UN agency and DFINITY launch soverign AI pilots for governments axios.com/2026/08/25/un-sove…
#InternetComputer $ICP: What Comes After SaaS?
Free, Business-Owned Software
Software is free because the infra it runs on is the product.
Businesses deploy their own Cloud Engine subnets with preinstalled open SaaS apps: CRM, HR, Accounting, Docs, Slack, Drive ... free forever/no per-seat licensing.
Every preinstalled app is remixable. It belongs to the business, can be modified at will, and customized using built-in AI.
AWS cannot easily copy this model. A free CRM would cannibalize Salesforce, while SaaS companies remain among the biggest drivers of hyperscaler demand.
@dominic_w how far along is Snorkel to help with migration from Salesforce, QuickBooks, HubSpot and other top SaaS apps? (is that even planned?)
It would dramatically lower the friction to adoption.
Any business could test it in practice before committing, no expensive migration team, no six-month implementation...
Deep liquidity, not features, will decide whether MultiDEX.ai changes the game for $ICP.
I'm bullish on the product. It'll mature, unlock new functionality, and improve with every iteration.🔥
My question is whether the liquidity will match the ambition. How much liquidity will @dfinity and/or its partners actually commit?🤷♂️
Launching in Play Mode is the right call. Find the bugs and polish the UX before real money is on the line.
But the real test comes after. If a $10M $ICP trade still gets crushed by slippage, we're still missing a huge part of the market.
Today, large ICP trades are forced onto KYC exchanges. That shuts out an entire class of whales, funds, and DeFi-native traders.
@dominic_w one feature I'd love to see is private OTC trades. Let users create a trade that only the intended counterparty can see and execute.
Large blocks could move without touching the public order book, no market impact and no unnecessary slippage. Ship that, and onchain becomes the default venue for large block trading.
👉Head over to multidex.ai, grab your $100K in play money, and see what it can do.
MULTI/DEX is live on multidex.ai in Play Mode. The community eval process has begun...
Trade to win prizes!! Send bugs and exploits to multidex@dfinity.org for bounties!! GitHub link to follow.
100% onchain. Owned by the world. DeFi 3.0. NNS autonomy for safety 🔥📈
Some complain about $ICP terminology: "why canisters, why neurons, why nervous system... just use normal crypto terms"
Here's why the normal terms don't fit;
⚠️"Smart contract" undersells what a canister is
A smart contract is a script that mutates balances.
A canister bundles code + state + its own gas balance, persists memory orthogonally (no database needed), serves web directly to browsers & can hold its own frontend. That's not a contract, that's a full compute unit.
⚠️"DAO" undersells what the NNS is
A DAO votes on treasury spending & parameter tweaks, then a multisig executes.
The NNS upgrades the entire protocol, admits & ejects node providers, creates subnets & recovers broken ones, fully onchain, no multisig, no foundation veto. It's not a DAO bolted onto a chain; it's the system that RUNS the chain. Hence: nervous system
⚠️"Staking" undersells what neurons are
Staking = lock tokens, earn yield.
Neurons carry dissolve delay, age bonus & vote following, your commitment literally shapes your governance power. It's a governance instrument, not a savings account.
✅Crypto terminology was invented for token settlement blockchains. Move tokens, secure ledger, done.
#InternetComputer is a general-purpose compute blockchain, forcing 2015 vocabulary onto it would be like describing a smartphone as a "portable telegraph".
New things deserve new names. The confusion isn't the terminology, it's expecting a blockchain cloud to look like a payment rail.
Learn 4 new words, understand a whole new design space, too big an ask for some, apparently.🤷♂️
p.s. The flat-earther mindset is mistaking your own unfamiliarity with a subject for everyone else being deluded.😉
Haha $ICP got rrrrrrrekt. I especially agree with the point that it obfuscates things by using goofy terminology. This leads the community to think that only they have discovered some secret truth. Typical flat earther mindset
Your crypto wallet shouldn't be a browser extension waiting to be exploited or a mobile app controlled by an app store.
