v2060Jacob Cloete · updated 29 Sep 2026
v2060 · what is worth building by 2060

Where physics leaves room is rarely where the world is stuck.

Sixteen engineering and maths problems, ranked three ways. Five come out on top.

Three ways to rank a problem

Room left under the physics or maths limit, what blocks progress now, and expected value by 2060. The five picks are in blue; robots and learning from little data together make the first.

Limit headroomphysics & mathsCurrent bottleneckwhat binds nowValue if solvedexpected, by 2060benchmark, not a limitopen or no ceilingover $1T a year$0.1–1Tunder $0.1T288Quantum speedupsQuantum speedups31110Carbon removalCarbon removal41213Strong materialsStrong materials7912Battery weightBattery weight81314Weather predictionWeather prediction91516Error-correcting codesError-correcting codes?57Proving software correctProving software correct?109Turbulence simulationTurbulence simulation?1615Matrix multiplicationMatrix multiplication?1411P vs NPP vs NP175Compute per wattCompute per watt513Clean firm power & gridClean firm power & grid666Industrial heatIndustrial heatB32Learning from little dataLearning from little dataB21RobotsRobotsB44Building fasterBuilding fasterengineeringmathematical

B: the gap is measured against a benchmark (humans, or other countries’ costs), not a law of physics. ?: an open question, or no known ceiling. Value is order-of-magnitude, so neighbouring ranks are ties.

The five problems I'm building toward

Each chart shows the number that binds today, with its limit or the best benchmark.

In use today Best lab result Milestone that unlocks value Limit or benchmark Limit of one technology Reference point Around 2010 Every chart: log scale, better is always to the right.
  1. 1

    Robots that learn

    The best robot policy of 2025 finished only 12.4 % of household chores in new, randomised layouts. Unsupervised use needs 99 %+, and hands and rare edge cases are most of the gap.

    Why first: a robot that reliably learns new physical tasks turns AI into labour in factories, care and homes, the biggest prize on the list. Failure rates need to fall about 90×.

    See the roadmap
    A 20-step chore works only if every step doesstep 90 %12 %≈ todaystep 99 %82 %step 99.9 %98 %≈ a personeach step needs ~10× fewer errorslearning a new skill: robot 50–100 demos · person 1–5
    Small error rates compound. That is why the long tail, not the easy steps, decides whether a robot is useful.
  2. 2

    Clean firm power and the grid

    A solar farm takes about seven months to build. In 2025 the median US project waited 61 months to connect, and only 13 % of requested capacity ever gets built. South Africa had a record 335 days of load-shedding in 2023.

    Why second: cheap clean generation already exists, so connection speed sets how fast every other energy fix lands. The fix is mostly process and hardware supply, not new physics.

    See the roadmap
    Building a solar farm vs connecting it (US)0123456yearsbuild≈ 7 monthswait61 monthsin the queue2,060 GW13 % builtthe plant is ready ~4½ yearsbefore the grid is
    The machines are not the slow part. The paperwork, studies and transformers are.
  3. 3

    Compute per watt

    A logic gate still burns about 3×10⁴–3×10⁵ times the Landauer minimum, while data centres head from 485 TWh (2025) to about 950 TWh (2030). Every joule saved here shortens the power queue.

    Why third: the largest proven headroom on the list. It ranks below the first two because AI build-out today waits more on grid connections and memory chips than on efficiency.

    See the roadmap
    Today: charge, then dumpVevery switchthrows away ½CV²as heat≈ 100,000× LandauerFuture: recycle the chargeenergyenergy swings back and forthonly erased bits cost kT ln 2
    The whole roadmap is about not throwing energy away on every switch.
  4. 4

    Building faster

    US construction output per worker was about 40 % lower in 2020 than in 1970. New York’s next subway costs about $2.5 billion per km, ten times the world average.

    Why fourth: construction cost multiplies the price of grids, chip fabs and housing. Much of the gap is permits and procurement, so engineering can close only part of it.

    See the roadmap
    Cost of 1 km of subway, drawn to scaleNew York$2,500 MIstanbul$126 MMadrid*$50 Msame rock, same machines: the 20–50× gap ismostly design, permits and procurement*1995–99
    Engineering can close part of this gap: repeat designs, factories and robots. The rest is how projects are run.
  5. 5

    Clean industrial heat

    Industrial heat is almost a fifth of world energy, and steel alone emits 7–8 % of greenhouse gases; the world average is 1.92 t CO₂ per tonne. Hydrogen-made steel works in pilots but costs about $225/t more in China at $5/kg hydrogen.

    Why fifth: the chemistry already works, so what is missing is cheap clean electricity and hydrogen. Proving software correct is a near tie for this place.

    See the roadmap
    Making iron means pulling oxygen off the oreBlast furnace: coal takes the oxygenFe₂O₃+3 CO2 Fe+3 CO₂≈ 2.3 t CO₂ per tHydrogen DRI: hydrogen takes itFe₂O₃+3 H₂2 Fe+3 H₂OHYBRIT ≈ 0 t fossilore minimum 8.4 GJ/t · all-steel avg 21
    Swap the molecule that carries the oxygen away and the carbon dioxide becomes water.