〔Overnight Brief〕Gemini 4 Argon Re-enters the Race at $2/$10 per Million Tokens but Stays Behind a Cybersecurity Gate; FTC Probe and Anthropic-vs-Nvidia Split Reshape the White House Pledge

0. Overnight Arc

The model race and the regulatory perimeter moved in the same 24 hours. Google pushed Gemini 4 Argon to the top of 13 of 18 benchmarks behind a cybersecurity-only gate ^[1], the FTC moved from probe to drafting civil investigative demands against OpenAI and Anthropic ^[2], and an OpenAI-Synopsys deal gave electronic design automation a flagship AI-workflow story ^[3][4]. The White House pledge, meanwhile, surfaced the split that the safety crowd had warned about — Anthropic on one side, Nvidia, Meta and SpaceX on the other ^[5][6].

1. The Model Race

• **[NEW] Google Gemini 4 Argon:** top or tied in 13 of 18 benchmarks, strongest in knowledge work and long-context reasoning, weaker on coding ^[1]. Pricing: $2 per million input tokens, $10 per million output tokens, per CEO Sundar Pichai ^[7][8].
• **[NEW] Gated rollout:** access limited to select cybersecurity partners pending system-patch work; the model includes monitoring that can terminate the run if the model oversteps ^[1]. Gemini senior director and product lead Tulsee Doshi says the safeguards "are effective" ^[1]. (Single-source detail on safeguards ^[1].)
• **[NEW] OpenAI × Synopsys:** multi-year deal to build GPT-Synopsys, an EDA-trained model that runs on Synopsys tool subscriptions with revenue-share tied to chip-design gains ^[3][4]. OpenAI co-founder Greg Brockman says it will cut weeks-to-months from design cycles; Synopsys CEO Sassine Ghazi frames it as incremental, not cannibalizing ^[3]. Synopsys shares rose 7% intraday, closed +4.78% ^[3].
• **[ONGOING] OpenAI reliability:** ChatGPT Space page error rate elevated, mitigation deployed, recovery being monitored ^[9].

2. The Regulatory and Safety Split

• **[ESCALATED] FTC probe:** Chair Andrew Ferguson launched the investigation weeks ago; the agency is drafting Civil Investigative Demands (CIDs), subpoena-equivalent orders targeting Anthropic, OpenAI and other frontier labs, issuance expected in the coming weeks ^[2][10]. An FTC official: the probe "should not in any way hinder" US AI dominance ^[2].
• **[NEW] White House pledge, the Anthropic-vs-Nvidia/Meta rift:** at a Trump meeting with roughly a dozen AI CEOs, Nvidia's Jensen Huang and Meta's Mark Zuckerberg pressed Anthropic CEO Dario Amodei on why he publicly flagged risks of cyberattack and mass unemployment, calling the messaging "overly aggressive" ^[5]. Amodei replied that the industry should describe capabilities and risks honestly, not minimize them ^[5]. A prior plan for a finance-style industry self-regulatory body collapsed after Huang, Zuckerberg and SpaceX's Elon Musk feared it would concentrate power in leading labs ^[5]. Zuckerberg's role in shaping the pledge: ^[6].
• **[NEW] White House AI lead David Sacks:** says many AI firms recently reached consensus on new internal controls and audits ^[11].
• **[NEW] President Donald Trump:** says the US should stop using the term "artificial intelligence" because the technology is not "artificial" but real ^[12].
• **[NEW] Bill Gates (interview, single source):** renews call for a robot tax, proposing a "labor unit" subject to the same FICA payroll tax whether the work is done by a human or a robot ^[13].
• **[NEW]** An OpenAI executive has withdrawn a planned second $25 million donation to an AI super PAC ^[14].

3. Compute, Capital and Cooling

• **[NEW] GPU scarcity, CoreWeave:** CEO says the company does not currently have enough Nvidia GPUs to meet customer demand ^[15]. (Single-source CEO remark ^[15].)
• **[NEW] GMI Cloud:** $668M equity-plus-debt round, with the $223M equity leg led by ARCHIV and Nvidia participating; the company cites $600M+ annualized contracted revenue ^[16].
• **[NEW] LG Electronics:** 150 billion KRW investment in AI data center cooling ^[17].
• **[NEW] ST Doushen subsidiary:** 60-month cloud services contract worth about $119M (≈RMB 801M) to provide GPU compute rental, deployment and operations to a VB customer ^[18].

4. Distribution: Wallets, Travelers, Farmers, Tourists

• **[NEW] Robinhood "Robinhood Agents":** native in-app AI agent for natural-language strategy and 24/7 auto-execution, launched at HOOD Summit, paired with planned 24-hour stock trading, perpetual futures and earnings contracts ^[19]. (Complex product claims, single source ^[19].)
• **[NEW] Meta Muse:** AI travel agent with booking capability; Truist says it poses a greater disintermediation risk to Expedia and Booking than prior LLMs because it can complete reservations, not just plan ^[20].
• **[NEW] Doubao:** named official AI tour guide for Shanxi cultural-tourism sites, covering 100+ attractions including Yungang Grottoes and Jinci Museum ^[21].
• **[NEW] "Shennong Zao":** Swahili-language vertical model for Kenyan smallholders, launched by China Agricultural University associate professor Wang Yaojun; Gates Foundation's Stewart Collis argues generalist models underperform for smallholders ^[22].
• **[ONGOING]** Caixin feature on the open-source vs closed-source split in the large-model industry ^[23].

5. What Would Falsify the Arc

• The cybersecurity-gated Argon rollout is the soft spot: if Google widens access to general developers within days, the safety framing loses credibility and pricing becomes the operative number ^[1]. If the FTC stalls the CIDs past "coming weeks," the regulatory tail risk shrinks to rhetoric ^[2]. CoreWeave's GPU shortage and LG's cooling capex are both single-source corporate statements; treat as directional until confirmed by filings or third-party data ^[17][15].