⛓️ 100% onchain with @oisy
🔐 Passkey login, no seed phrase to lose
🧩 Keys are never stored in one place
🕶️ Privacy mode
🌐 Multichain: BTC, ETH, SOL, ICP, BNB Chain, Base, Polygon & Arbitrum...
💸 Borrow & lend crypto
🖼️ NFTs
🤖 Built-in AI assistant
🛒 Even works for payments at SPAR supermarkets in Switzerland 🇨🇭
Try it: oisy.com
Less then sixty seconds to set up. Peace of mind every time you log in.
OISY 2.5 is live with three new features:
1) OISY Earn welcomes Liquidium
Borrow and lend directly from your OISY Wallet. True DeFi lending without centralized bridges.
2) OISY Notes
Store your private notes with the same level of confidence as your digital assets. Encrypted and only readable by you (thanks to vetKeys). Share notes with links that expire after a single view.
3) OISY Trade limit orders
Buy and sell your digital assets at prices you decide. Set your target price and let the order fill itself when it the price target hits. Live for ICP and ckUSDT, with more pairs coming.
In addition, we implemented several design improvements
#InternetComputer $ICP If you're curious how it works, this post is well worth the read.🧠= good
Right now the Internet Computer dashboard shows the network finalizing blocks at 201 blocks per second.
That number sits right there in the live metrics, not some marketing slide from last year. It is what the protocol is actually doing at this moment across the whole system.Most blockchains run one chain.
Every validator or node lines up to agree on the next block in sequence. That creates a hard limit on how fast the whole thing can move. A few seconds per block is normal. Even the faster ones rarely push past a dozen or two in sustained real world conditions.
The Internet Computer works differently. It splits the workload into separate subnets. Each subnet runs as its own independent replicated state machine with its own set of nodes reaching consensus. Right now there are around 50 of these subnets live. Each one produces and finalizes its own blocks on its own timeline. Add them together and you get the aggregate rate the dashboard displays. A single subnet does not have to carry the entire network load.
When one application or group of canisters gets busy, the impact stays mostly contained. Other subnets keep moving without waiting. The Network Nervous System coordinates new subnets and upgrades when needed, but day to day operation stays distributed.That design choice directly explains the high block rate. It is not magic or overclocking. It is parallel execution built into the architecture from the start.
More subnets means more parallel block production. The network scales capacity by adding more of these units rather than trying to make one giant chain faster and faster.On the transaction side the dashboard also shows several thousand transactions per second moving right now, with update calls (the ones that actually change state) running in the low thousands on average and much higher in bursts.
Queries, which are read only, move even faster because they often execute locally on one node without full subnet consensus. The combination lets the system handle real interactive workloads at speeds that feel closer to normal web apps than typical blockchain experiences.For the network itself this level of block production means genuine headroom. Over a million canisters are already deployed. Activity spans data centers in dozens of countries with hundreds of independent node providers.
The cycle burn rate reflects actual compute being used, not just theoretical capacity sitting idle. Congestion does not build up in one place because the load spreads across independent pieces. When demand grows, the system can add another subnet instead of fighting over space on a single shared chain. This changes what the broader blockchain space can realistically deliver. For years the discussion has circled around throughput limits.
Can a decentralized system ever match the performance people expect from centralized infrastructure without cutting corners on security or adding layers of complexity?
Many projects answer with rollups, side chains, or sharding proposals that still leave coordination overhead or new trust points.The Internet Computer shows one working path through its subnet model. Everything stays under one protocol with shared cryptography and governance, yet execution happens in parallel. Finality stays fast per subnet. State remains consistent within each unit.
Applications built as canisters get full on chain execution for compute, storage, and state without relying on off chain servers. That directly supports the sovereign cloud idea. A sovereign application runs entirely on the protocol. No company controls the backend. No single jurisdiction can easily shut it down. No hidden API keys or centralized databases sit between the user and the logic. With block production at this pace the system can support the kind of sustained, data heavy, interactive use that used to require big tech cloud providers.
