Meta just dropped a frontier model priced to undercut everyone. Circle got a national trust charter with ten days to go before the GENIUS Act deadline. BlackRock shuffled another billion in Bitcoin through Coinbase custody. And a US CBDC ban is quietly becoming law tonight without Trump's signature. Bitcoin is hovering near 64,000, up almost 10% for July, but traders are twitchy about what August brings. Let's get into it.
Meta released Muse Spark 1.1 this week, and the story here is really about pricing pressure on the frontier model market. On the benchmarks, Meta claims Spark 1.1 is competitive with Claude Opus 4.8, Gemini 3.1 Pro, and GPT-5.5 across the agentic and coding tests everyone cares about right now — SWE-bench Verified, Terminal-bench, BrowseComp, OSWorld. Standard flex.
But look at the price tag. $1.25 per million input tokens, $4.25 per million output tokens. Compare that to GPT-5.5 at $5 in and $30 out, or Claude Opus 4.8 at $5 and $25. That's not a small discount. On output tokens, Meta is roughly seven times cheaper than OpenAI's flagship. Gemini 3.1 Pro sits in the middle at $2 and $12.
On top of that, Spark 1.1 ships with a 1 million token context window, zero-shot tool use, and multi-agent orchestration built in. Meta is throwing $20 in free API credits at developers to get them experimenting. Zuckerberg is calling it high value, low price. That's the pitch.
Here's the interesting tension. Analysts are pointing out — correctly — that price isn't the whole game for enterprise. Governance, security, reliability, support contracts, indemnification. Those matter more than saving a few thousand dollars on inference. And frontier models are ludicrously capital-intensive to train and serve. Meta may be pricing below cost to grab share, and history says that pricing eventually normalizes.
But if you're a CTO negotiating your next Anthropic or OpenAI contract, you now have leverage you didn't have a week ago. Multi-model procurement just got easier to justify. Expect quiet price adjustments from the incumbents in the next couple of quarters — probably not headline cuts, but committed-use deals and volume tiers that suddenly get more generous.
While everyone argues about model pricing, the actual deployment stories in enterprise are getting concrete. And the numbers are starting to look serious.
Siemens is now handling 90% of inbound calls autonomously through Amazon Connect Customer AI Agents. Ninety percent. Either resolved directly or intelligently routed, no human touching it. The AI queries their internal employee directory in real time through Lambda, pulls CRM data, orchestrates callbacks so callers don't wait on hold. When something's too complex, it escalates with full context so the human doesn't start from zero.
Commonwealth Bank of Australia is running an AI orchestration agent built on Microsoft Copilot Studio and Dynamics 365. In May, 84.6% of self-service messaging interactions were resolved end-to-end without a human. They're now extending it beyond retail banking into the rest of the bank.
Naturgy in Spain is over 50% autonomous on technical assistance calls, targeting 90% by year-end, with customer satisfaction sitting at 9.4 out of 10. They tell callers upfront they're talking to an AI and offer a human option — and satisfaction still hits 9.4.
Deutsche Telekom is going further. They're not just adding AI to customer service, they're rewiring the whole operating model. 50,000 monthly active users on ChatGPT Enterprise internally. A 546% increase in AI tool usage since early 2026. Live translation and post-call summaries baked into the voice network itself.
Here's what's worth noticing. Twelve months ago, agentic customer service demos were mostly theater. The 90% number was aspirational. Now Siemens is in production at that level. The technology crossed a threshold, and the enterprise buyers who moved early are locking in real cost structure advantages. The ones still running pilots are about to find themselves competing against companies whose unit economics look fundamentally different.
Let's talk about the plumbing underneath Bitcoin ETFs, because there's a lot of noise about big BlackRock transfers and most of it is being misread.
