Section 1 — The Temperature

Forty-eight hours after the GPT-5.6/Grok 4.5 double launch, the price war stopped being rhetorical and became policy — inside the leaders’ own companies. On Friday, Elon Musk, CEO of Tesla (TSLA — Google · Fidelity) and SpaceX (SPCX — Google · Fidelity), sent Tesla staff a memo telling them to switch to Grok “when possible,” days after Tesla capped employee spending on third-party AI tools at $200 a week — a cap that applies to Anthropic, OpenAI, and Google models but pointedly exempts Grok (Electrek). This is the same Musk who conceded on X this week that Fable is better than Grok 4.5; the memo converts that admission into strategy — if the product can’t win on merit yet, win on captive distribution and $0.13-per-task economics. His public posture stayed sunny: after a mathematician posted that Grok 4.5 had constructed a genuine counterexample in his own research area, Musk replied “Grok only gets better from here” (@elonmusk, July 10).

The second signal came from the market’s scoreboard rather than anyone’s benchmark chart. SK Hynix priced the largest U.S. listing ever by a foreign company and jumped 13% in its Nasdaq debut Friday, closing at $168.01 after raising $26.5 billion (CNBC). Chairman Chey Tae-won’s framing was the purest demand statement of the week: even after announcing a doubling of capacity within five years, he said customers told him “that’s not enough, man.” Meanwhile Meta shares finished their best week since early 2024, up 15%, on the Muse Spark 1.1 launch and cloud-rental musings (CNBC). Capital is rewarding memory, infrastructure, and distribution this week — not model launches. Notably, Alphabet gets no such credit while Gemini 3.5 Pro sits in preview entering the second week of July.

Third: the roster has physically converged on one lodge in Idaho, and the quiet on X is the tell. Sun Valley’s “summer camp for billionaires” is running through the weekend with Sam Altman, CEO of OpenAI (private; IPO reportedly filed), Tim Cook, CEO of Apple (AAPL — Google · Fidelity), alongside incoming CEO John Ternus, Mark Zuckerberg, CEO of Meta (META — Google · Fidelity), Sundar Pichai, CEO of Alphabet & Google (GOOGL — Google · Fidelity), and Jeff Bezos all photographed on site; Dario Amodei, CEO of Anthropic (private; IPO filed), and OpenAI president Greg Brockman are on the guest list (Fortune). Everything there is off the record, which means the next two weeks of “sudden” partnership and M&A headlines were likely drafted this weekend over hikes. For a voice-first brief, the most important conversations of the cycle are currently the ones nobody is posting.

Finally, a crack opened in the biggest closed market. Beijing plans to let Alibaba, ByteDance, and DeepSeek buy a limited number of Nvidia H200 chips — reportedly under 200,000 units total, with each company required to justify quantity and use case (Japan Times, via The Information). Set that against Zuckerberg pricing Meta’s API at roughly a quarter of rivals’ rates and Palo Alto Networks’ CEO arguing AI pricing needs to fall 90% as token costs soar (CNBC), and the through-line of the week is unmistakable: every leader is now optimizing for cost-per-task and access, not peak capability. The contradiction to watch: the same leaders selling “good enough, cheaper” externally — Musk above all — are the ones mandating their own tools internally precisely because employees, left to choose, keep picking the more capable ones.

Section 2 — By Company

Microsoft. Satya Nadella, Chairman & CEO of Microsoft (MSFT — Google · Fidelity): Quiet on X again; his “social permission” framing keeps compounding in the commentary layer, with a July 10 Inc. essay building a leadership playbook around his warning that AI companies must earn societal trust and reorganize rather than eliminate work (Inc.). Mustafa Suleyman, CEO of Microsoft AI (@mustafasuleyman), and Kevin Scott, CTO (@kevin_scott): quiet. Silence flag: this is the second consecutive edition with no meaningful public signal from Scott — and Jeff Dean on the Google side — so the two-week silence watch is now formally on for both. Context worth holding: Nadella’s Build-era “agent-native stack” thesis and the Copilot super app (with Autopilot and the Scout agent) are due this summer (Fortune), which makes Microsoft the quietest company on the roster with the most product exposure to the next sixty days.

