Section 1 — The Temperature

The weekend belonged to a fight, and the fight tells you more about the state of the industry than any launch this month. On Saturday morning, Elon Musk, CEO of Tesla (TSLA — Google · Fidelity) and SpaceX (SPCX — Google · Fidelity), reshared an old post calling Sam Altman, CEO of OpenAI (private; IPO reportedly filed) “Scam Altman” and added “He takes scamming to a whole new level” (@elonmusk). Altman fired back within minutes: “homeboy you’re the one selling public market investors on short-term space datacenters” (@sama) — a direct shot at the orbital data centers Musk has pitched to SPCX shareholders. Musk’s retort, that the space data centers start flying next year, referenced Altman’s legal troubles (Benzinga). Then Altman turned the feud into product marketing, posting that many benchmarks suggest 5.6 Sol is the best model in the world, “but the most reliable way to tell is that elon is obsessed with me again” (@sama, 814K+ views). Two CEOs of companies courting public investors are now running competitive positioning through a personal grudge — and both know exactly what they’re doing.

Beneath the theater, the substantive front moved to the courtroom. Apple (AAPL — Google · Fidelity) sued OpenAI in the Northern District of California on Friday, alleging a coordinated trade-secret theft campaign directed by OpenAI’s hardware chief Tang Tan — a 24-year Apple veteran — including asking Apple employees to bring “actual parts” to job interviews and coaching departures on evading exit security (TechCrunch). The filing calls OpenAI’s hardware business “rotten to its core.” OpenAI’s response: “We have no interest in other companies’ trade secrets” (CNBC). The pattern of the past month is now unmistakable: the frontier fight has moved from benchmarks to price, from price to Washington, and now from Washington to litigation — Musk’s $150B suit against OpenAI, a dismissed xAI poaching claim, and now the $4.6 trillion iPhone maker versus its own AI partner, months before an expected historic IPO.

The third signal is a walkback that maps the gap between what leaders say and what their companies do. After The Information reported a Friday memo telling Tesla staff to switch to Grok 4.5 “when possible” — days after Tesla capped employee spending on rival AI tools at $200/week while exempting Grok — Musk posted Saturday that “legacy media is misrepresenting the situation”, saying he merely asked Tesla and SpaceX to try Grok and that teams should keep using rival models that outperform it (@elonmusk). The memo and the spending cap are documented; the walkback is the post. Meanwhile the benchmark record he himself conceded last week — Fable ahead of Grok 4.5 — hasn’t moved: on the neutral DeepSWE 1.1 coding benchmark, Grok 4.5 resolves 53% of real GitHub issues versus 70% for Claude Fable 5 (Electrek).

Fourth: the application story delivered its bluntest consent lesson yet. Meta (META — Google · Fidelity) killed its Muse Image Instagram feature Friday — three days after launch — after CAA and SAG-AFTRA erupted over a default opt-in that let anyone generate AI images from public accounts’ photos; Meta’s statement conceded the feature “missed the mark” (Reuters). It’s the fastest retreat of the Superintelligence Labs era, and a reminder that shipping velocity — the thing every leader on this roster is currently selling — has a rights-and-consent speed limit that Washington didn’t even have to enforce.

The week ahead is stacked: TSMC earnings Thursday, July 16; the reported July 17 double-launch of Gemini 3.5 Pro and DeepSeek V4; and a White House voluntary frontier-model framework that FT reporting has placed as imminent, ahead of the executive order’s August 1 deadline (Reuters/FT). Quiet Sunday; loud week coming.

Section 2 — By Company

Microsoft. Satya Nadella, Chairman & CEO of Microsoft (MSFT — Google · Fidelity): Quiet on X through the weekend. The company’s live thread is its role in the five-lab working group (with Anthropic, OpenAI, Google, and Amazon) building a shared jailbreak severity scoring system ahead of the White House framework’s August 1 deadline (TechTimes) — Microsoft positioning itself as infrastructure for the review regime rather than a subject of it. Mustafa Suleyman, CEO of Microsoft AI (@mustafasuleyman), and Kevin Scott, CTO (@kevin_scott): both quiet; the Copilot super app Nadella promised for this summer remains the next scheduled signal.

Google / DeepMind. Sundar Pichai, CEO of Alphabet & Google (GOOGL — Google · Fidelity): Quiet on X. The July 17 Gemini 3.5 Pro date is now widely reported as confirmed — a deliberate ground-up pre-training rebuild rather than a fine-tune of the 2.5 base — and it lands the same day as DeepSeek V4’s official release, a direct East–West flagship collision (AIbase). Google still hasn’t published a model card or date, so hold the specifics loosely. Demis Hassabis, CEO of Google DeepMind (@demishassabis): quiet; the talent-bleed subplot continues, with Gemini contributors Jonas Adler and Alexander Pritzel reportedly poised to follow Jumper and Shazeer out the door — to Anthropic (AI Weekly), even as Google expands its coding strike team’s remit into midtraining to close the gap. Jeff Dean (@JeffDean): quiet. Alphabet bets note: no new signal this window; the EU Android antitrust loss remains the parent’s regulatory drag.

OpenAI (private; IPO reportedly filed). Altman’s weekend was the feud, covered above — but note the substance he attached to it: the claim that 5.6 Sol is the best model in the world is his own benchmark reading, made on X during a flame war, not an independent evaluation. The heavier item is the Apple lawsuit: beyond Tan, the complaint names former Apple engineer Chang Liu, alleging he kept an Apple laptop and downloaded confidential files while at OpenAI (Axios). With the io Products device promised for the first half of 2026 and an IPO in the wings, discovery in this case is now a business risk with a court schedule. Divergence flag: Apple’s allegations are one side of a filing; OpenAI denies them.

