JoxCap: Quantum-Safe Money, ASML's Order Book, and Why Inference Isn't Money

No. 01

Project spotlight: Quantus (QTC), money built for the day after Q-Day

Ticker check first. This is Quantus Network, the post-quantum proof-of-work chain. It isn't Quant (QNT), and it isn't the two other "QTC" tickers on CoinGecko (Qubitcoin and Qitcoin).

What it is

Quantus is a new layer 1 that tries to be Bitcoin with quantum-resistant cryptography. Every transaction is signed with ML-DSA-87 (Dilithium), the NIST post-quantum standard, with no ECDSA fallback. Those signatures are large, so Quantus rolls transactions into zero-knowledge proofs to keep blocks manageable. The same trick gives optional privacy: send coins to a "wormhole" address, re-mint them elsewhere with a proof, and the onchain link is broken. No smart contracts, no staking. Just money.

The monetary policy borrows from Bitcoin: a 21M cap, GPU mining, and a block reward that decays smoothly rather than halving. Mainnet went live on 9 September. Founder and CEO Christopher Smith previously built BitMesh, Lunyr and FactoryDAO. Quantus reportedly raised about $2.4M across two rounds, the second led by Balaji Srinivasan. The whitepaper lists audits from Eiger, Neodyme and Hashcloak.

Trading opened on 6 October through near.com and NEAR Intents, with no major centralised exchange listing, so any wallet plugged into NEAR's 1Click Swap API can offer it. On Thursday morning CoinGecko had QTC at about $112, a $32M market cap and roughly $7.6M of daily volume. That's about 40% below the $186 high it printed on 8 October.

Why now

Last week Justin Drake asked the industry to calmly plan for "bunker mode", moving coins to fresh addresses in case AI-assisted maths or quantum computers break ECDSA sooner than expected. Ledger CTO Charles Guillemet pushed back hard: there's no evidence of a practical break. Whoever's right, the question has left the conference panels. This week Coinbase, which holds about 5% of all bitcoin in custody, published its own priorities for a multi-year Bitcoin post-quantum migration. Quantus is the purest bet on the question, because it has nothing to migrate. In the same week it co-hosted Quantum & Privacy Day at TOKEN2049 with NEAR, and Project Eleven named it a design partner for its post-quantum custody product.

The risks

The premine. 27% of supply (5.67M QTC) was minted at genesis: 23% for investors, founders and team, and 4% for the company. Insider tokens are locked until September 2027, then vest over 36 months. On the whitepaper's own numbers, that's roughly 5,000 QTC a day unlocking against about 2,200 a day mined.

Thin float, big valuation. About 284K QTC are in circulation, roughly 1.4% of the cap. At Thursday's price the full 21M would be worth about $2.35B. Small floats move fast both ways, as the last week showed.

Self-reported numbers. Hashrate, user counts and the team's estimate of the mining cost per coin ($45–80 in late September) all come from Quantus itself.

It's a month old. No track record, liquidity mostly on one route, and competition from QRL, Quip and bigger chains like NEAR adding post-quantum signatures of their own.

Jox's take: The thesis is real and the engineering looks serious: native post-quantum signatures, published audits, open-source code. But it's a one-month-old chain with a 27% premine and an unlock wall that starts next September. You're buying the quantum narrative here, not a hedge against it. If I touched it, I'd size it like an option, not a savings account, and I'd have September 2027 circled.

No. 02

Inference is not money. It's the newest commodity.

There's a line doing the rounds in the Bittensor community: "inference is the new money". It's a good line. Taken seriously, it's also wrong, and the difference matters for how you value TAO.

The argument, fairly

Inference is AI being used: every prompt, every agent call. Training builds the model; inference is the model working. The bull case runs like this. Demand for inference is exploding. It's controlled by a handful of large labs. An open network that supplies it cheaper is valuable. And TAO, capped at 21 million with Bitcoin-style halvings, is the asset you need to access that network. The first three steps are broadly right. The leap is the last one, from "valuable resource" to "money".

Three tests inference fails

Money is what people hold when they don't need to spend it yet. That gives three practical tests.

Does it hold value over time? No. a16z's Guido Appenzeller called it "LLMflation": for a model of equivalent capability, the cost falls about 10x a year. GPT-3-level performance cost $60 per million tokens in late 2021 and about $0.06 by late 2024. Stanford's AI Index found a more than 280-fold drop in the cost of GPT-3.5-level performance between November 2022 and October 2024. Epoch AI puts the decline for a given level of performance at about 47% a quarter, or roughly 13x a year, since 2023. A store of value that loses most of its purchasing power every year isn't one.