Google / DeepMind. Sundar Pichai: no posts, but photographed among the Sun Valley arrivals — his most consequential audience this week is off the record. The Gemini 3.5 Pro situation is now a discipline story: the model remains in limited Vertex AI preview entering the second week of July, with reporting pointing to token-efficiency concerns, coding performance below flagship standard, and long-horizon reasoning short of the I/O bar as the reasons for the slip (MarketScale). The widely circulated July 17 date remains unconfirmed by Google — no model card, no API docs — so treat it as a rumor with a countdown attached. Demis Hassabis, CEO of Google DeepMind (@demishassabis): quiet on X; his “biggest and broadest research bench” defense from the June exodus still stands as the company line while Gemini 3.5 Flash quietly carries the load in production. Alphabet bets note: no significant bet-level news in the window; the parent’s narrative remains hostage to the Pro launch.

OpenAI (private; IPO reportedly filed). Sam Altman (@sama) spent launch week’s aftermath in Idaho, photographed at Sun Valley on July 7 — no major posts since “happy building.” The GA reception is the story: testers are converging on “GPT-5.6 for reliability, Fable for raw intelligence,” with investor Matt Shumer writing that Fable was better for almost every task he tried, while T3 Chat’s CEO called Sol world-leading in computer use (Axios). Two under-covered ships in the same window: GPT-Live, a voice generation that listens and speaks simultaneously (CNBC), and a second acquisition by OpenAI’s Deployment Company — applied-AI firm Northslope — extending the $4 billion bet that forward-deployed engineers, not benchmarks, win enterprise (MarketingProfs). Divergence flag stands: Altman’s 54% agentic-efficiency claim versus Anthropic remains OpenAI’s own launch-day figure, still without independent verification.

Anthropic (private; IPO filed — no ticker yet). Dario Amodei (@DarioAmodei): quiet on X; he’s on the Sun Valley guest list though not yet photographed on the ground (Fortune). The hard news is up-funnel: Micron (MU — Google · Fidelity) disclosed a multi-year partnership making it the first-choice memory and storage supplier for Anthropic’s next AI systems, co-developing HBM to match Claude workloads, and taking a strategic stake in Anthropic’s latest financing (TradingKey) — read alongside the Samsung 2nm talks, Anthropic is methodically securing its own silicon-to-memory supply chain. On the research voice: the company’s July 6 interpretability work describing a readable “global workspace” inside Claude (the “J-space”) is the week’s most substantive lab publication (NeuralBuddies recap). The post-restoration billing shift on Fable 5 — usage credits beyond plan allowances since July 7 — continues drawing developer grumbling even as Anthropic still captures over half of enterprise token spend on major routing platforms (TechCrunch).

Tesla / SpaceXAI. Musk’s Friday memo is the item of record: Tesla staff should move to Grok “when possible” given lower token costs, with engineers asked to email him feedback directly — issued days after the $200 weekly cap on rival AI tools that exempts Grok (Electrek, via The Information). The tension he created himself: “In fairness, Fable is definitely better than Grok 4.5, but most tasks don’t require Fable-level capability” (TheStreet) — a candid pitch externally, but internally it means steering Tesla’s engineers, who reportedly kept choosing Claude, onto the weaker tool because the CEO owns the company that makes it. Critics are already calling it self-dealing between two companies with different shareholders. On capability, his July 10 post amplifying a mathematician’s report that Grok 4.5 produced a legitimate counterexample to a hypercontractivity result — “Grok only gets better from here” — is a genuine, externally-sourced signal, though one result is not a trend. The weekly-gains promise for Grok Build and the monthly foundation-model cadence both remain forward-looking claims, not shipped facts.

Apple. Tim Cook: quiet on X, but present where it counts — he and successor John Ternus are attending Sun Valley together, a deliberate continuity display two months before the September handoff (MacDailyNews). No new Siri AI news in the window; the standing picture holds — developer testing now, public beta next month, fall consumer release, EU and China excluded pending Cook’s regulatory diplomacy with Brussels. The quiet itself is informative: Apple’s AI voice this cycle flows through deals (the Gemini-based foundation models, the Broadcom extension below) rather than posts.

Nvidia (NVDA — Google · Fidelity). Jensen Huang, Founder & CEO (no personal X account): no fresh public remarks, but two developments move his book. First, Beijing plans to let Alibaba, ByteDance, and DeepSeek buy limited H200s — under 200,000 units reportedly, each purchase individually justified to regulators — reopening, barely, a market Nvidia’s own 10-K described as effectively foreclosed (Japan Times). Second, Samsung chairman Lee Jae-yong is set to visit the U.S. this month to meet Huang on Korean semiconductor and AI investment — their first formal meeting in nine months (GuruFocus). And the quiet tax on everyone else’s news continues: SK Hynix’s record listing is, at bottom, a bet on HBM feeding Nvidia accelerators.

