Proof of Work vs Proof of Stake: what each one actually protects

Before 2009 nobody knew how to make thousands of mutually unknown computers agree on the order of a set of ledger entries without somebody in charge. The problem is not technical in the sense of being complicated. It is that if copying a message is free, then creating fake identities is free too, and with unlimited fake identities you win any vote.

Proof of Work and Proof of Stake are two answers to the same problem, and at bottom they do the same thing: they turn the right to propose a block into something that costs real money. If voting has a price, forging the vote stops being free. The difference lies in how that cost is paid and what happens to it afterwards.

This guide compares the two mechanisms using the real Bitcoin and Ethereum figures as of 28 August 2026, and finishes by working out what it would cost an attacker to try to rewrite a single block on each network.

The problem they solve: double spending

A digital file copies for nothing. If money is a file, nothing stops you sending the same pound to two people at once. The traditional solution is a trusted third party: your bank keeps the ledger and decides which transaction arrived first.

A network without a central authority needs something else: a procedure that makes every participant converge on the same history, and makes redoing that history prohibitively expensive. That is the job, and everything else is implementation detail.

Both mechanisms share the same underlying rule: the valid chain is the one that has accumulated the most cost. In PoW that cost is computational work; in PoS it is committed capital. Changing the past means redoing all of that cost faster than the rest of the network can extend it, and that is the barrier.

How Proof of Work works, step by step

1. Miners gather transactions from the mempool into a candidate block.
2. They hash the block header. A hash function returns a fixed-length value that changes completely with any minimal change to the input, and cannot be run backwards: the only way to find a hash meeting a condition is to try.
3. They change a number — the nonce — and repeat. Trillions of times per second.
4. They are looking for a hash below a target. That target is the difficulty. The lower the target, the more attempts are needed on average.
5. Whoever finds one publishes the block and collects the reward: currently 3.125 BTC per block after the halving of 20 April 2024, plus the fees on the transactions included.
6. Everyone else verifies it instantly. Finding the hash is enormously expensive; checking that it is valid is immediate. That asymmetry is the whole invention.

The difficulty retarget. Every 2,016 blocks — about two weeks if all goes to plan — the network compares how long production actually took against the ten-minute-per-block target and adjusts difficulty up or down. If new hashrate arrives, blocks come faster and difficulty rises until the pace returns to ten minutes. If machines switch off, it falls. It is a thermostat, and it runs without anyone administering it.

As of 28 August 2026, with the chain at block 964,393, hashrate is around 893 EH/s and difficulty around 125.81 T according to mempool.space. Translated: 893 quintillion hash attempts per second, in aggregate, to produce one block every ten minutes. Everything that misses is discarded. The operational detail of that industry is in what crypto mining is.

How Proof of Stake works on Ethereum

In PoS you do not compete by computing. You compete by having deposited.

1. You deposit 32 ETH into the deposit contract and activate a validator. Pectra introduced balance consolidation up to 2,048 ETH per validator for large operators, but 32 ETH is still the activation minimum.
2. You wait in the entry queue. The protocol limits how many validators activate per epoch through the churn rate, fixed at 256 per epoch. As of August 2026 there are 2,144,850 ETH waiting in the entry queue: roughly 37 days and 6 hours.
3. Once active, the protocol assigns you work. Time is divided into 12-second slots and epochs of 32 slots. In each slot one validator is chosen as proposer and a much larger committee attests to what it sees.
4. You are paid for attesting correctly and on time. The yield as of August 2026 is 2.67% APR, with 902,935 active validators and 42.4M ETH staked, 34.79% of total supply.
5. If you go offline, you bleed slowly. Inactivity penalties are small and symmetrical to rewards: the network does not punish you for switching the machine off, it stops paying you and charges you for the gap.
6. If you attack, your deposit is destroyed. That is slashing, and it is the essential difference from PoW.

Slashing: the mechanism with no PoW equivalent

A Bitcoin miner who behaves badly simply wastes electricity: the network ignores the block and the miner keeps the machines intact, ready to try again tomorrow.

A validator who signs two different blocks for the same slot, or issues contradictory attestations, is cryptographically identifiable by anyone. The protocol ejects them and confiscates part of the deposit. And the penalty is proportional to how many validators do the same thing at the same time: an isolated fault is cheap, but a coordinated attack in which a third of the network signs falsely destroys a devastating fraction of the capital involved. The design makes attacking alone an expensive accident and attacking in concert a route to ruin.

Getting out. According to validatorqueue.com, the exit queue as of August 2026 is 2,685 ETH: one hour and seven minutes, plus a sweep delay of about 7.8 days before funds land. In other words, entering currently takes more than a month and leaving is nearly immediate. That relationship inverts during market stress, when everyone wants out at the same time. I develop this in what staking is.

Comparison table

Worked example: what it costs to try to rewrite a block

On Bitcoin. To reverse a transaction you have to produce an alternative chain longer than the existing one, starting from the block you want to change. If you control less than 50% of hashrate, the probability of succeeding falls exponentially with every confirmation. So the starting point is assembling more than 893 EH/s: replicating the planet's entire installed base and powering it.

There is also a cost people routinely forget: opportunity cost. Every block you devote to your secret chain is a block whose reward you are not collecting on the honest chain. At 3.125 BTC per block, with BTC around $79,560 as of August 2026, that is roughly $248,625 per block in foregone income. To undo a transaction with six confirmations you need to produce seven blocks: about $1.74M in surrendered subsidy alone, on top of the energy and on top of having acquired the hardware first. These are approximate calculations derived from mempool.space figures, not an estimate of the total cost of an attack.

