Bitcoin is bleeding through institutional channels. Spot Bitcoin ETFs just logged a tenth straight day of net outflows, dragging the cumulative total close to 3 billion dollars. BlackRock's IBIT alone accounts for over 2 billion of that, with one single session in late May seeing 527 million walk out the door. Bitcoin price slid to around 72,978 dollars after Trump's posture on Iran shook risk markets, though dip buyers are now stacking over 500 million in bids near the 70,000 level. Meanwhile Microsoft is making the biggest enterprise AI move of the year, the CFTC just greenlit US Bitcoin perpetuals, and a soft fork proposal called BIP-110 has the Bitcoin community arguing about chain split risk heading into August. Let's get into it.
Microsoft is no longer pitching AI as a feature. It's rebuilding Windows and Azure as a runtime for autonomous agents, and the scale of this bet is hard to overstate.
Here's the architecture. There's a Windows Agent Runtime shipping to Insiders in June, which sandboxes agents with per-capability grants for file system, network, and app launch access. There's the unified Microsoft Agent Framework, which finally merges AutoGen and Semantic Kernel into one commercially supported SDK. Copilot itself is being rebuilt as a multi-model orchestration layer, routing tasks across OpenAI's GPT-5.5, Anthropic's Claude, and lightweight on-device models, configurable at the enterprise tenant level. And there's a Windows Agent Store coming as a distribution channel.
Copilot Studio's computer-use agents just went generally available across all commercial geographies. These agents can click, type, and navigate any graphical interface, meaning automation without APIs. Pricing runs about 4 cents per step through Copilot Credits, so a four-step workflow costs 16 cents. The enterprise hook is audit logging into Microsoft Purview, Azure Key Vault for credentials, and isolated execution through Windows 365 Cloud PCs.
The rollout is already happening inside Office. Excel got edit with Copilot, and agentic capabilities are now in PowerPoint, Word, and Outlook. The team's key insight was treating Excel as a low-resource programming language, similar to how they trained models on Kusto Query Language internally.
And then there's the EY deal. EY and Microsoft just committed over 1 billion dollars over five years to deploy AI across enterprise functions. EY rolled Copilot to 150,000 users with reported 15% productivity gains, and Microsoft 365 E7 Frontier Suite is going to over 400,000 EY colleagues. Internal numbers EY is publishing include 95% lead time reduction in finance operations and 90% cuts in manual tax data extraction work.
The strategic read is brutal for competitors. With 300 million Microsoft 365 seats and native connectors into Graph, Fabric, SharePoint, and Azure AI Foundry, the gravitational pull toward staying inside the Microsoft stack is enormous. Temporal and LangGraph are competing on portability and model agnosticism, but Microsoft is competing on lock-in disguised as convenience. Build 2026 next week, June 2nd and 3rd, will fill in the rest.
Spot Bitcoin ETFs are in their longest outflow streak since launch. Ten straight days of net redemptions, roughly 3 billion dollars gone, and BlackRock's IBIT is responsible for about 62% of that. On May 27 alone, IBIT shed 527.8 million, the second largest single-day outflow on record. Fidelity's FBTC, Grayscale's GBTC, ARK, and Valkyrie all contributed, but this is overwhelmingly an IBIT story.
Ether ETFs are worse on duration, with 14 consecutive sessions of outflows totaling around 694 million. Meanwhile, smaller and newer products are pulling capital. XRP funds attracted 35 million between May 20 and 29. Hyperliquid's HYPE ETFs have crossed 100 million in cumulative inflows. There's a clear rotation happening at the margin away from the core BTC and ETH funds toward selective alt themes.
What triggered the bleed? Macro and geopolitics. Bitcoin dropped about 11% from above 82,000 on May 6 to under 73,000, with the Iran framework uncertainty and Trump's Hormuz commentary providing the catalyst. One analyst is calling the streak a contrarian indicator, and Cory Klippsten at Swan reminded everyone that retail sentiment still matters because BlackRock doesn't own Bitcoin, not by a long shot.
There's also a Strategy story worth watching. Michael Saylor's firm moved 411 BTC to Coinbase Prime on May 29 in two transfers of roughly 205 and 206 BTC. It hasn't been confirmed as a sale, but it's putting fresh attention on whether the treasury financing model holds up under sustained price pressure. BSTR's co-founder Sean Bill went further, saying much of the Bitcoin treasury space is full of carnival barkers who can't actually deploy capital effectively.
