March second, twenty twenty-six. A lot happening across AI and Bitcoin today. DeepSeek is days away from dropping a trillion-parameter open-source model built entirely on Chinese chips. Bitcoin is holding around sixty-six thousand as U.S.-Israeli strikes on Iran send oil up six percent and trigger hundreds of millions in liquidations. Stablecoins keep quietly eating the world, with Circle reporting record revenue and USDC circulation hitting seventy-five billion. And crypto venture capital is shifting hard toward infrastructure plays. Let's get into it.
So DeepSeek V4 is expected to drop sometime between March third and seventh, timed to coincide with China's Two Sessions parliamentary meetings, which is not subtle. This is a trillion-parameter mixture-of-experts model with roughly thirty-two billion active parameters per token. That's a similar active footprint to what we've seen from frontier models, but the total parameter count is massive.
What makes this interesting beyond the raw scale is a few things. First, it's natively multimodal from the ground up. Text, image, video, and audio all handled simultaneously, not bolted on after the fact. Second, it introduces some genuinely novel architecture. There's something called Manifold-Constrained Hyper-Connections for training stability, an Engram Conditional Memory system for context retrieval, and a new sparse attention mechanism for handling its one-million-token context window. Third, and this is the geopolitical angle, the entire thing is optimized for Huawei Ascend chips. No NVIDIA hardware. This is China demonstrating it can build competitive frontier AI without American silicon.
DeepSeek also plans to release V4 under an open-source license. That matters. It means global developers can access and modify the model without licensing fees, which is a direct competitive challenge to OpenAI and Google's closed approaches.
Meanwhile, Meta has launched Llama 4, pushing its own multimodal capabilities forward. Llama 4 is the latest in Meta's strategy of releasing powerful open models to commoditize the layer above their social platforms. And Mistral has a new three-billion parameter reasoning model out.
The pattern here is clear. The open-source frontier is advancing fast. You've got LLaMA 4, Qwen 2.5, DeepSeek V4, Mistral's reasoning models, all shipping within weeks of each other. The gap between open and proprietary models continues to narrow, and for many practical applications, it may not matter anymore. Organizations that want to control their AI stack, fine-tune for specific domains, or avoid vendor lock-in have never had better options. The question isn't whether open-source models are good enough. It's whether the closed model providers can justify their premium when open alternatives are this capable.
Bitcoin dropped to sixty-six thousand seven hundred on Sunday night as traditional markets got their first chance to price in U.S. and Israeli airstrikes on Iran. Oil surged to seventy-seven dollars. Asian equities dropped one and a half percent. About three hundred million dollars in crypto positions were liquidated over the weekend.
But here's what's actually notable. Bitcoin recovered to sixty-six five hundred and outperformed equities in Monday's risk-off session. That's meaningful. During the initial shock, Bitcoin sold off alongside everything else, but it bounced faster than stocks did. Some are reading this as Bitcoin starting to behave more like a geopolitical hedge than a pure risk asset. Arthur Hayes made the argument that the longer the U.S. engages in expensive military operations, the more likely the Fed is to expand the money supply to finance it, which would ultimately be bullish for Bitcoin.
Strategy, Michael Saylor's company, made its hundred and first Bitcoin purchase last week. Three thousand fifteen Bitcoin for two hundred and four million dollars at an average price of sixty-seven seven hundred per coin. That's below their overall cost basis of about seventy-six thousand. Total holdings now sit at seven hundred twenty thousand seven hundred thirty-seven Bitcoin. Say what you want about the leverage, but they keep buying below their average and they now hold more Bitcoin than most countries.
On the ETF front, the picture is genuinely mixed. Last week saw a billion dollars in inflows into crypto ETPs, led by seven hundred eighty-seven million into U.S. spot Bitcoin ETFs, snapping a five-week outflow streak. But zoom out and over nine billion has fled Bitcoin and Ether ETFs over the past four months. European investors appear to be buying the dip while U.S. institutional money keeps bleeding out.
Citigroup announced plans to integrate Bitcoin custody and servicing for institutional clients in twenty twenty-six, joining the wave of traditional banks building Bitcoin infrastructure. And NYDIG published research suggesting that if AI drives productivity-led disinflation, it could prompt easier monetary policy, which would create tailwinds for Bitcoin. An interesting thesis connecting the two biggest technology stories of the decade.
