Meta just charged money for one of its own AI models for the first time ever. US regulators blew past a major stablecoin deadline. Bitcoin miners are getting squeezed hard, with one publicly traded miner dumping its entire treasury. And PwC and OpenAI are teaming up to replace call center scripts with autonomous agents. That's what we're getting into today.
Let's start with Meta, because this is a real shift. Muse Spark 1.1 launched as Meta's first paid AI model. That's a genuinely big deal for a company that built its entire AI reputation on giving Llama away for free.
The pricing is aggressive. $1.25 per million input tokens, $4.25 per million output tokens. That undercuts Claude and OpenAI's comparable tiers by roughly 50 to 75%. New developers get $20 in free credits, and the API is OpenAI-compatible, meaning you can literally swap it in without rewriting code.
On paper, the model is built for agentic work. Multimodal, 1 million token context window, active context management, tool use, and it can directly control a computer. It supports MCP servers and custom skills. Meta is positioning it against Claude Haiku 4.5 and GPT-5 mini, and internally it's meant to gradually replace the Llama models running WhatsApp, Instagram, Facebook, and the smart glasses.
Here's the interesting part strategically. Meta spent years arguing that open-weight models were the future and that closed APIs were a losing bet. Now they're running the exact playbook they used to criticize. Muse Spark is closed-weight, US-only at launch, and monetized per token. Zuckerberg is out on X pitching it like any other API vendor.
Why the flip? Two reasons, probably. First, the Superintelligence Labs reorg needs a revenue story that isn't just ads. Second, the frontier of agentic coding is where the actual developer money is right now, and you can't fund that kind of R&D by giving weights away. The benchmarks are mixed. Strong on agent coordination, weaker on pure coding tests like SWE-Bench Pro and Terminal-Bench. So it's not a Claude Sonnet killer. It's a price-war entrant.
What this really signals is that the AI pricing floor is collapsing. If Meta can profitably sell agentic inference at $1.25 per million input tokens, everyone else has to justify their markup. Expect Anthropic and OpenAI to respond, probably not on price directly, but on capability tiers and enterprise bundles.
Related theme, different angle. PwC and OpenAI announced a joint offering they're calling agentic contact and service solutions. Translation: they want to replace scripted IVR trees and rule-based chatbots with goal-oriented AI agents that actually resolve problems.
The pitch is an AI voice and digital agent built on OpenAI's multimodal APIs, connected to enterprise systems, capable of taking action instead of just routing tickets. They've stood up a joint Center of Excellence to shorten deployment timelines, and the framing is what they call an agentic front office, where marketing, sales, commerce, and service all share context.
This isn't happening in a vacuum. Deutsche Telekom moved from pilot to production this year, with AI embedded in live voice calls across the network. Their Magenta AI Call Assistant does real-time translation, note-taking, and post-call summaries. Internally, 50,000 plus employees are now on ChatGPT Enterprise, and AI incident management is auto-resolving about 70% of network incidents, targeting 90%. And over in Japan, transcosmos upgraded its CX platform and reported that swapping a voice bot for a voice AI agent lifted self-service resolution from 19% to 52%. That's not a marginal improvement. That's a category shift.
The pattern to notice: OpenAI's enterprise strategy is now clearly embedding through consulting firms with existing client relationships. Anthropic is doing the same thing with Blackstone. The frontier labs figured out that selling APIs to developers is fine, but the real enterprise dollars come through Accenture, PwC, and Deloitte's client books.
One caveat worth flagging. Live-call AI in customer service is where hallucinations become expensive very quickly. Deutsche Telekom is emphasizing European data centers and GDPR compliance, and regulators in Germany and Brussels are already circling. Retraining and workforce impact remain vague across all these announcements. So the technology is arriving faster than the governance around it.
Yesterday marked one year since the GENIUS Act was signed. That was supposed to be the deadline for US regulators to finalize the rules governing stablecoin issuers. They missed it. All of them.