Do non-users hold it? No. You buy inference because you need to run a model, and the moment you don't, you don't want it. That's an input, not a reserve asset.

Is it the unit of account? No. Inference is priced in dollars per million tokens, by every provider and every subnet that sells to outside customers. Even Gamma, the usage-credit idea Bittensor's Const floated in September, would denominate its credits in dollars.

What inference actually is

It looks far more like oil, electricity or bandwidth: essential, consumed and getting cheaper. Commodities have a consistent history. Electricity got steadily cheaper through most of the last century, and the value went overwhelmingly to the users, with producers doing well only from protected or lowest-cost positions. Today's energy prices are a reminder that commodities can swing back, but the long-run pattern held. The cleaner warning is fibre. Around $90 billion of fibre was laid in the US in four years, and in 2001 Merrill Lynch estimated only 2.6% of the capacity was in use. Bandwidth prices collapsed, and Global Crossing went bankrupt in January 2002 with about $12.4 billion of debt. The demand forecast was right. The producers still lost.

That isn't a bear case. It's a different valuation method. A monetary asset is valued on scarcity and belief. A commodity producer is valued on demand, cost of supply and how much of the value it keeps.

TAO as a producer

Demand: real, and growing. At the Exploit Summit on 28 September, Bittensor co-founder Const cited SubConnect's Revenue Index, which estimates that 24 subnets earn roughly $28–35 million a year from paying customers. SubConnect raised that to more than $50 million in a 2 October update, a figure investor Mark Jeffrey repeated on 11 October alongside a target of $100 million by end-2026. These are industry estimates, not audited accounts, and the $50 million update came without a subnet-by-subnet breakdown. SubConnect is not a wholly neutral scorekeeper either: it announced an investment in Leadpoet on 8 October. Still, a few years ago there was no outside revenue to point to at all.

Cost of supply: well behind. Since the December 2025 halving the network issues about 3,600 TAO a day. At Thursday morning's price of about $262, that's roughly $344 million a year of new tokens paid to miners, validators and subnet owners (my calculation: 3,600 × 365 × $262). Against $28–35 million of revenue, the network pays out about $10–12 of new tokens for each dollar customers pay. On the self-reported $50 million it's still about 7 to 1. Against a market cap of about $3.0 billion, TAO trades at roughly 85–106x paying revenue on Const's figures, or about 59x on the self-reported number.

Emissions are a bootstrapping subsidy, and every new producer starts subsidised. The largest subnet shows the scale of it: a March analysis by Pine Analytics, reported by Startup Fortune, found that Chutes receives roughly 14.4% of all emissions against an estimated $1.3–2.4 million of outside revenue, a subsidy of roughly 40 to 1. The problem is the cost curve: to close the gap, revenue has to grow faster than the price of what it sells is collapsing.

Value capture: indirect. Under dynamic TAO, each subnet has its own alpha token, and since June 2026 emissions per subnet follow a moving average of alpha prices. When a subnet wins a customer, the subnet, its miners and possibly its alpha token benefit first. TAO benefits second-hand, when people buy alpha with TAO. Gamma, under which alpha would be burned to mint dollar-denominated usage credits, is the right direction, though it is still a concept with no formal specification or launch date. And the protocol can't yet verify that the service was delivered, so burns are hard to tell apart from activity a subnet manufactures for itself.

What would change my mind

Three signposts. Outside businesses or agents pricing things in TAO, with no dollar peg in the middle. A large holder base that never touches the network, the first test of which is whether Grayscale's GTAO trust, still quoted over the counter with its ETF registration pending, gets approved. And TAO holding value while inference prices keep falling, decoupled from revenue. Even then, that would be TAO becoming money, not inference.

The scorecard

Four rows for a spreadsheet, checked quarterly. Revenue versus emissions in dollars, today about 7–12x, with under 3x on verified revenue as the bar that would impress me (note the ratio also improves if the TAO price falls, which isn't progress). Revenue concentration across the top subnets. Named customers confirming on their own channels. And how much subnet revenue flows back to TAO itself.

Jox's take: Inference won't be money. It'll be the oil of the AI economy, and that's a perfectly good story. But commodities teach the same lesson every cycle: the commodity gets cheaper, the users get rich, and producers win only if they're low-cost and keep the value they create. Value TAO like a producer, not like digital gold.

