Monday, July 20. Here's what's moving. Elon Musk says xAI's next model, Grok 4.6, wraps initial training this week — 2 trillion parameters aimed squarely at China's Kimi K3. Bitcoin is flat near $64,000, weighed down by lingering Kimi selloff and oil spiking on U.S.-Iran tensions. Michael Saylor's Strategy sold another $263 million in stock and paused Bitcoin buys for a second straight week, building a $3.2 billion cash pile. Saylor also came out swinging against BIP-110, the proposed spam filter fork, calling it a bad idea. And Strategy just launched a Bitcoin Banking Adoption Index — Fidelity leads at 71%, the average bank sits at 32%. Let's dig in.
The frontier model race just got louder. Musk announced over the weekend that Grok 4.6 — a 2 trillion parameter model — finishes its initial training run this week. He's claiming it will outperform Moonshot's Kimi K3, which itself is a 2.8 trillion parameter open-weight model that dropped just days ago and rattled chip stocks hard enough to spill into Bitcoin.
The interesting claim isn't the parameter count. Kimi K3 is bigger on paper. What Musk is pitching is efficiency — that Grok 4.6 matches the inference speed and token economics of the 1.5T Grok 4.5 while scaling reasoning capability well past it. Some analyses floating around suggest Kimi costs roughly three times more to run than xAI's architecture. If that holds up under real benchmarks, it's a meaningful gap.
The whole thing runs on Colossus, xAI's Memphis compute cluster, which is pushing toward a million GPUs in its second expansion. That's the actual moat here — not the model, the infrastructure underneath it.
Meanwhile, Moonshot is reportedly pushing for a Hong Kong IPO at a 30 billion dollar valuation, and Alibaba just made Qwen open-weight. The Chinese labs are moving fast and giving the weights away, which is squeezing the closed-source players from a direction OpenAI didn't really plan for.
Watch what happens when the benchmarks actually land. Musk has a habit of pre-announcing. But if Grok 4.6 delivers Kimi-level performance at a third of the compute cost, that reshapes the economics of running frontier AI — and the market reaction to Kimi last week suggests investors are finally pricing in that efficiency, not scale, is the real variable.
While everyone argues about AGI timelines, agentic AI is quietly getting deployed in hospitals. And not in the toy demo way — in the FDA-cleared, doctor-oversight, actually-changes-outcomes way.
Bunkerhill Health just closed a 55 million dollar Series B led by Khosla Ventures for its Carebricks platform. It lets health systems build and deploy their own clinical AI agents — over 20 live at UTMB alone. One of them, a coronary calcium detection agent, flagged a case that led to a life-saving triple bypass. Nephrology wait times dropped more than 50%. Lung nodule follow-ups are 80% faster.
Separately, UpDoc has an FDA-cleared agentic platform for chronic care running at Cleveland Clinic, Allegheny Health, and UCSF. It handles medication adjustments and follow-up testing between visits, with physicians in the loop.
And Novant Health — 900 locations, 21 hospitals — just partnered with K Health to launch 24/7 AI-enabled virtual primary care across the Carolinas. Autonomous intake before the visit, EHR-grounded recommendations during it.
There's also a new benchmark worth knowing about: RadLE 2.0, from CRASH Lab. Instead of just measuring diagnostic accuracy, it scores AI systems on when they should defer to a human — confidence calibration, safety, handover readiness. Because the actual question for autonomous medical AI isn't can it be right, it's does it know when it's wrong.
The pattern here is clear. The AI agent story in 2026 isn't chatbots. It's narrow, regulated, workflow-embedded systems doing specific clinical tasks with measurable outcomes. That's where the actual revenue is going to show up first.
Strategy did something unusual last week. For the second week running, Michael Saylor's company sold shares — 263.5 million dollars worth of MSTR — and bought zero Bitcoin. Instead, cash reserves ballooned to 3.225 billion dollars.
The treasury still sits at 843,775 BTC, worth around 54.7 billion at current prices. But here's the uncomfortable number: cost basis is roughly 63.7 billion. That's about 9 billion dollars in unrealized losses on the Bitcoin position. CEO Phong Le said Strategy would only reassess its debt exposure if Bitcoin fell to the 8,000 to 10,000 dollar range — which reads as either bravado or a stress-test signal, depending on how you squint.
The pause matters because it breaks the flywheel narrative. For years the pitch was: issue stock, buy Bitcoin, Bitcoin goes up, market cap goes up, issue more stock. Now Strategy is issuing stock and holding cash to service preferred dividends. That's a different company.
And the corporate treasury landscape is fragmenting. Coinpedia's breakdown of the top 20 public holders shows three distinct camps. The static holders — Galaxy, Bullish, Twenty One Capital — haven't touched their Bitcoin since 2025. The active buyers — Metaplanet up to 43,000 BTC, Strive at 19,900, American Bitcoin Corp, Block — kept accumulating. And the trimmers — MARA sold over 18,000 BTC, Riot cut 2,325, even Strategy sold 3,588 earlier this year.
Meanwhile SpaceX's IPO filing revealed 18,712 BTC on the balance sheet. Combined with Tesla, Musk-controlled companies hold over 30,000 BTC. Saylor's takeaway: 25% of the Magnificent 8 now holds Bitcoin. The other 75% — Apple, Microsoft, Alphabet, Amazon, Meta, Nvidia — still don't.
The era of one uniform corporate Bitcoin thesis is over. What replaces it is treasurers actually making trade-offs — between accumulation, debt service, and share buybacks. That's messier, but it's also more honest.
Strategy also just launched something worth paying attention to: the Bitcoin Banking Adoption Index. It scores 25 major global banks on how deeply they've integrated Bitcoin products — trading, custody, digital asset offerings, financing.
The overall average across those 25 institutions is 32%. So the honest read is: banks are engaged, but nowhere near fully committed.
Fidelity dominates at 71%. That's not really a bank comparison — Fidelity Digital Assets launched back in 2018, they run FBTC, they do institutional custody and trading. They've had a seven-year head start on peers.
The real story is the tight cluster of Wall Street incumbents. BNY at 46%. Goldman Sachs at 45%. JPMorgan, Morgan Stanley, and Citigroup all at 43%. And each is playing a slightly different hand. JPMorgan is leaning into tokenized deposits and payments through Kinexys. Morgan Stanley is pushing regulated wrappers — the Morgan Stanley Bitcoin Trust and ETP access through E*TRADE. Citi's building institutional payment rails via Citi Token Services.
Europe trails. Santander and Société Générale around 35%. Japan's SMBC and Royal Bank of Canada down at 13%.
And separately, BNY Mellon is pursuing regulated Bitcoin and Ethereum custody for UAE clients through Abu Dhabi Global Market, partnered with Finstreet and the ADI Foundation. That's the direction — jurisdictions with clear frameworks pulling institutional custody offshore, or at least offshore-plus.
Worth flagging the obvious conflict: Strategy publishes this index while holding 843,000 BTC. They benefit directly from every bank moving up the scoreboard. But the methodology is transparent, the scoring is public, and even accounting for the bias, 32% average adoption is a real number. Bank-level Bitcoin integration is happening — just slower than the ETF flow numbers might suggest.
One prediction to close on: the Bitcoin story of the next twelve months isn't the price. It's whether Strategy's flywheel breaks permanently, and what fills the gap when the largest single buyer of the last four years spends more weeks holding cash than stacking sats.