Developers get performance without giving up ownership or censorship resistance.The industry feels the difference. Projects that still advertise future scaling roadmaps look slower by comparison when live numbers like these are available to check. Teams building on ICP can ship things that need reliable high volume execution today instead of waiting for promised upgrades. The total transaction count crossing hundreds of billions shows real usage accumulating over time, not just testnet experiments. It is not flawless.
Subnet management, canister limits per subnet, and ongoing protocol tweaks all require attention. But the current block rate is concrete evidence that the core scaling approach works in production. The network keeps adding subnets and refining parameters as load increases, and the metrics keep reflecting that growth.If you watch the dashboard over a day or two you see the finalization rate fluctuate with activity but stay in a high range. It is not a static marketing number.
It moves with actual demand across those independent subnets.This is what decentralized compute at web relevant scale starts to look like. Not every chain needs to copy the exact model, but anyone claiming to build the next layer of open infrastructure has to reckon with performance delivered at this level rather than promised for later. The Internet Computer is running the numbers now.
$ICP Mingleberrymedia.com
Ter Ber retweeted
🚨 $ICP HOPIUM JUST TURNED INTO GOVERNANCE REALITY
Proposal 142268 is open.
This is the kind of proposal people will look back on and say:
“That was when $ICP started opening the serious compute layer to the world.”
What it does:
• Opens the SEV subnet to public use
• Moves confidential compute closer to developers
• Expands authorized subnet access
• Pushes $ICP further beyond normal blockchain limits
Most chains are still fighting over:
• TPS
• Fees
• Bridges
• Sequencers
• Wrapped assets
• Cloud-hosted frontends
$ICP is moving at the subnet layer.
Confidential compute.
Public access.
On-chain governance.
Real infrastructure.
This is not another meme upgrade.
This is $ICP quietly building the foundation for sovereign apps, AI, enterprise systems, secure data, and tamper-resistant compute.
The market is asleep.
The NNS is awake.
$ICP by @dfinity is not trying to be the best blockchain.
It is trying to replace the stack underneath them all.
Vote accordingly. ♾️
Support the work / donate $ICP:
1e672d038cebc619d93186418fa98f6499dbdb9cfdfac54f366c61a4a4ee4362 – at Riyadh, Kingdom of Saudi Arabia
Ter Ber retweeted
🇨🇭Today I posted Part 2 of the Mini Series about the @SwissSubnet: youtube.com/watch?v=2FRRFIgz…
I had the chance to sit down with Dennis Platzl, CEO of Swiss Subnet.
We had a great conversation about $ICP technology, its real-world use cases, institutional adoption, cybersecurity, and decentralized cloud infrastructure.
We also talked about the solid progress Swiss Subnet is making both in Switzerland and internationally. With strong interest in tailored AI solutions and secure decentralized cloud on blockchain, Swiss Subnet is currently hiring developers and members for their Business & Development team. They’re looking for people who speak German, French, Italian, and English.
These roles will help them engage with institutions across Switzerland and beyond, and support their onboarding process. In my next video, I’ll share more details about their client pipeline and the jurisdictions their future clients are operating in.
Swiss Subnet is staying focused on sustainable growth and strong execution. Part of that includes signing Letters of Intent with various institutions in Switzerland, some of which are large organizations operating at the cantonal level. Stay tuned for Part 3.
These Unreal Metrics Place ICP over Solana
Internet Computer (ICP) (@dfinity) shows much stronger performance than @Solana in several key areas.
Right now, ICP processes 2,131 transactions per second. Solana manages only 1,344.
In the last hour, ICP handled 7.67 million transactions compared to Solana’s 4.84 million.
ICP’s peak speed reaches over 25,000 TPS, nearly four times higher than Solana. Its theoretical maximum is more than three times faster.