No. 03

Markets: CPI came in hot, and bitcoin slipped to the bottom of its range

Wednesday's US September CPI came in above forecast. Headline rose 0.6% on the month, in line, but 3.7% on the year against 3.6% expected, up from 3.4% in August and the highest annual rate since May. Core, the number that matters for the Fed, rose 0.3% on the month and 2.6% on the year, against forecasts of about 0.2% and 2.4–2.5%. Energy explains the headline. It doesn't explain core.

Bonds moved first. The 10-year Treasury yield spiked above 5.32% after the print and the 2-year touched 4.95%, before a tech sell-off pulled money back into Treasuries: the 10-year closed near 5.19%, slightly lower on the day, and the 2-year near 4.80%. Stocks didn't get the reprieve. The S&P 500 closed down 1.11% at 7,718.94, its worst day since July, and the Nasdaq fell 1.88% to 26,692.87, with chips and the AI trade leading losses. Odds of a quarter-point hike at the 27–28 October FOMC rose from about 22% to about 40%.

Bitcoin went with it. It was trading near $83,600 on Wednesday morning, still inside the week's range of roughly $82,000 to $85,500. By the evening it was near $81,700, below the $82,000 floor. Overnight it dipped to about $80,700, briefly under the $81,000 low from Tuesday's liquidation flush, and on Thursday morning it was back near $81,100.

Jox's take: Headline was always going to look ugly because of oil. The problem is core, and core ran hot. With an October hike back on the table and the 10-year still near 5.2%, risk assets are trading off rates, not narratives. $81,000 is the line for bitcoin now. It wobbled overnight and is only just holding. If it fails on a closing basis, I wouldn't be in a hurry to catch the next leg.

No. 04

ASML: demand already outruns 2028 capacity

ASML reported Q3 net sales of €11.7 billion and net income of €5.0 billion, with a gross margin of 57.5%. Installed-base management came in at €3.3 billion. Q4 sales are guided at €14.2–15.2 billion. Full-year 2026 sales are now expected at €44–45 billion, with a gross margin around 56%. The sharper line was capacity: management said planned supply for 2027 and 2028 will not fully meet customer demand, and it is looking at further capacity increases for 2028.

Jox's take: This is the AI build-out showing up in the bottleneck that matters most. TSMC has now answered from the other side of the trade: record Q3 net profit of NT$763.1 billion, up 69% on the year, on revenue of NT$1.49 trillion, with an operating margin of 58.7% that cleared the top of its own guidance. With high-performance computing at 68% of sales, the wafer side is matching the tool side, and doing it at a margin that says demand is still outrunning supply (CNBC).

No. 05

Banks: JPMorgan misses, Goldman clears the bar

Four big US banks reported on Tuesday. JPMorgan posted net income of $15.3 billion, or $5.57 a share ($5.75 excluding securities losses), a rare miss against consensus near $5.90. Goldman Sachs reported diluted EPS of $17.01 against expectations near $13.02, and revenue of $18.68 billion against about $16.71 billion, led by equities trading. Wells Fargo and Citi also printed. The tape split: trading and markets strength for Goldman, a softer print from the biggest bank just ahead of CPI.

Wednesday's round kept the split going. Bank of America's net income fell to $6.9 billion, or $0.91 a share, on $30.0 billion of revenue, weighed down by a $1.2 billion regulatory penalty charge. Morgan Stanley earned $4.9 billion, or $3.03 a share, on $20.3 billion of revenue, with equities revenue up 41% to $5.8 billion. BlackRock reported adjusted EPS of $15.93 and $205 billion of quarterly net inflows, taking assets under management to $15.5 trillion.

Jox's take: One miss and one blowout doesn't rewrite the cycle. It does set the tone for the rest of earnings season: pipelines still look fine, but the bar for "strong" is high.

No. 06

Coinbase pushes Bitcoin toward post-quantum prep

Coinbase published a technical note on Bitcoin's post-quantum readiness. It holds about 5% of all bitcoin in custody and wants the community to start a multi-year migration before any cryptographically relevant quantum computer exists. Preferences include flexible output designs such as P2MR (BIP-360) and P2TRv2, and staged hybrid paths that keep classical and post-quantum spending options during the transition. Abandoned and unmigrated coins are treated as a separate governance question that shouldn't block the engineering work.

Jox's take: When the largest US exchange starts writing migration prefs, the quantum story stops being conference theatre. That doesn't make Q-Day imminent. It does make Quantus's "nothing to migrate" pitch easier for newcomers to understand.

On the radar · this week

The calendar that matters

Source  ThriveInMarkets week ahead →