The deeper question is whether the ETF channel still absorbs supply. For most of 2024 and 2025, ETFs were the demand sink that mattered. Right now they're the supply source. If macro stabilizes, this reverses fast. If it doesn't, corporate treasuries and direct purchases need to pick up the slack.
There's a fight brewing in Bitcoin that could matter more than the price action. It's called BIP-110, the Reduced Data Temporary Softfork, and it's pushing the long-running debate about data inscriptions from relay policy into consensus rules.
Here's what it does. BIP-110 would cap new output script lengths at 34 bytes and limit data-heavy outputs during a one-year deployment window. The target is data uses that fill block space without serving payments. Activation requires miner signaling on Bit 4, hitting a 55% threshold over a 2,016 block period, with an August activation window in view.
The controversy isn't subtle. Supporters frame it as cleanup, restoring Bitcoin's focus on monetary use at the consensus layer. Critics warn it alters Bitcoin's neutrality and risks a chain split. If miner signaling is contested, you end up with upgraded nodes rejecting blocks that non-upgraded miners produce, which means deposit and withdrawal disruption on exchanges, wallet inconsistency, and confirmation chaos right when liquidity needs to be reliable.
Bitcoin Knots has implemented support, but miner alignment remains the open question. And the timing is awkward, because just this week the full text of the US Constitution was inscribed onto Bitcoin via Ordinals at block 951,492. Whatever side you're on, that inscription is a real economic event paying real miner fees, and it's exactly the kind of activity BIP-110 wants to constrain.
Zoom out and there's a broader theme. Cloudflare's CEO Matthew Prince argued this week that AI bot traffic will exceed human traffic by early 2027, and he's proposing stablecoin-based micropayments and the HTTP 402 model to monetize agent access to web content. He estimates the internet may need 10 to 100 million transactions per second of payment throughput to make this work. Bitcoin can't do that on-chain, and Lightning is the obvious answer, but the irony is sharp. Just as the world might need Bitcoin to settle massive volumes of machine-to-machine micropayments, the Bitcoin community is debating whether to tighten what can even sit in a transaction. The block space allocation debate is about to get a lot more practical.
Two regulatory and competitive moves worth flagging. First, the CFTC just brought Bitcoin perpetual futures onshore. KalshiEX got approval to list its BTCPERP contract as a CFTC-registered designated contract market, and Coinbase Financial Markets received staff-level relief to access certain Deribit products. This is the offshore perpetuals liquidity question, which has been dominating crypto derivatives for years, finally getting a regulated US answer.
The CFTC paired the approval with an advisory clarifying that round-the-clock trading is appropriate for crypto but might not fit other sectors. Coinbase is moving fast on the opening, integrating with Deribit to give eligible US institutional clients access to global perpetuals and options. JPMorgan's Jamie Dimon, meanwhile, is escalating the fight over stablecoin yield in the CLARITY Act debate, telling Brian Armstrong directly that banks will not accept stablecoin issuers offering deposit-like rewards. The bank versus crypto firm divide on yield is now the defining policy question.
On the AI side, Perplexity raised 500 million dollars in a Series D, valuing the company around 14 billion. The focus is shifting to commerce. They're building what they call a Buy Layer, a structured product graph beneath conversational search that returns real-time merchant cards. Allbirds, Ridge Wallet, and several Shopify Plus stores are in closed beta. Conversion tracking is cookieless and GDPR-compliant.
The market context matters. Perplexity is now in second place behind ChatGPT in chatbot market share, somewhere between 8% and 11% globally, with around 85 million monthly active users in the US. Google is responding with AI Overviews and Shopping AI cards, which appear in about 41% of product-intent queries, but click-through rates are 2.1% versus 8.4% for traditional Shopping carousels. That gap is the entire opportunity Perplexity is chasing.
The catch is legal. CNN just sued Perplexity, alleging unauthorized copying of more than 17,000 stories, videos, and images. The complaint targets both ingestion via PerplexityBot and the retrieval-augmented generation outputs, plus Lanham Act trademark dilution claims. This is now the landmark test case for whether RAG-based AI search is fair use or systematic infringement, and the answer will shape the economics of AI search for years.
One observation to close. Microsoft is racing to make agents persistent, governed, and locked into its stack. Bitcoin ETFs are bleeding while the protocol layer argues about what data deserves to exist on-chain. These look like separate stories, but they're the same story told twice. The next phase of both AI and Bitcoin is about who controls the runtime, not who builds the best product. Watch where the orchestration layer settles, because that's where the margin lives.