Most impressively, ICP offers instant finality with zero-second wait time. Solana needs about 13 seconds.
Transaction fees on ICP are extremely low at just $0.000046.
These metrics prove ICP’s superior speed, efficiency, and scalability. While Solana leads in market cap and users, ICP clearly wins on raw blockchain performance.
Ter Ber retweeted
The future of blockchain is not just transactions, it’s full internet-scale infrastructure.
ICP is redefining Web3 with fully on-chain applications, ultra-low fees, sub-second finality, and true decentralized computing.
@digitalauthpk @DfinityToday @dfinity
🤖 Made with AI
Ter Ber retweeted
We burned another 20K $ICP 🔥
I got many comments from people moaning about the CBR. I evaluate facts not feelings.
Fact: In Q4 25 we burned 125.9 ICP
Fact: In 15 days of May we burned 74K ICP = 3.5X better progression 🔥
These are just facts. DYOR.
Ter Ber retweeted
The Met's open-access collection runs to 500,000 works. Free to view. Hard to navigate.
Mareen W built VAULT to fix that - one clean interface for the entire collection. Rembrandt next to Egyptian faience, sorted however you want.
Yours to try: caffeine.ai/app-market/creat…
Ter Ber retweeted
Your app lives in Caffeine. Now your code can too - on your terms.
A lot of you have been asking for this. You want to open your codebase in your own editor, version it properly, work with your existing tools, and bring it back into Caffeine when you're done. That's exactly what GitHub import lets you do.
Export to GitHub, tweak it however you want, and pull it straight back in.
Your workflow, your tools.
Ter Ber retweeted
☁️Internal Cloud Engines demo today!
You configure your own engine, select nodes across AWS, Google, Azure, or independent hardware, and migrate between them without downtime. No vendor lock-in.
Security is built into the network protocol, not bolted on. No firewalls, no Cloudflare, no security team. Tamper-proof by default.
Any AI agent (Perplexity, Claude, Cursor, Caffeine) can deploy production apps to it directly. When AI builds software, the infrastructure it runs on needs to be hackproof. Cloud Engines solve that.
Governments and enterprises need digital sovereignty, but it usually means picking one cloud vendor over another. Cloud Engines give you actual sovereignty: you control the nodes, the jurisdiction, the providers, and you can switch any of them without interrupting your apps.
Super exciting!
Ter Ber retweeted
#Mission70 update: Proposal 141328 has just been approved and executed. Effective today, Gen-1 node rewards are computed using the reduced values (−40%).
Ter Ber retweeted
#Mission70 update: Wondering how the proposed voting reward reduction is computed?
I built an interactive simulator deployed on ICP that lets you explore the parameters and see the resulting inflation reduction and neuron rewards in real time.
Based on the current neuron snapshot, the reduction is 36.71%. Try it here: fe5jc-aaaaa-aaaam-aivbq-cai.…
Ter Ber retweeted
Caffeine v3.0 ☕️, Tuesday, April 7th
caffeine.ai
A new era begins on ICP in a few days. A huge Caffeine upgrade will take place. This is about more than the new features it packs: what can be built on the Internet Computer through chat alone shall advance by a quantum leap.
It will blow your mind, because you can go so much further: those wanting to build apps and services on self-writing cloud, where every technical task is automated and directed by natural language interaction, may rationally decide that the Internet Computer is the place to be thanks to the work of Caffeine Labs.
But, stunning as the v3.0 advance is, this only marks the beginning of a new beginning for ICP. Caffeine is just getting started, and will be joined by other major advances unfolding in the ICP ecosystem. Several things are coming together:
1) The architecture that powers Caffeine v3.0 will ensure it continues to grow more powerful as time goes on. We expect that it will be possible for non-technical people to build almost anything on the Internet Computer in the future. Eventually special features like "Caffeine Snorkel" will also make it possible to auto-migrate legacy apps and services to the Internet Computer.
