The protocol has already written most of it. Physics writes the rest.
TOKEN2049 named itself after a year. Here is that year.
Bitcoin's issuance is fixed in code. Nobody votes on it and nobody can speed it up. So some facts about 2049 are already known.
0.049BTCThe block subsidy in 2049. One sixty-fourth of today's 3.125.
99.9%of all bitcoin mined by the 2048 halving
6halvings between now and then
2049 is a receivables problem
Miners keep a proof-of-work network running, and today almost everything they receive is new coins. That schedule of receipts is written in the code, and it halves every four years. By 2049 what miners receive has to come from fees, or they go under. The industry calls this the security budget. Every holder in this room owns an asset that depends on it.
20243.125
20281.5625
20320.78125
20360.390625
20400.195
20440.098
20480.049
New BTC paid per block at each halving. Bars drawn to scale. Halving years are approximate.
Fees only work at volume
Last week fees were 0.7% of what Bitcoin miners earned. For fees to carry the network, there have to be far more transactions: billions of small ones, not a few large ones.
0.7%Share of miner revenue from fees, week to September 28, 2026 (Hashrate Index). The other 99.3% is the subsidy that halves away.
The volume is coming, and it isn't people
In 2049 most payments are made by software. An AI agent buys an answer, a rendered frame, a slice of data, and pays for each one as it goes. The rails are already here. On July 14, 2026 the Linux Foundation launched the x402 Foundation with 40 members, including AWS, Google, Visa, Mastercard, Stripe and Cloudflare, to make payments as easy for software as exchanging data.
There is one condition. A payment worth a fraction of a cent only makes sense if checking and settling it costs even less.
Today the transistor is the cost driver
Every mining machine is a block of transistors turning electricity into heat. To do more work, you need more transistors and more electricity, so the bill runs faster and faster. And miners are now being outbid for that electricity: AI pays more for the same megawatts.
Then the 2028 halving cuts revenue per hash in half again. The industry's answer, bigger farms on cheaper remote power, is running out of road.
Revenue per PH/s per day, today$39.87
Power cost, best air-cooled machine$21.00
Power cost, crypto computing in light~$2.10
Hashprice from Hashrate Index, September 28, 2026. Best air-cooled ASIC at 12.5 J/TH. Power at $0.07 per kWh. Photonic figure modeled at about 90% less power.
What you do, today and on light
Everyone in this room feels the same cost in a different form. Each of these will be delivered as a service; nobody has to buy a machine.
Hold bitcoinFunds, ETFs, treasuries
TodayWhat the network receives moves from new coins to fees.
On lightA seat in the partner network: machines nobody outside it can buy, doing the work at a tenth of the power.
Mine
TodayThe 2028 halving cuts revenue per hash in half while AI bids for your power.
On lightThe same hashrate on about a tenth of the power, and the rest of your power to sell.
Run an exchange or trading venue
TodayMatching and settlement cost money and time on every trade.
On lightSettlement as a service, faster and cheaper per trade, from compute pods in buildings near your users.
Build a chainProof of stake and layer 2s included
TodayProof generation and verification are your heaviest compute bill.
On lightProving and verification bought by the job, at a fraction of the power.
Build payments or AI agents
TodayA payment of a fraction of a cent costs more to settle than it is worth.
On lightPayments small enough for software to pay by the request.
Build the machines
TodayEach new chip generation costs more to make and gives back less, and the loud, power-hungry farms that run them are losing goodwill with the communities around them.
On lightA photonic stack to build on: switch, gates and memory.
Light closes the circle
All six gains come from one change: True Photonic computes with light instead of transistors, for about a tenth of the power. It runs all the way around the problem.
Cheap workA hash for about a tenth of the power.
Cheap verificationThe same work checks and settles transactions.
Tiny paymentsA fraction of a cent becomes worth taking. This changes things in more ways than can be covered here.
Fee volumeBillions of payments from software, every day.
Miners paidThe fees carry the network after the subsidy is gone.
The fifth step funds the first.
Zero waterNo cooling towers, no chillers.
No fansSilent, so it can sit on a floor of an occupied building.
~1/10 the powerSized to the electrical service a city building already has.
Singapore paused new data centres in 2019 because of the power they draw, and since 2022 has released capacity in rounds, the latest of at least 200 MW with half from green energy. The compute that wins is the compute that fits inside the power a city already has.
In 2049 the block reward rounds to zero. The power bill doesn't. Light is how the fees pay the bill.
Why now
2028The next halving. Less than two years away, and the first real squeeze.
DoneThe Poovey Switch, clocked at 150 to 200 femtoseconds at the Technion Israel Institute of Technology. Optical gates tested. OS version one ready.
ModeledAbout 90% less power per hash and per verification.
NextAn operating 64-bit photonic computer, targeted for December 2026 to February 2027.