Big day for AI competition — SpaceXAI dropped Grok 4.5, priced aggressively to undercut Anthropic and OpenAI. Bitcoin is drifting near a long-term power law support line Fidelity's been tracking since 2015, unmoved by fresh US strikes on Iran. The BIP-110 fork debate is heating up with Saylor and Adam Back on one side and miner support sitting at zero. Adam Back's own 30,021 BTC treasury deal just lost its funding structure. And Hyperscale Data crossed 1,000 BTC in treasury. Let's get into it.
SpaceXAI — the company formerly known as xAI, now folded into the SpaceX umbrella — just shipped Grok 4.5. This is their first model since going public, and Elon is calling it an Opus-class model. That's a specific claim. He's saying it's roughly comparable to Anthropic's Opus 4.7 in capability, but faster and dramatically cheaper.
The pricing is where this gets interesting. Grok 4.5 is $2 per million input tokens and $6 per million output tokens. Compare that to Opus 4.7 at $5 in and $25 out. That's a 4x discount on output pricing against the model they're benchmarking themselves against. OpenAI's GPT-5.6 Sol variant runs $5 and $30. Even OpenAI's cheaper Luna tier at $1 and $6 is basically matched by Grok on output.
The model was trained on tens of thousands of NVIDIA GB300 GPUs, and it's the default engine inside Grok Build, their coding agent. There's also a deep Cursor partnership — reports of investment talks up to $10 billion, or an acquisition offer as high as $60 billion. Nothing confirmed.
On benchmarks, Grok 4.5 is competitive but not outright best-in-class. That's the honest read. It's not beating Opus or GPT-5.6 on every eval. But if you're building products and burning tokens at scale, price and latency matter more than winning benchmarks by two points. Twice the token efficiency at less than half the cost is a real wedge.
The broader picture: SpaceXAI is also shipping Grok integrations into Word, PowerPoint, Databricks, Amazon Bedrock, and Warp. They're not just selling a model — they're wiring Grok into the actual surfaces where knowledge workers spend their day. EU rollout is delayed to mid-July. And this is now a four-horse race at the frontier: OpenAI, Anthropic, Google, and SpaceXAI. Nobody was betting on Musk's shop being in that group 18 months ago.
Something quieter but arguably more consequential is happening in healthcare. Real agentic AI is moving into clinical workflows — not chatbots, not summarizers, actual decision-support systems influencing patient care.
East Kent Hospitals in the UK became the first NHS trust to deploy a tool called MEMORI. It ingests routine clinical data — blood tests, vitals, medications — and generates an infection risk score for each patient. The retrospective evidence suggests it can flag deterioration up to 72 hours earlier than standard early warning systems. That's a huge intervention window for sepsis, which is one of the biggest killers in hospital settings.
In Taiwan, NVIDIA and Foxconn are backing a $1.5 billion initiative called Healthy Taiwan. They're building what they describe as an AI team for every clinician — multiple specialized agents, an ECG agent, a diagnosis agent, all coordinating on a platform called CoDoctor. They're using NVIDIA's Omniverse to run digital-twin simulations before anything touches a real patient.
Israel is going further. They're running regulatory sandbox trials at three major hospitals for AI that actually makes decisions — not just suggests them. Home-based heart failure monitoring that can adjust treatment against a physician-defined protocol. Home ultrasound analysis where the AI decides if a scan is normal enough to skip specialist review.
And in Hong Kong, a radiology department is running an entirely private, on-premise AI on AWS Bedrock behind PrivateLink — no external API calls — handling patient WhatsApp inquiries in mixed Chinese-English, quoting prices across 1,000+ exam types.
Here's what ties this together. The debate has shifted. It's no longer whether AI belongs in clinical settings. It's about liability, oversight, and when the AI should refuse to decide. That's a much more mature conversation than we were having 12 months ago.
Some real signal in Bitcoin venture funding this week, and it's all pointing in one direction — back to the base layer.
Ask Labs, backed by Tether, closed a $5.2 million seed round to advance Arkade — a programmable execution layer built directly on Bitcoin. Tether led it. Ego Death Capital, Anchorage Digital, and Epoch VC participated. Total institutional backing on this project is now $7.7 million. The pitch is settling payments, lending, escrow, and conditional transactions on Bitcoin's base layer without wrapped tokens or cross-chain custody.
Meanwhile, Tether is finally bringing USDT back to Bitcoin natively via the RGB protocol. This is a full-circle moment — USDT originally launched on Bitcoin's Omni Layer back in 2014 before migrating to Ethereum and Tron. The commercial rollout is being led by UTEXO, which raised $7.5 million in March with Tether as a primary investor. Launch is expected within weeks. RGB uses client-side validation, anchors to Bitcoin for security, and is Lightning-compatible for near-instant settlement.
Think about what this means. The largest stablecoin issuer on earth — $184 billion in market cap — is deliberately routing its future infrastructure through Bitcoin instead of expanding further on Ethereum or Solana. That's not sentiment. That's a capital allocation decision.
On the treasury side, Adam Back's BSTR deal to launch with 30,021 BTC hit a wall. Cantor Fitzgerald and BSTR are renegotiating terms because the original financing package stopped being binding. It's a live test of whether investor appetite for pure-play Bitcoin treasury vehicles is still there at scale, or whether Strategy has already absorbed most of that demand.
And there's now the BIP-110 fight. It would cap arbitrary data on Bitcoin for a year to push back against Ordinals-style spam. Miner support is currently zero. Saylor and Adam Back are warning that turning a spam dispute into a consensus fight is a bigger risk than the spam itself. That's probably right.
Corporate Bitcoin treasuries are moving from a story about accumulation to a story about capital structure. Three data points from this week make that clear.
First, SpaceX joined the Nasdaq-100. That triggers roughly $4.3 billion in passive inflows from index-tracking funds. SpaceX holds 18,712 BTC and disclosed about $1.2 billion in Bitcoin fair value on its balance sheet. So every pension fund, every 401k index allocation, is now buying a slice of Bitcoin exposure through standard equity infrastructure without anyone approving a new mandate. That's the passive-flow adoption channel finally opening.
Second, Strategy Inc. — Saylor's company — issued a $5.6 billion digital credit facility backed by its Bitcoin reserves. It was oversubscribed 3.2x in 48 hours. The mechanism ties borrowing capacity to Bitcoin-collateralized stablecoins, so companies can access working capital without selling BTC. This is the piece that's been missing. Bitcoin as productive collateral, not idle treasury.
Third, Block rolled out on-chain proof-of-reserves for its 8,883 BTC, covering both the corporate treasury and the Cash App and Square rails. That makes them the 14th-largest corporate Bitcoin holder with independently verifiable holdings. Post-FTX, that kind of transparency is becoming table stakes.
But the picture isn't uniformly bullish for every treasury vehicle. Eric Trump's American Bitcoin — 8,000 BTC in the treasury, 17th-largest corporate holder — just did a 1-for-15 reverse split to avoid Nasdaq delisting. The stock is down roughly 94% since its September 2025 IPO. Q1 net loss of $81.8 million, even with mining margins above 50%. The treasury grows, the equity value collapses. That's the risk of the pure-play model when liquidity is thin and dilution is heavy.
And separately, Empery Digital actually sold Bitcoin from its treasury to fund an AI data center pivot. Shares rose on the news. So the market isn't uniformly rewarding the accumulate-forever strategy anymore. It's starting to price in what the Bitcoin is being used for.
One prediction: the companies that survive the next Bitcoin drawdown won't be the ones with the biggest stack. They'll be the ones who figured out how to borrow against it without selling it.