On the same figures the network issues roughly 450 BTC of subsidy per day (144 blocks × 3.125), somewhere near $35.8M a day. That is the security budget an attacker has to outspend on a sustained basis.

On Ethereum. You do not need to out-compute anyone; you need to control enough stake. Preventing finality requires a third of deposited ETH; controlling the chain requires more than half. With 42.4M ETH staked and ETH around $2,505 as of August 2026, the deposited total is worth roughly $106bn. A third is about $35.4bn, and more than half about $53bn.

The critical difference is not the number, it is what happens afterwards. The Bitcoin attacker keeps the machines and can resell them or go back to mining honestly. The Ethereum attacker watches the protocol destroy the capital they put up — and faces a community capable of coordinating a fork that excludes them. One rents the attack; the other buys it outright with no resale value.

The centralisation criticism, on both sides

In PoS the serious criticism is not about game theory, it is about concentration. If you need 32 ETH and a server running around the clock, most people delegate. And delegation accumulates: Lido, a liquid staking protocol, is the largest DeFi protocol in the world with $24.20B of TVL as of August 2026. SSV Network adds $12.57B and Binance staked ETH $9.34B. Taking the value of all deposited ETH as the reference, around $106bn, Lido represents roughly a fifth of the total. That is an approximation, because a protocol's TVL and the value of stake are not measured identically, but the order of magnitude is what matters.

The risk is not that Lido "attacks" Ethereum. It is that a small set of operators concentrates censorship power, and that a software bug or a court order affects an enormous fraction of the network at once.

PoW has its own version of the same problem and it is no smaller: mining concentrates into pools that aggregate hashrate from thousands of participants, and it is the pool, not the individual miner, that chooses which transactions go into the block. Add to that geographic concentration wherever energy is cheapest. Nobody has solved this on either side; anyone telling you otherwise is selling something.

The energy debate is framed badly by both camps

What the critics get wrong. Calculating Bitcoin's consumption "per transaction" means nothing. The energy is not consumed by the transaction: it is consumed by the competition for the block, and that competition spends the same on an empty block as on a full one. Consumption depends on miner revenue, not on usage. If transactions doubled tomorrow, consumption would not move.

What the defenders get wrong. "The spend is the price of security" is true, and it does not answer the question of whether that level of security warrants that level of spend. Nor is it resolved by pointing out that some of the energy is renewable or otherwise stranded: that improves the balance, it does not remove it.

On The Merge, the verifiable fact is that Ethereum eliminated mining on 15 September 2022 and its consumption became that of running a few hundred thousand validator nodes on domestic or ordinary server hardware. The reduction was several orders of magnitude. I am not giving a precise percentage because estimates of PoW-era consumption vary widely by methodology, and a gap is preferable to a figure I cannot verify.

What The Merge changed, and what it did not

• It changed the consensus mechanism: from miners to validators, without halting the chain for a single block.
• It changed issuance: the mining reward disappeared, and it was far larger than current staking rewards.
• It changed the participation profile: from an industrial activity intensive in physical capital to one intensive in financial capital.
• It was completed in April 2023 with Shapella, which enabled withdrawals of staked ETH. Until then you could deposit but not leave.

• It did not make gas cheaper. Transaction cost is set by the EIP-1559 fee market, not by consensus. What collapsed L2 costs was Dencun, on 13 March 2024.
• It did not raise base-layer transactions per second.
• It did not make Ethereum deflationary by itself: that depends on the burn exceeding issuance, which is a function of network usage.

Where the debate stands in 2026

For Bitcoin, the open question is what will sustain the security budget as the block reward keeps falling. As of 28 August 2026 fees sit at 1–2 sat/vB with a very quiet mempool, so they contribute a marginal fraction of miner revenue. With the April 2028 halving cutting the subsidy to 1.5625 BTC, that arithmetic has not improved.

For Ethereum, the live front is stake distribution and censorship resistance, which is exactly what the coming upgrades target: Glamsterdam, with ePBS, is aimed at 4 November 2026, and Hegotá, with FOCIL, at 19 May 2027.

On the regulatory side, the joint interpretation issued by the SEC and the CFTC on 17 March 2026 expressly clarified the treatment of both protocol mining and protocol staking, a point that had gone years without a clear answer in the United States. Broader market-structure legislation is still pending: the CLARITY Act passed the House on 17 July 2025 and cleared the Senate Banking Committee on 14 May 2026, with a cloture vote expected on 15 September 2026. In the European Union, MiCA regulates service providers and stablecoin issuers rather than consensus mechanisms themselves, so neither mining nor staking is licensed as such — what is licensed is the business of offering staking services to clients.

How mining or staking rewards are taxed is a separate question again, and it varies substantially between jurisdictions: when income is recognised, at what value, and how a later disposal is treated are all national matters. Check the rules where you are resident and take professional advice rather than assuming a treatment you read about elsewhere. If you want the full contrast between the two networks, it is in Bitcoin vs Ethereum.

Frequently asked questions

Sources and references

1. mempool.space — Bitcoin hashrate and difficulty
2. Ethereum.org — The Merge
3. Ethereum.org — Proof of Stake
4. validatorqueue.com — validator entry and exit queues
5. Bitcoin: A Peer-to-Peer Electronic Cash System (whitepaper)
6. DefiLlama — DeFi protocol TVL
7. CFTC — Press release 9198-26 on the joint crypto-asset taxonomy
8. SEC — Joint clarification on crypto assets (17 March 2026)

Related guides