Anthropic (private; IPO filed — no ticker yet). Dario Amodei (@DarioAmodei): quiet on X, in keeping with his post-blackout posture. The company’s best marketing this weekend was involuntary: Musk’s standing concession that Fable is better than Grok 4.5, the DeepSWE 1.1 gap (70% vs. 53%), and reporting that Tesla engineers broadly prefer Claude for day-to-day development even as the $200/week cap steers them elsewhere (Electrek). Continuity items: usage-based billing on Fable 5 remains in effect post–July 7 as a capacity measure, and Anthropic’s five-lab jailbreak framework work feeds directly into the August 1 White House deadline. The Samsung custom-chip talks remain open and undecided.

Tesla / SpaceXAI. Musk (@elonmusk) produced the roster’s loudest 24 hours for the third edition running: the Altman feud, the Grok-mandate walkback, and a standing promise of weekly Grok Build gains. The context his posts don’t carry: SPCX has slipped below its $135 IPO price amid heavy multiyear put-option volume (Barchart via Yahoo), meaning the “good enough, cheaper” Grok strategy now has to show up in SpaceXAI revenue with the market watching skeptically. Also in the record: Cursor disclosed that an earlier snapshot of its own codebase leaked into Grok 4.5’s training data, inflating one launch-day coding score before being excluded (Electrek). Grok 4.5’s European rollout is expected mid-July.

Apple. Tim Cook, CEO: quiet on X, as always — but Friday’s lawsuit is the loudest thing Apple has “said” about AI all year, and it reads as a strategic decoupling: the updated Siri shipping this fall is built on Google’s Gemini models, not OpenAI’s, even as ChatGPT remains integrated in Apple Intelligence (CNBC). Apple’s statement frames the suit as protecting unreleased-product IP; it seeks return of materials and a bar on use, not (yet) damages that would rupture the partnership outright. Read against the September handoff to John Ternus, Cook is spending his final quarter as CEO litigating the company’s AI perimeter.

Nvidia. Jensen Huang, Founder & CEO of Nvidia (NVDA — Google · Fidelity) (no personal X account): no fresh public remarks this weekend — and none needed. Every combatant above trains and serves on his silicon, and the week’s quiet infrastructure signal points his way: TSMC is moving silicon photonics toward mass production, the manufacturing base for Nvidia’s Spectrum-X Photonics and Quantum-X networking switches aimed at the copper interconnect wall (Yahoo Finance). The feud economy runs on GB300s either way.

Section 3 — Funnel Watch

↑ Up-Funnel — Silicon & Equipment

– TSMC (TSM — Google · Fidelity) — holding: earnings Thursday, July 16 kick off tech’s reporting season; the Street wants confirmation that Q1’s 41% revenue growth and the $52–$56B capex plan hold, and there’s a new angle — silicon photonics moving to mass production as the answer to AI’s copper bottleneck (Barchart). Third consecutive edition of relevance; two more and the promotion rule triggers.

– SK Hynix (SKHY — Google · Fidelity) — holding: no fresh news in the weekend window; Monday opens the first full Nasdaq trading week after Friday’s record $26.5B debut. Third consecutive edition.

– Samsung (foreign-listed) — holding, quiet weekend: the two live threads — chairman Lee Jae-yong’s planned U.S. meeting with Jensen Huang this month and the undecided Anthropic 2nm talks — both remain open. Third consecutive edition.

– Broadcom (AVGO — Google · Fidelity) — (EXITED: news flow paused) after the Jalapeño unveiling and the Apple wireless extension, no new items this window. Returns when the custom-silicon roadmap produces its next milestone.

↓ Down-Funnel — Model Rivals

– DeepSeek (private) — holding: V4’s official launch is set for July 17 — deliberately or not, Gemini 3.5 Pro’s reported date — with peak-hour pricing at 2x baseline and a hard July 24 retirement of the legacy deepseek-chat/reasoner API aliases that every integrated dev team must migrate around (TechTimes). Third consecutive edition; two more and the promotion rule triggers — and Thursday will likely settle it.

– Z.ai / GLM-5.2 (private) — (ENTERED: analyst validation) Jefferies’ Christopher Wood called GLM-5.2 “almost equal to Anthropic” as a corporate-market competitor at roughly a quarter of the per-token cost (Edgen). The Chinese open-weight bench is now two-deep behind DeepSeek.

– Moonshot / Kimi K3 (private) — (ENTERED: parameter-race escalation) Moonshot confirmed Kimi K3 ships this month at a reported 2.5 trillion parameters — larger than DeepSeek V4-Pro’s 1.6T — crowding an already saturated July launch calendar (AIbase). Reported specs, not yet benchmarked.

– Meta AI (META — Google · Fidelity) — the Muse Image shutdown (covered above) is the down-funnel lesson: Meta’s first Superintelligence Labs image model survived three days of consumer contact. The underlying Muse Image model itself remains available; the Instagram @-mention feature does not (Gizmodo).

↓ Down-Funnel — Tooling & Applications

– Cursor / Anysphere (private; SpaceX acquisition pending, expected Q3) — holding: the training-data contamination footnote (its own codebase in Grok 4.5’s training set, inflating a headline coding score) is a governance preview of what model-lab vertical integration looks like in practice (Electrek). Third consecutive edition of relevance.

Bench heat: TSMC and DeepSeek both hit edition three and both have hard catalysts inside five days — if Thursday’s earnings and V4’s launch land as expected, both reach edition five by mid-week next week and the promotion rule will trigger; start deciding now whether DeepSeek belongs in the core roster. Micron and CoreWeave rotate off after quiet windows; Broadcom exits until the next silicon milestone. On the core roster, Kevin Scott and Jeff Dean log a third consecutive edition without meaningful public signal — the two-week silence flag is approaching for both.