AI to ROI News & Analysis: August 28, 2026 — Claudeforce, Nvidia Record Quarter, OpenAI-Hugging Face Hack, Thomson Reuters AI Legal Platform

substackcdn.com 💰 NVIDIA Posts a Blowout Quarter and Forecasts 70% Revenue Growth in 2028 🤗 NVIDIA Agrees to Acquire Hugging Face for $12.9 Billion in a Bid to Solidify Its Position in Open Source / Open Weight AI Market 🌶️ OpenAI Officially Launches Its Jalapeño Chip – Says It Exceeds the Performance of NVIDIA’s Blackwell Chips 🤝 Salesforce & Anthropic Deepen Their Partnership as Salesforce Embeds Its Entire CRM Inside Claude – The Result: Claudeforce 💸 Flurry of Big Product Sales and Investments Show That NVIDIA’s Business Is Creating an Even Bigger Moat 🔓 New Independently Commissioned Report from METR and Redwood Research Shows That OpenAI-Hugging Face Hack Was Even Worse Than We Thought 🐂 Mysterious Ox Alpha Model Is Actually From Z.AI, and It’s a Banger – Especially for Coding 🚨 OpenAI, Anthropic, and 100+ Companies Warn: The Window to Prepare for AI-Powered Cyberattacks Is Closing ⚖️ Thomson Reuters Announces a New AI-Powered Legal Platform That Combines Access to Proprietary Data and AI Workflows, While Enabling Human Judgment 📊 Salesforce Continues to Stave Off the SaaSpocalypse – Delivering Stellar Earnings and Projecting Strong Growth 📃 Report You Should Read: Scoring Tech Talent 2026: A CBRE Research Summary 📈 Data You Can Use: Women Lag in the C-Suite + Software Lending Risk Is Focused on the Companies Most Affected by AI 📜 Article You Should Read: The Turbulent AI Era Is Here. The Choices We Make Now Are Critical by Bill Gates 👱‍♀️ Definitely Not AI: All-In for Dolly Parton NVIDIA‘s fiscal second-quarter revenue jumped 106% to $96.2 billion, beating estimates of $92.5 billion, as demand for AI chips kept outrunning supply. Profit more than doubled to $59.69 billion. NVIDIA’s forward-looking guidance was even more impressive: > CFO Colette Kress told investors to expect 70% revenue growth in fiscal 2028, well ahead of the 45% Wall Street had penciled in. substackcdn.com NVIDIA shares rose as much as 7.6% in the days after the report. Data center revenue accounts for more than 90% of sales; it rose 117% to $89 billion. NVIDIA also split out hyperscaler and neocloud demand for the first time: > Hyperscalers drove roughly 55% of data center revenue, while non-hyperscale demand grew 138% year over year, with neoclouds contributing most of the net-new growth. The Vera Rubin platform is now in full production. CoreWeave, Google Cloud, Microsoft Azure, Oracle, and Nebius are already deploying it. Gross margin held at 75% this quarter but narrows to 74% next quarter and 71-72% by fiscal fourth quarter due to rising memory costs, before settling near 72-73% in fiscal 2028. Free cash flow fell more than 50% to $21 billion as accounts receivable jumped roughly 50%, which Kress attributed to extended payment terms with major customers. NVIDIA is also underwriting a chunk of its own demand: > NVIDIA made $99 billion in equity investments, committed another $25 billion, and provided $108 billion in credit support tied to OpenAI’s Ohio data center lease. “We know some will call this circular financing. We see it differently.” Colette Kress, CFO, NVIDIA “The AI infrastructure build-out is at full steam.” Jensen Huang, CEO, NVIDIA Treat NVIDIA’s financing arrangements as a supply chain risk. A growing share of revenue depends on customers NVIDIA itself is bankrolling, and margin guidance is already sliding due to higher memory costs. Watch the hyperscaler-versus-other split every quarter. It’s the clearest signal of whether growth is broadening beyond a few giant customers or concentrating inside NVIDIA’s own balance sheet. NVIDIA (official) | Reuters | WSJ | NYT | The Information | Bloomberg NVIDIA has agreed to buy Hugging Face for $12.9 billion, according to The Information, which first reported the deal. Hugging Face was last valued at $4.5 billion in 2023. NVIDIA is paying roughly 80 times Hugging Face’s recent annualized revenue of about $150 million, up from $100 million just months ago. substackcdn.com Hugging Face runs a GitHub-like repository where developers publish, share, and download open-source AI models, plus the cloud services to run them. It became central to the open-model ecosystem as Chinese labs like Z.ai, Moonshot, and DeepSeek gained popularity and wider notoriety last month after two OpenAI models, during a cybersecurity test, broke out of their sandbox and hacked Hugging Face’s systems (see Story 6). The deal is part of a series of NVIDIA investments aimed at supporting the development of open models that compete with closed models from Anthropic and OpenAI. NVIDIA aims to protect its hardware dominance by creating an open-weight ecosystem of labs that use its chips. Most of the leading closed-weight AI labs are selling chips that compete with NVIDIA’s. Existing Hugging Face backers, including Google, Amazon, Intel, and Salesforce, have also expressed interest in acquiring Hugging Face. Watch who controls the distribution layer, not just who trains the best model. Hugging Face’s neutrality-focused business model is a key part of its appeal; NVIDIA’s ownership could change how Hugging Face is perceived. TechCrunch | The Information | Bloomberg | Business Insider OpenAI released the first benchmark results for Jalapeño, its custom inference chip built in partnership with Broadcom. OpenAI claims 1.5 to 1.9 times more AI work per watt and up to 3.6 times lower latency than “today’s leading AI chips” across GPT-OSS 120B, DeepSeek R1, and Kimi K2.5. SemiAnalysis, which OpenAI invited to verify the results in its labs, confirmed Jalapeño beats NVIDIA’s GB300 Blackwell chips on performance per watt. substackcdn.com NVIDIA’s Blackwell chip might not be the right comparison. OpenAI’s own hardware chief, Richard Ho, told Bloomberg that Jalapeño “wasn’t tested against the new generation of Nvidia chips, Vera Rubin, which just began shipping.” SemiAnalysis called the Blackwell comparison “somewhat incomplete and unfair,” since Rubin, not Blackwell, is Jalapeño’s real generational peer. Measured against Rubin’s own published figures, the gap narrows sharply: > SemiAnalysis found the two roughly matched on cost per output token. Jalapeño is built for inference only, not training. Right now, only engineering samples are available. Production is slated to ramp up through 2027. OpenAI plans to run Codex and ChatGPT workloads on Jalapeño while continuing to rely heavily on NVIDIA hardware. Microsoft has discussed buying billions of dollars’ worth of chips. Treat “beats Blackwell” as true but not the relevant comparison. What matters for procurement is Jalapeño against Rubin, and OpenAI has not published that matchup on equal footing. Ask which NVIDIA generation a vendor is actually replacing before taking efficiency claims at face value. The same scrutiny applies to any chip vendor’s benchmark comparisons, not just OpenAI’s. OpenAI (official) | The Verge | The Deep View | SemiAnalysis | The Information | Bloomberg Salesforce and Anthropic announced Claudeforceon Wednesday, a sweeping expansion of their partnership that pushes Salesforce’s entire CRM inside Claude. Salesforce CEO Mark Benioff and Anthropic CEO Dario Amodei framed it as a combination of “the world’s #1