While everyone watches Bitcoin's price, stablecoins are having a moment that deserves more attention. Circle reported seven hundred seventy million dollars in Q4 twenty twenty-five revenue, up seventy-seven percent year over year. USDC circulation reached seventy-five point three billion, up seventy-two percent from the end of twenty twenty-four. Net income hit a hundred thirty-three million for the quarter, up from four million a year earlier. They're targeting a forty percent compound annual growth rate for USDC circulation. Those are serious numbers for what is essentially a payments infrastructure company.
On the Tether side, the U.S.-focused stablecoin USA-T just got its first reserve report from Anchorage Digital Bank. As of January thirty-first, there were about seventeen and a half million tokens in circulation with reserves of seventeen point six million, showing a small surplus. The reserves are held in segregated fiduciary trust accounts, primarily U.S. dollar cash and overnight reverse repos backed by Treasuries. It's a conservative, transparent structure designed to set a standard for onshore digital dollars.
Visa launched USDC settlement on Solana back in December, enabling near-instant transaction finality. Sony Bank in Japan signed an agreement with JPYC to study real-time transfers letting customers buy a yen-pegged stablecoin directly from bank accounts. The European banking consortium Qivalis is in talks with crypto exchanges ahead of a euro stablecoin launch in the second half of this year. Hong Kong and Shanghai authorities are testing blockchain for cross-border cargo trade data under Project Ensemble.
The total stablecoin market now exceeds three hundred billion dollars. The GENIUS Act in the U.S. has established a federal regulatory framework. Europe has MiCA. These aren't experiments anymore. This is financial infrastructure being rebuilt in real time. And the interesting thing is that stablecoin adoption doesn't depend on Bitcoin's price going up. It's driven by the simple reality that moving dollars on blockchain rails is faster and cheaper than legacy systems. That's a use case that sells itself.
Venture capital in crypto has deployed over two billion dollars so far in early twenty twenty-six, and the mix tells you where the smart money thinks value is accruing. The biggest rounds are going to stablecoin infrastructure, custody, and tokenized real-world assets. Not Layer 1 blockchains. Not meme coins. Infrastructure.
Rain raised two hundred fifty million for stablecoin payment infrastructure. BitGo pulled in two hundred thirteen million through its IPO for digital asset custody. BlackOpal raised two hundred million for a tokenized credit product. Dragonfly launched a six hundred fifty million dollar fund specifically targeting financial infrastructure, stablecoins, and DeFi. Their managing partner Haseeb Qureshi explicitly said the future lies in financialization, not non-financial crypto applications.
Smaller rounds are happening too. Based, a web trading and payment app, closed an eleven and a half million dollar Series A led by Pantera Capital with participation from Coinbase Ventures. t54 Labs raised five million for AI trust infrastructure with backing from Ripple. The International Finance Corporation invested forty million in Zetrix AI's digital infrastructure.
Coinsilium, a digital asset treasury company, announced it's shifting strategy toward prediction markets and event-driven finance while maintaining a non-dilutive Bitcoin accumulation strategy with a hundred eighty-two Bitcoin on its balance sheet. ProCap Financial, Anthony Pompliano's company, bought four hundred fifty Bitcoin, making it the nineteenth largest publicly traded holder.
What's happening is a maturation cycle. The bear market washed out speculative capital, and what's flowing back in is targeting companies that generate revenue, serve institutional clients, and build the plumbing that connects crypto to traditional finance. The VCs betting on this cycle aren't betting on number-go-up. They're betting that the financial system is being rewired, and they want to own the infrastructure layer.
Here's something to sit with. The two biggest stories in technology right now, AI and stablecoins, are both infrastructure stories. DeepSeek V4 isn't exciting because it's a chatbot. It's exciting because it's open infrastructure that anyone can build on. USDC isn't exciting because it's crypto. It's exciting because it's payment rails that work better than what banks offer. The value is accruing to the plumbing, not the applications. If you're thinking about where to pay attention, look at what's becoming infrastructure. That's where the durable value lives.