Bitcoin's mining layer has three facts that are hard to hold together with its reputation.
Three pools have mined a majority of its blocks in recent measurements. Around forty percent of hashrate has run through one pool and its proxies. More than eighty percent of the chips doing the work have come from a single manufacturer.
None of that is a scandal. It is the natural end state of a single- algorithm proof-of-work network: the reward goes to whoever buys the most efficient hardware, the most efficient hardware comes from one supplier, and the pools that coordinate the largest hashrate win the most blocks. Bitcoin still works. But the claim that "no one controls it" has to be understood as no one controls the protocol rather than no one controls the production.
Epic's answer to this is the most unusual part of its design, and the least discussed.
Three algorithms, one hardware class each Epic produces blocks with three different proof-of-work algorithms, each aimed at a different kind of machine:
RandomX, for general-purpose CPUs — the chips already in every laptop, desktop and workstation.
ProgPoW, for commodity graphics cards — the hardware in hundreds of millions of gaming PCs.
Cuckatoo31+, for specialized hardware — retained at a small share.
The block production is allocated 48 percent, 48 percent and 4 percent respectively.
The purpose of the split is not novelty. It is that a miner cannot dominate Epic by being excellent at one thing, because two- thirds of the blocks cannot be reached with the hardware that wins at the third. An industrial ASIC operation, however efficient, competes for four blocks in a hundred.
The allocation is enforced, not encouraged Here is the detail that separates this from every other "multi- algorithm" coin that quietly became single-algorithm in practice.
At each block height, Epic's consensus policy engine — called Feijoada — fixes which algorithm the next block must use, drawn from the policy's proportions. Miners do not choose. The mix is a property of the chain, not of miner behavior, and it holds whether miners like it or not.
Without that enforcement, the split would collapse. Miners follow profitability, profitability follows hardware prices, and hardware prices follow whatever is cheapest to produce at scale. A voluntary three-algorithm network becomes a one-algorithm network as soon as one algorithm pays better, which is precisely what has happened to other projects that tried this.
Difficulty is tracked per algorithm rather than globally. Each new block retargets the difficulty of its own algorithm from the preceding sixty observations of that algorithm's blocks, so the three lanes do not contaminate each other's adjustment. At the target mix that is roughly a two-hour lookback for RandomX and ProgPoW, against Bitcoin's fixed two-week epoch.
The 48/48/4 split itself was set by a network upgrade in June 2023. The framework is open-ended: algorithms can be added and proportions rebalanced by consensus if the hardware landscape changes. The numbers are an instrument, not a doctrine.
Why 4 percent of ASIC is not a contradiction The obvious objection: if ASIC resistance is the goal, why allow ASICs at all?
Because fighting them has a poor record. Every coin that promised ASIC resistance and depended on it for its security has either forked repeatedly, watched its hashrate collapse when a manufacturer shipped a machine anyway, or ended up trusting a small group of developers to keep changing the algorithm faster than silicon can be designed.
Epic takes the other route: give specialized hardware a small, fixed seat at the table and keep the other two seats for hardware that already exists in billions of homes. The ASIC operator is a participant rather than an existential threat, and the network does not depend on winning an arms race it would eventually lose.
That is a design judgment, not a law of nature. It is at least an answer to the question, rather than a slogan that dissolves the first time someone tapes a hundred mining boards together.
The honest part: latent is not the same as realized The tempting claim here is that Epic is secured by six and a half billion devices — every phone, laptop and gaming PC on earth.
The 2026 technical paper explicitly refuses that argument, and the refusal is worth quoting because it is the kind of honesty that makes the rest of a document credible:
The installed base of CPU and GPU devices worldwide is roughly 6.6 billion, but that base "is not committed defensive hashrate, and this paper does not treat it as such: realized security is only the hashpower that mining rewards actually attract."
That is the correct way to think about it. A billion laptops that could mine are not security. They are potential. What secures the chain today is the hashpower actually pointed at it, and that number is a function of the reward, the electricity price, and whether anyone finds it worth doing.
A reader should hold the design claim and the realized number separately: the design makes it possible for ordinary hardware to participate, which is a structural property. Whether ordinary people actually do is an empirical question, answered only by looking at the network.
The cost of a deterministic schedule There is a vulnerability in this design, and the same technical paper names it.
Because Feijoada decides which algorithm must produce the next block, the chain's liveness depends on miners for that algorithm showing up. If the required algorithm's hashpower disappears — because it became unprofitable, because a jurisdiction banned it, or because someone deliberately withheld it — the chain can stall even if the other two algorithms have plenty of miners.
A probabilistic design would let the remaining miners carry on. A deterministic one cannot, by construction.
The mitigation is that the three algorithms run on different hardware in different places, so a single cause is unlikely to remove all three. But "unlikely" is not "impossible," and a book that pretended otherwise would be doing what this book has criticised everywhere else.
What this buys, stated without inflation Not "unhackable." Not "unstoppable." Not "no one can attack it."
What it buys is narrower and more useful: no single hardware vendor, and no small group of pools, can decide what gets mined on this network. The production of blocks is distributed across hardware classes by consensus rule rather than by market share, and the mix survives miner preferences because miners do not get a vote on it.
For a chain whose entire argument rests on nobody being in charge, that property is not decoration. It is the difference between a claim about governance and a claim about reality.
The next chapter takes the other half of that reality: what it costs to verify the chain yourself, and why four gigabytes matters more than any of this.
Three numbers worth watching If you want to check these claims rather than take them, three figures do most of the work.
Hashrate per algorithm. The policy fixes 48/48/4. Reality is whether miners actually show up in all three lanes. A lane that is thin relative to its allocation is a lane where the chain depends on fewer participants than the design assumes — and it is visible in any block explorer that labels the algorithm.
Node count. Not miners: nodes. Miners produce blocks; nodes decide whether to accept them. A chain that is cheap to validate attracts validators who do not mine, and those are the people who enforce the rules when a miner tries something. Four gigabytes is what makes that number plausible.
Fee revenue. As the block reward declines, fees become the security budget. Watching fee revenue against hashrate tells you whether the transition to a fee-funded model is happening or stalling — the single most important open question about any proof-of-work chain's long-term security, including this one.
None of the three is a price. All three are checkable today, by anyone, without asking permission.