What Is Starknet? The Cairo-Powered ZK Rollup Scaling Ethereum, Explained

Almost every Ethereum Layer 2 makes the same trade to scale: copy the EVM as closely as possible so Solidity developers feel at home. Starknet made the opposite bet. It threw out the EVM entirely, built its own language and virtual machine, and used a proof system — STARKs — that most of the industry considered too heavy to run in production. The wager was that if you design the whole stack around provability instead of retrofitting it, you can prove vastly more computation per proof and eventually hit throughput no EVM-equivalent rollup can match. Years in, that bet looks more interesting than ever. This is how Starknet actually works, why the Cairo detour matters, and where it stands in 2026.

Who built it and why

Starknet comes from StarkWare, a company founded in 2018 by a team that includes Eli Ben-Sasson — one of the academic co-inventors of the STARK proof itself. That pedigree is the whole story here. Where most rollup teams are engineers adapting existing cryptography, StarkWare is building around a proof system they helped invent, which is why they were willing to take the harder road of a custom VM.

The company's first product was StarkEx, an application-specific scaling engine that powered exchanges like dYdX and Immutable X — private, permissioned, but battle-tested at real volume. Starknet is the general-purpose, permissionless successor: a public Layer 2 anyone can deploy to. It went live on Ethereum mainnet in late 2021, opened to permissionless contract deployment in 2022, and has been shipping toward decentralization ever since. The motivation was never "cheaper Ethereum" for its own sake — it was to prove that validity rollups could scale general computation, not just payments and trades.

The architecture: a ZK validity rollup

Start with the category, because it's the most misunderstood thing about Starknet. It is a ZK rollup — more precisely a validity rollup — which is the opposite design from optimistic rollups like Arbitrum and Optimism.

Optimistic rollups assume transactions are valid and rely on a fraud-proof window (typically seven days) during which someone can challenge a bad batch. That's why withdrawing from an optimistic rollup back to Ethereum takes a week. Starknet doesn't assume anything. Every batch of transactions is accompanied by a STARK proof — a cryptographic object that mathematically demonstrates the batch was executed correctly — and Ethereum verifies that proof before accepting the new state. If the proof checks out, the state is final. There's no challenge period and no need to trust the sequencer's honesty; you trust the math.

The proof system is where Starknet genuinely differs from other ZK rollups. Most use SNARKs, which are compact but require a trusted setup — a one-time ceremony that, if compromised, could let someone forge proofs — and rest on pairing-based cryptography that a future quantum computer could break. STARKs need no trusted setup at all, and they're built on hash functions considered resistant to quantum attacks. The trade-off is that STARK proofs are larger, which historically made them expensive to verify on Ethereum. StarkWare's bet was that proving efficiency would improve faster than the cost mattered — and by 2026 that's largely borne out.

On data availability, Starknet is a true rollup: it posts transaction data (as compressed state diffs) to Ethereum itself, so anyone can reconstruct the chain's state from L1 data. Since Ethereum's proto-danksharding upgrade added blob space, that data goes into blobs, which is what makes rollup fees cheap.

Proving is done by StarkWare's prover infrastructure, historically branded SHARP (the Shared Prover), which aggregates many programs into shared proofs. In 2025 StarkWare rolled out Stwo, its third-generation prover, replacing the older Stone prover and delivering roughly a 100x improvement in proving throughput by the team's benchmarks — the single biggest lever on how much Starknet can process.

Cairo: why it isn't an EVM chain

Here's the part that trips up people coming from other L2s. You cannot deploy a Solidity contract to Starknet unchanged. Starknet doesn't run EVM bytecode. Its smart contracts are written in Cairo, a language StarkWare created specifically so that program execution can be proven cheaply with a STARK.

This is the crux of the whole design philosophy. An EVM-equivalent ZK rollup has to prove the execution of the EVM, an environment that was never designed to be provable — every quirky opcode has to be modeled inside the proof. Cairo is designed the other way around: the language and its virtual machine exist to make provability efficient. The result is that Starknet can prove a lot more computation per proof, which is the foundation for its throughput ambitions.

The cost is ecosystem friction. Solidity has by far the largest developer community in crypto, and Cairo asks builders to learn something new. StarkWare has narrowed that gap — modern Cairo borrows heavily from Rust's syntax and safety model, and tooling has matured — but it remains a genuine adoption hurdle and the honest reason Starknet's ecosystem is smaller than the EVM rollups'.

Native account abstraction

One thing the custom-VM decision unlocked for free is native account abstraction, and it's Starknet's most underrated user-facing feature. On Ethereum and most L2s, your everyday wallet is an "externally owned account" controlled by a single private key; smart-contract wallets are a bolt-on that needs standards like ERC-4337 to work smoothly.

Starknet inverts this: every account is a smart contract by default. There is no separate class of key-controlled accounts. Because your account is a contract, wallets can build in things that are painful elsewhere — passkey and biometric logins, social recovery if you lose a device, session keys that let a game transact on your behalf for an hour without popping a signature every move, and paying gas fees in tokens other than ETH. This is why Starknet wallets like Braavos and Ready can offer genuinely Web2-feeling UX that mainnet Ethereum still struggles to match.

Fees, speed, and performance

Fees on Starknet are cheap and, since the v0.14 series of upgrades, denominated in STRK. Each block prices three separate resources — L1 gas, L2 gas, and L1 data gas — reflecting the real costs of settling to Ethereum versus executing on L2.