Section 3 — Funnel Watch

↑ Up-Funnel — Silicon & Equipment

– SK Hynix (SKHY — Google · Fidelity) — no longer (foreign-listed): the company’s ADRs debuted on Nasdaq Friday, +13% to $168.01 on a $26.5B raise, the largest U.S. listing ever by a foreign company; ticker converts from SKHYV to SKHY on Monday (CNBC). Chairman Chey Tae-won: customers say doubled capacity is still “not enough.” Second consecutive edition of relevance — and now directly investable, links live.

– TSMC (TSM — Google · Fidelity) — holding: earnings land Thursday, July 16, with the Street looking for roughly $39.8B in revenue, up from $30.1B a year ago (Benzinga, prior coverage). Second consecutive edition of relevance. Every launch covered above was fabbed here.

– Broadcom (AVGO — Google · Fidelity) — holding: after Jalapeño with OpenAI, Broadcom extended its multi-billion-dollar wireless chip supply agreement with Apple through 2031 (Distill briefing) — now supplying custom silicon programs on both sides of the model war. Second consecutive edition.

– Micron (MU — Google · Fidelity) — (ENTERED: Anthropic partnership) first-choice memory supplier to Anthropic with co-developed HBM and a strategic investment in Anthropic’s latest round; 16 non-cancelable contracts worth $22B+ in committed HBM revenue, sold out through 2026 (TradingKey). The SKHY listing gives U.S. investors a rival HBM proxy the same week.

– Samsung (foreign-listed) — holding: chip-division profit this cycle reportedly beat its prior 40 years combined on memory pricing (Distill briefing), and chairman Lee Jae-yong meets Jensen Huang in the U.S. this month. The Anthropic 2nm talks remain the swing item.

↑ Up-Funnel — Compute & Infrastructure

– AMD (AMD — Google · Fidelity) — (ENTERED: infrastructure alliance) shares jumped after landing a major AI infrastructure alliance this week (Distill briefing); with China cracking open for H200s and custom ASICs proliferating, Lisa Su’s window as the credible second source keeps widening.

– CoreWeave (CRWV — Google · Fidelity) — (EXITED: news flow ended) second consecutive quiet window; the Meta-rental two-way watch stands, but with no fresh items it rotates off until the story moves.

↓ Down-Funnel — Model Rivals

– Meta AI — Muse Spark 1.1 shipped Thursday with a paid developer tier priced at roughly 25% of competing models — “The pricing from some of the other labs is very extreme,” Zuckerberg told Bloomberg (Bloomberg via Yahoo) — plus Muse Image on Tuesday; the market answered with Meta’s best week since early 2024. The whiplash is the story: nine days after telling employees agentic progress disappointed, Zuckerberg is selling the same stack to developers at a discount.

– DeepSeek (private) — dual catalysts: named among the firms Beijing will allow to buy H200s, and V4’s reported July 17 launch still sets up the head-to-head with Gemini 3.5 Pro’s rumored date. Second consecutive edition; three more and the promotion rule triggers.

↓ Down-Funnel — Tooling & Applications

– Qualcomm (QCOM — Google · Fidelity) — (ENTERED: agentic-stack M&A) reportedly in talks to acquire AI software firm Modular for $4B (Distill briefing) — a chip company buying the software layer, the mirror image of SpaceX buying Cursor.

– Palantir (PLTR — Google · Fidelity) — holding: Alex Karp photographed at Sun Valley alongside the model-lab CEOs (Thought Catalog); the application layer now sits at the main table.

– Cursor / Anysphere (private; SpaceX acquisition pending, expected Q3) — no new items beyond Grok 4.5 integration; watch status pending deal close.

Bench heat: SK Hynix and TSMC are locked in through next week — SKHY’s permanent ticker Monday, TSMC earnings Thursday, and the rumored Gemini 3.5 Pro / DeepSeek V4 collision Friday make July 13–17 the densest calendar of the month. Micron enters hot on the Anthropic tie-up; Qualcomm enters on the Modular talks; CoreWeave exits until the Meta cloud question resolves. Promotion-rule ledger (edition two of the current run): TSMC, Broadcom, Samsung, and DeepSeek at two consecutive editions; the silence flag is now active on Kevin Scott and Jeff Dean after a second signal-free cycle.