2) For the first time, Caffeine will allow all users to publish the App Market, so that other users can install copies of their apps, and if they want, remix them, since apps from the market are mutable and can be changed as needed. This is important change will be followed by the activation of monetization features, which will kickstart a new open source onchain economy based around the AI-driven development of re-usable apps and services.
3) Caffeine apps use ICP Blob Storage (which is currently exclusively available to Caffeine owing to its ongoing development) to maintain copies of files and binary large objects. Coming advances will enable Caffeine apps to store data at the world's best $/GB — making the Caffeine the perfect plartform to use to build apps that must store and share large amounts of data (for example, for a photo sharing app, or a team communications platform where video meetings are routinely recorded and archived).
4) Caffeine will soon be joined by new ICP "cloud engine" functionality, and then, the Internet Intelligence Network (IIN) — two new parts of the Internet Computer. Cloud engines will provide enterprise users with much more control over how the Internet Computer hosts their apps and services. IIN will provide everyone with access to inference on open weights models at lower cost than is possible today (for example allowing a business to create an e-commerce website on the Internet Computer where every customer can be helped through product exploration by AI, without worrying about the cost).
5) Cloud engines, which are essentially private subnets that can be controlled via a control panel, enable the Internet Computer to provide a cloud experience that feels more familiar to prosumers and enterprises used to centralized clouds. Users will be able to select the nodes combined to power their cloud engines, and nodes can include cloud-on-cloud nodes, such as AWS instances, as well as traditional sovereign node hardware. These nodes can be chosen to match precise throughput needs, or regulatory requirements (e.g. GDPR). The Internet Computer is thus transforming into a mass market cloud solution.
6) Powerful additional advantages of cloud engines include a) since they are essentally Internet Computer subnets, as usual hosted apps and services are tamperproof and unstoppable, and don't require support by security and systems administration personnel, b) hosted apps can have their query throughput auto-scaled horizontally by adding nodes, which can be located in proximity to demand, and can have ther update throughput auto-scaled by splitting the engines — enabling auto-scaling where software does not have to be updated — and c) owners can traverse underlying compute vendors and geographies and jurisdictions, simply by adding and deleting nodes, which can be done without interrupting hosted apps.
7) The Internet Intelligence Network is an extension to the Internet Computer that we plan to propose to the NNS soon (which I have mentioned before). This will be powered by community-operated hardware devices, which crucially, in combination with new technology we are developing, shall shall facilitate provision of lower per-token inference costs than AI clouds can currently achieve within targeted market segments (low costs that are derived from the technology involved, in contrast to certain contemporary decentralized GPU networks that use token appreciation to drive unsustainable subsidization flywheels).
8) A key unique additional aspect of IIN is that inference is verifiable, just like onchain compute — that is, those supplying prompts and context will know that their results are correct, and shall not have to trust that the IIN nodes powering inference haven't changed their results (for example, by inserting advertising, or malicious content). The strength of the verification is configurable. In the default mode, there will be no discernible difference in cost or performance/time thanks to the techniques developed. In higher security modes, useful for special high-security purposes, verification cost will scale linearly with every node checking the inference, while the hit to performance/time remains very small. (Note that because inference is non-deterministic, inference verification is technically challenging and novel, and will advance the state-of-the-art).
In all, v3.0 is part of the Internet Computer's march towards the mainstream cloud market, which is predicted to generate $1 trillion in revenue in 2026, and $2 trillion in 2030.
While we see the Internet Computer as a revolutionary decentralized cloud platform, rather than a network beloning to the web3 industry per se, we note that after tokens (Bitcoin), and DeFi (Ethereum), onchain cloud is the first new decentralized network functionality to unlock mass market potential — and it's happeing exclusively on the Internet Computer, which is by far the most technologically advanced network the world has ever seen.
ICP is advancing to define an important new cloud sector, where the network is the cloud, and the cloud is self-writing.
