The most consequential recent change is v0.14.3, which went live on mainnet on June 22, 2026. It introduced a dynamic L2 gas base fee that adjusts automatically block-to-block based on how full the previous block was (an EIP-1559-style mechanism for L2 execution), cut the target gas per block by around 30%, and shortened block times. Together those made fees both lower and more predictable during congestion.

On raw throughput, Starknet posted a sustained record of 127 transactions per second over a 24-hour period back in October 2024 (during the Flippy Flop campaign) — a real-world figure, not a lab peak. The longer-term target is far higher: StarkWare is integrating Cairo Native (ahead-of-time compilation of Cairo, contributed largely by LambdaClass) into the sequencer, which is expected to push throughput toward the 10,000+ TPS range the roadmap talks about. Whether it gets there is unproven, but the architecture — proving-first, custom VM — is the reason the ceiling is plausibly that high.

Token and incentives

Starknet has a native token, STRK, which launched in February 2024. It does real work:

• Gas. Since the v0.14 upgrades, transaction fees are paid in STRK.
• Staking. Staking launched in 2025 and grew explosively — the network went from around 110M STRK staked at the start of the year to over 1.1 billion STRK staked, which by 2026 represents more than a fifth of circulating supply. Stakers help secure the network and earn rewards.
• Governance. STRK is used in Starknet's on-chain governance over protocol upgrades.

The 2026 headline is the dual-token consensus. Starknet now lets Bitcoin holders stake BTC to help secure the chain alongside STRK — the network's security is split between the two, with STRK carrying the larger share and BTC a meaningful minority stake. It makes Starknet one of the first rollups to fold Bitcoin directly into its security model, part of a broader "BTCFi" push that also includes bridging Bitcoin liquidity onto Starknet for DeFi.

The ecosystem

Starknet's DeFi is smaller than the top optimistic rollups — total value locked has hovered around $200M across DeFiLlama-tracked protocols in 2026, well below Arbitrum — but it's real and has a few standouts. Ekubo is the flagship AMM, known for highly concentrated, gas-optimized liquidity that makes it unusually capital-efficient. Nostra and Vesu anchor lending and borrowing. On the wallet side, Ready (formerly Argent) and Braavos are the main native smart-account wallets, with Cartridge Controller serving the on-chain gaming crowd.

Beyond DeFi, Starknet has leaned into consumer use cases: Ready has pushed toward being a crypto-native "neobank" with fiat on-ramps and a self-custodial debit card, and Bitcoin/Lightning integrations let some wallets spend in the real world. Gaming and provable-computation apps — a natural fit for a chain built around proofs — round out the ecosystem.

How to get on

The path is straightforward. First, install a Starknet wallet — Ready, Braavos, or Cartridge Controller are the native choices, and their account-abstraction features are a big part of the appeal. Then bridge assets in. StarkGate is the canonical bridge for moving ETH and ERC-20s from Ethereum to Starknet; it has expanded over time to add Bitcoin and Solana routes, so it's no longer Ethereum-only. Centralized exchanges also support direct STRK withdrawals to Starknet, which is often the cheapest way to arrive if you're starting from an exchange balance.

One caution that never gets old: bridges are the single most phished surface in crypto. Reach StarkGate through the official Starknet site or your wallet's built-in bridge, never through a search ad or a link someone DMs you.

Risks and what to watch

A clear-eyed accounting:

• Decentralization is still in progress. For much of its life Starknet ran a single StarkWare-operated sequencer. The v0.14 "Grinta" upgrade laid the groundwork for a multi-sequencer setup — several independent sequencers reaching consensus, each with its own mempool — and staking-based validation is on the 2026 roadmap. Until that's fully permissionless, the operator retains real influence over transaction ordering and liveness. This is the most important thing to track.
• Modest TVL and a thinner ecosystem. A TVL in the ~$200M range means less DeFi depth, wider spreads, and fewer blue-chip protocols than Arbitrum or Base.
• Cairo, not EVM. The custom language is Starknet's technical strength and its adoption weakness. A smaller developer pool means fewer apps and a distinct audit surface — Cairo bugs don't benefit from the years of EVM security tooling.
• Roadmap execution. The 10,000+ TPS ceiling and full decentralization are targets, not shipped facts. Cairo Native integration and the decentralized sequencer/prover layers still have to land.

Bottom line

Starknet is the purest expression of a contrarian idea: that the way to scale Ethereum isn't to imitate the EVM but to design a whole stack — language, VM, proof system — around being provable. That choice cost it the easy on-ramp of Solidity compatibility, and its ecosystem is smaller for it. But it also gave Starknet things EVM rollups can't easily replicate: STARK proofs with no trusted setup and quantum-resistant cryptography, native account abstraction that makes wallets genuinely usable, a throughput ceiling that's plausibly far higher than the competition's, and by 2026 a dual-token security model that pulls Bitcoin into the fold. It's still mid-journey on decentralization and modest on liquidity. But of all the Layer 2s, it's the one making the most technically ambitious bet — and increasingly delivering on it.

For related reading: what is DeFi, best Ethereum L2s, account abstraction explained, and how to bridge to Layer 2.

Related guides

• ZKsync Era
• Scroll
• Linea
• Status Network
• Arbitrum

Not financial advice. STRK is volatile and Starknet's decentralization and performance roadmap can change — always verify details on official Starknet channels.