Over four days ending around July 2, BlackRock moved about 20,359 Bitcoin — roughly $1.22 billion — into Coinbase. This week they added another 951 BTC, worth about $59 million. On-chain trackers immediately started asking: is BlackRock dumping? No. These are ETF operational flows. When IBIT creates or redeems shares, actual Bitcoin has to move in or out of Coinbase Custody, which is IBIT's principal custodian. Deposits usually mean share creations, not sales.
Separately, BlackRock also withdrew 883 BTC from Coinbase Prime — about $55 million — which is part of a diversification pattern. Back in 2024, BlackRock amended its custody agreement to require 12-hour withdrawal capability and started onboarding Anchorage Digital as a second custodian. So Coinbase is still dominant, but not exclusive. That's healthy market structure.
And on the government side, the US Marshals Service just signed a digital asset custody deal with Coinbase Prime for federally seized crypto. That's a meaningful institutional stamp. Coinbase Prime is now custodying assets for asset managers, corporates, and now the federal government.
Bigger picture — BNY Mellon just made USDC the first stablecoin supported on its Digital Asset Custody platform for institutional clients. Custody, transfers, minting, burning, all under one roof, alongside traditional cash management. This is the merger of TradFi and stablecoin infrastructure happening in real time, at one of the oldest banks in America.
The uncomfortable counterpoint: CoreWeave just raised $20 billion in financing for AI infrastructure. That's a lot of institutional credit appetite going to GPUs rather than Bitcoin treasuries. Some analysts argue AI is now directly competing with Bitcoin for speculative capital and macro liquidity. And Empery Digital, a Bitcoin treasury company, just sold about half its BTC stack to pivot toward — you guessed it — AI data centers. Not every treasury story ends well.
Ten days until the GENIUS Act rulemaking deadline. July 18 is when seven federal agencies — OCC, FDIC, Fed, NCUA, Treasury, FinCEN, and OFAC — are supposed to publish final implementation rules for US stablecoins. The draft rules already tell us where this is going.
1-to-1 reserves in eligible assets: cash, Fed balances, insured deposits, Treasury bills, overnight Treasury repos. No rehypothecation for most uses. Monthly reserve reports certified by both CEO and CFO, with third-party attestation. A $5 million minimum capital floor for new federal issuers, more for larger ones. Board-approved AML programs. Sanctions screening. And explicit token freeze and block capability as a regulated obligation.
On liquidity: at least 10% of outstanding stablecoins redeemable same-day, at least 30% within five business days, redemptions at par within two business days. Under stress — if 10% of issuance is requested inside 24 hours — issuers get up to seven calendar days to complete, with regulator notification.
Effective date is 120 days after final rules publish, or January 18, 2027 at the latest.
Circle just won final OCC approval to form First National Digital Currency Bank — a national trust bank. USDC becomes the most federally regulated stablecoin in the market. Circle crossed a $10 billion market cap, which triggers additional GENIUS Act oversight. They also hold about $73 billion in reserves earning around $3 billion a year in interest. Under GENIUS, that yield can't be paid to token holders — it stays with Circle and distribution partners.
Which brings us to the threat. A consortium called Open Standard announced Open USD, backed by more than 140 companies including Visa, Mastercard, Amex, Stripe, Coinbase, BlackRock, and Google. The pitch: redistribute nearly all reserve income back to consortium partners. That directly attacks Circle's core revenue model. OUSD isn't live yet, but the math is menacing enough that CRCL stock has been under pressure, trading near $63 with technical support around $61.
And almost as a footnote — the 21st Century ROAD to Housing Act is becoming law tonight without Trump's signature. It includes a ban on a US CBDC through 2030. So the American stablecoin path is now formally locked in: private, regulated, dollar-backed tokens. No government-issued digital dollar. That's the settled US position.
One prediction to close on. If Meta's pricing holds for six months and Siemens-style 90% autonomy becomes the enterprise baseline, the cost of a customer service interaction is about to fall by an order of magnitude. Companies that don't rebuild around that number are going to look expensive very quickly.