Six agencies were on the hook: the Fed, OCC, FDIC, NCUA, Treasury, and the interagency coordinators. Instead of final rules, they've issued 10 proposed rulemakings. Treasury put out four, covering framework consistency, foreign issuer registration, and AML compliance. The OCC has two, the FDIC one, the NCUA one, plus an interagency proposal to harmonize supervision.
Here's the mess. The act still takes effect January 18, 2027, or 120 days after final rules are issued, whichever is later. So issuers have to prepare for a regime whose actual requirements haven't been written down yet. There's no enforcement mechanism for the missed deadline. No penalty. No automatic fallback. Just regulatory silence.
The substantive rules that remain unfinished include customer identification, anti-money-laundering, sanctions programs, reserve asset requirements, redemption processes, capital, and risk management. Basically all the things that determine whether running a stablecoin business is actually viable.
One interesting piece is the direct versus indirect customer distinction in the AML proposals. Direct customers of a permitted stablecoin issuer would trigger full customer identification and monitoring. Indirect customers, meaning users who only touch the stablecoin through an exchange or a wallet provider, put the issuer in a weird position of having to understand the intermediated structure without necessarily knowing the end user. Nobody has really figured out how that works at scale, especially with unhosted wallets.
Meanwhile the industry is moving anyway. Sony Bank in Japan got OCC approval for a national trust bank plan to handle USD stablecoins. Circle, Ripple, BitGo, Fidelity Digital Assets, and Paxos all got conditional approvals to convert from state to national trust bank charters. And Tether, notably, is on a 2-year countdown before it potentially loses its position on US crypto platforms under the GENIUS framework.
The pattern is familiar. Congress passes a bill, sets an ambitious timeline, regulators fall behind, industry consolidates around the entities that can navigate the ambiguity. Circle and Paxos win by default. Smaller and foreign issuers get squeezed. Whether that's the outcome Congress intended is a different question.
Bitcoin mining is having a rough stretch, and it's telling us something about where this cycle sits.
The network just went through its 14th difficulty adjustment of the year, dropping 5% to 127.17 trillion. That followed a 7.9% hashrate collapse over 10 days, partly driven by an ice storm in Texas that forced miners offline. Hashrate fell from about 1.13 zettahash to roughly 690 exahash before partially recovering. Hashprice sits around $31 per petahash per day, which is 37% below the October 2025 peak.
All 8 of the most recent adjustments have gone in one direction or the other with sizable moves, averaging over 5%. That's unusual volatility for the difficulty algorithm and it reflects miners cycling on and off as margins swing.
The more revealing story is Bitdeer. The Singapore-based miner just liquidated its entire Bitcoin treasury. Sold 189.8 BTC recently, offloaded 943.1 BTC in total, and now holds zero Bitcoin on the balance sheet excluding customer deposits. This is a company that just became the largest publicly traded self-miner by hashrate, passing Marathon at around 63 exahash per second. And they're dumping every coin.
Why? They're pivoting to AI. Raising $325 million in convertible senior notes, potentially $375 million, to fund high-performance computing, AI cloud services, proprietary ASIC development, and data center expansion. The old hodl-everything mining playbook is being replaced with a treasury-lean, diversify-into-AI-compute playbook.
On-chain data backs up the stress. Miner-to-exchange transfers surged 2,150% versus the 90-day baseline. The Puell Multiple is at 0.71, well below the profitability threshold. Hash Ribbons are showing capitulation.
But here's the counterweight. Long-term holders now own a record 14.85 million BTC. The coins miners are selling are being absorbed. Older wallets, 7 to 10 years old, are moving, but analysts characterize it as organic distribution, not panic. Price has held a range in the low 60s despite all this supply pressure.
That's actually the story. Miner capitulation without price collapse means demand is quietly stronger than the headlines suggest. Whether that leads to the next leg up or just a longer grind sideways depends on macro conditions the miners can't control.
If Meta is now charging for AI, PwC is selling agents to call centers, and the biggest miner just dumped every coin to buy GPUs, the boundary between the AI economy and the Bitcoin economy is getting a lot thinner than it was 12 months ago. Watch which side the capital flows toward next quarter.