Pulse AI Briefing

Nvidia’s Jensen Huang told President Trump, live on speakerphone, that the industry will not accept an AI slowdown — hours after Trump called Anthropic’s Dario Amodei a “perfect little angel” and named himself AI’s only guardrail.

The signal

The slowdown fight leaves the labs and lands in the White House

Speaking at the All-In Summit in Los Angeles on Monday, Nvidia chief executive Jensen Huang took a phone call from President Trump on stage and put him on speaker for the room. “I’m onstage with the besties,” Huang said, before telling the President plainly that “we’re not going to let [an AI slowdown] happen” — a direct rebuff of the pacing argument Anthropic’s Dario Amodei has spent a week building. Trump had moved first that morning, calling Amodei a “perfect little angel” on Truth Social and declaring that “the only control or guardrails that AI needs is a STRONG AND SMART (High IQ!) PRESIDENT.” A safety debate that began inside the labs is now a loyalty test run from Washington.

Seven stories

Today's Briefing

Markets · 14 September

Cybersecurity leads as the AI-risk trade replaces chips

The safety warnings that dented chipmakers over the weekend sent a very different signal to security names, on the logic that a more dangerous threat environment expands security spending. CrowdStrike closed 13.85 per cent higher at an all-time high, Zscaler jumped 16.52 per cent, Palo Alto Networks gained 13.09 per cent — its biggest one-day move since April 2025 — and Fortinet added 9.04 per cent. The cybersecurity ETF CIBR rose just 4 per cent, leaving the bid concentrated in pure-plays and reading as sentiment around agent-swarm risk rather than fundamentals.

Source: 24/7 Wall St.

Product · 14 September

Anthropic opens Claude for Financial Advisors

Anthropic launched a suite of AI connectors and “workflow skills” aimed at advisory practices, wiring Claude into Charles Schwab, BlackRock, Addepar, Envestnet, iCapital, Orion, SS&C Black Diamond, Vanguard, Morningstar, S&P Global, FactSet, Salesforce and DocuSign, among others. Skills cover advisor onboarding, alternative-investment briefing, compliance and AI-policy review, estate and tax briefing, portfolio rebalancing and prospect intake. Enterprise pricing runs roughly 70 to 120 US dollars per user per month, with the plug-in itself free and a one-time usage credit for firms licensing before 30 September.

Source: Yahoo Finance

Hardware · 14 September

OpenAI buys smartphone camera maker Glass Imaging for $300M

OpenAI has acquired Glass Imaging, a Los Altos company founded in 2019 by former Apple engineers Ziv Attar and Tom Bishop, who previously led the team behind Apple’s Portrait Mode. Rather than editing photos after the fact, Glass uses neural networks to model individual camera systems and improve images from the moment the shutter fires. The deal, first reported by The Wall Street Journal, values the startup above 300 million US dollars and follows OpenAI’s recent purchases of TBPN and an open-source developer-tools firm as it preps consumer hardware.

Source: TechCrunch

Infrastructure · 14 September

Cornelis raises $205M to break Nvidia’s networking lock-in

Cornelis Networks, spun out of Intel in 2020, closed a 205 million US dollar round led by IAG Capital Partners and unveiled Active Compute Fabric — an open, GPU-agnostic interconnect that lets accelerators process and send data at the same time, targeting the GPU time lost waiting on data to arrive. It simultaneously announced a strategic collaboration with Qualcomm for rack-scale inference. Its CN5000 400 Gbps switch ships now, with the multi-protocol CN6000 800 Gbps due in the fourth quarter.

Source: TechCrunch

Infrastructure · 14 September

SemiAnalysis: Vera Rubin hits 7x Blackwell’s tokens per megawatt

Pre-release testing of Nvidia’s Vera Rubin NVL72 shows 7x better token throughput per megawatt than Blackwell on DeepSeek V4 Pro’s 1.6-trillion-parameter model, well above Jensen Huang’s public 3x claim. At 100 tokens per second of interactivity the setup reaches 59.4 million tokens per second per megawatt against 28.5 million for GB300, which SemiAnalysis models as 149.9 billion US dollars of annual profit per gigawatt versus 105.3 billion for GB300 on Dynamo SGLang. The authors’ verdict: “Jensen needs to stop sandbagging.”

Source: SemiAnalysis

Privacy · 14 September

Project Lily: contractors are reading real ChatGPT prompts

404 Media reports that OpenAI runs an internal programme, codenamed Project Lily, that hires hundreds of contractors to read real ChatGPT users’ prompts — many containing sensitive personal information — and rate the responses on a 1-to-7 scale. Workers are recruited through Crossing Hurdles and paid via Mercor, with one reporting over 50 US dollars an hour. ChatGPT has more than 900 million users and model training is on by default for consumer plans; the report notes that Anthropic and Google run comparable human-review programmes.

Source: 404 Media

Geopolitics · 14 September

Beijing calls Amodei’s essay “fearmongering” before a Trump–Xi summit

China’s Foreign Ministry pushed back on Dario Amodei’s weekend essay, which argued that a Chinese lead in AI would pose “grave danger” and urged continued export controls on the most advanced chips and chipmaking equipment. Spokesperson Guo Jiakun said that “fearmongering, confrontation and vicious competition will only disrupt the process of global AI governance — which serves no one’s interest.” State-run Global Times dismissed the piece as a “Cold War playbook” for AI. Trump and Xi are expected to discuss AI governance at a meeting planned for 24 September.

Source: NPR

Calendar

What to Watch

  • This week

    Z.AI settles a US$5 billion raise

    The Chinese lab’s concurrent financing — a Hong Kong share placement plus zero-coupon convertible bonds — is set to settle on 16 September, with 60 per cent earmarked for next-generation GLM foundation models and training infrastructure.

  • 24 Sep

    Trump–Xi summit

    The two leaders are expected to discuss AI governance at a meeting in Washington, days after Beijing dismissed Amodei’s pacing essay as fearmongering and a top Chinese security official accused “certain nations” of weaponising AI advantages.

  • October

    Anthropic’s Nasdaq listing

    An October debut is targeted after the 1 June confidential filing. Nvidia’s reported anchor commitment and final pricing will set the reference point for every frontier-lab valuation that follows.

  • 13–15 Oct

    TechCrunch Disrupt

    Moscone West, San Francisco. The agent-platform and infrastructure cohort shipped over the past two weeks will be benchmarked here.

  • 20–21 Oct

    AI Expo Europe

    Amsterdam. European enterprise deployment and the sovereign-AI procurement agenda now dominated by Cohere and Mistral.

  • 4–5 Nov

    FT Future of AI Summit

    Regulation, agentic systems and scaling. The first major policy gathering after the White House’s open turn against safety-driven slowdowns.

  • 17 Nov

    Nvidia Q3 results

    The clearest read on whether the infrastructure buildout is still compounding — and on how much of the Vera Rubin performance story the company chooses to tell publicly.

  • 2 Dec

    International AI Summit

    Brussels. Policymakers and industry on global governance and access, with the EU AI Act’s high-risk obligations legally in force since 2 August pending the Digital Omnibus delay.