the active gap between the current build and the fuller company-tycoon loop.
Model creation
Staged blueprint creator with live projections for capability, training time and cost.
Research
Technology tree, architecture programmes and unlock gates tied to both money and calendar.
Company setup
Founder skills, traits, company identity, region and country trade-offs.
Business layer
Pricing, audience, market share, marketing, funding, debt, ranking and intelligence foundations.
Physical site
The starter home exists as a 3D orthographic scene and is being turned into a living company view.
Current screens
Still early. Already recognisable.
These are development screenshots, not store mockups. Layout, spacing and art will continue to move as systems leave prototype state.
Built in public
The game is not waiting for a reveal trailer to exist.
Scaling Laws is being built openly by Marcin "HCK" Firmuga under HCK Labs. The source, engineering notes and development history stay visible while the game changes.