The Agentic Enterprise. AK · Daily Edition · 9 min read | Thursday, November 12, 2026 · Daily Edition Nvidia loses the only CEO it ever hadJensen Huang resigned as Nvidia's CEO and director on Thursday, effective immediately, and CFO Colette Kress is running the company while the board looks for a successor. For 33 years the most important supplier in enterprise AI had one boss, and he ran it through dozens of direct reports. Nvidia's board says customer commitments and the product roadmap will continue without interruption. Chips already in the pipeline won't notice. What changes is how the hard calls get made, starting with who gets scarce supply and at what price. | The Big Story Leadership Nvidia's board takes the wheel five days before earnings| N | vidia's board announced Thursday that Huang had resigned as president, CEO and board member. Kress, CFO since 2013, becomes interim CEO and keeps the finance job. Harvey C. Jones, a former Synopsys CEO, becomes the company's first-ever board chair. The search will consider internal and external candidates. NVIDIA |
It moved fast. On Sunday, November 8, the board opened an independent review of reports that a foreign intelligence effort had targeted Huang, and said it had no reason to believe national security or company secrets were compromised. Huang admitted an affair in a staff memo. US officials suspect the woman of ties to Chinese intelligence, according to press reports; China's foreign ministry called the claim unsubstantiated. On Wednesday The Wall Street Journal reported that the board was preparing to remove him, and the stock fell about 8% after hours. NVIDIA board statement · TechCrunch · The Next Web For buyers, the near-term risk is small. Jones says operations, customer commitments and the roadmap continue "without interruption." Most enterprises buy Nvidia through a cloud or a server maker, and those contracts don't name the CEO. Third-quarter results arrive Tuesday, November 17, as scheduled. The longer risk sits in how Nvidia decides. Huang has said he runs about 60 direct reports, TechCrunch notes, and the company had no known succession plan. Allocation during a shortage and early access to new silicon ran through one man. An interim chief who is also the CFO will likely make those calls more by the spreadsheet, at least for a while, which may suit customers fine. Expect to negotiate with a different company. Nvidia's roadmap is the same as it was yesterday. The person who settled its arguments is gone. The Spearhead Take Don't move GPU orders on a headline. Run the review you'd run on any single-source supplier: which workloads could run on a second accelerator path, and what your cloud contracts say about substitution if allocation tightens. |
Sources: NVIDIA · TechCrunch · The Next Web · ABC News | The Obvious & The OverlookedWhat is easy to see, and what is easy to miss. The Obvious Nvidia's governance is now an investor problem. Shares fell about 8% after hours Wednesday on the report of the board's plans. TechCrunch. | Airlines don't want another middleman. United has written a ban on AI agent bookings into its contract of carriage. View from the Wing. | Vendors want to sell agents by the thousand. IBM and Microsoft opened their new platform with more than 6,000 prebuilt agents and assets. Microsoft. |
|
The Overlooked Delta took the opposite bet. It lets AI assistants book when they sign in with the customer's own Delta login, and puts approval on the traveler. Forbes. | Short GPU contracts now pay better than long ones. CoreWeave sells short-term capacity above $40 million per megawatt a year, against $20 million to $25 million on four- and five-year deals. TechCrunch. | Open model hubs may start closing doors. Hugging Face policy chief Irene Solaiman warned attacks could push open platforms to tighten access, which could mean upload limits or ID checks. TechCrunch. |
|
|
| Moving PiecesA booking ban, prepaid GPUs, code no AWS engineer reviewed, a 6,000-agent catalog and a CRM challenger. Commerce United writes AI agents out of its contractUnited revised its contract of carriage on Wednesday to bar AI agents from making, changing or servicing tickets, including through travel agencies, and gave itself the right to cancel what they book. Agents can still show United's flights and fares, per the Wall Street Journal. The airline cites exposed account data, unintended purchases, MileagePlus misuse and check-in rules that need a person to answer hazardous-materials questions. CEO Scott Kirby gave a plainer reason at a Chicago event, the Journal reported: he doesn't want another middleman. The ban lives in a contract; nothing technical stops an agent yet. Every company with a customer login will have to pick a policy, and security belongs in that meeting. Forbes · View from the Wing | Infrastructure CoreWeave would rather have your cash than a bond buyer'sCoreWeave posted third-quarter revenue of $3.66 billion, up 168%, with a $709 million net loss and $846 million in interest for the quarter. Rather than sell unsecured debt, some of which yielded above 15% last month, it expects about $9 billion in customer prepayments and deposits over the next three months. "We will not use the unsecured debt market to fund our business at current yields," CEO Michael Intrator said. If you rent GPUs at scale, expect a request for money up front. Weigh it like a loan you're making to your supplier. TechCrunch | Engineering AWS shipped a cloud emulator its own engineers never reviewedAWS released a free local emulator of its services, built from 438 AWS API models by three teams of agents that write tests, find failures and fix them. Agent reviewers also run the merge gate. "Our team hasn't written or reviewed any of the code," AWS senior principal engineer Matthew Goldberg said. The code is public domain under CC0, and fidelity varies by service. It's a useful test bed, and AWS has put its name behind agent review backed by tens of thousands of fidelity tests. Use it in development. Production still gets your own tests. Amazon · Help Net Security | Platforms IBM and Microsoft open with 6,000 agents and assets, and no priceIBM and Microsoft launched Imagine Enterprise, a platform to build, run and govern industry agents, starting with financial services, insurance and retail. It ships with more than 6,000 prebuilt agents and assets from IBM Consulting, on Microsoft Foundry, Copilot Studio, Fabric, Entra and Purview. IBM can run it for you, build it with you or hand it over. There's no price and no availability date, and the one named client, SEI, is exploring the technology. Count a 6,000-agent catalog as inventory. Before you buy, ask how many of them run in production today, and who maintains each one after the engagement ends. Microsoft | Deals Rox raises $100 million to work around SalesforceRox raised a $100 million Series B at a $1.3 billion valuation, co-led by General Catalyst and March Capital, with Databricks Ventures among the new investors. Its agents find prospects, book meetings, manage deals and flag upsells inside a company's existing systems. Customers include Puma, New Relic and Ramp. Rox says ARR grew more than tenfold in its first year on the market and enterprise net revenue retention tops 700%; neither figure is audited. The CRM fight now runs through the agents working on the data. If sales pilots one, settle first which system stays the record. American Bazaar |
| On the RadarEight signals, one line of context each. Travel OpenAI acquired Odessia, an AI travel agent accredited to issue airline tickets, and shut the service down the same day. Its founder now works on travel at OpenAI. TechCrunch · The Next Web |
Product Databricks updated Genie, its agent for business leaders, to run long multi-step jobs without supervision on top of an ontology built from customer data. Its Unity Gateway can cap usage and swap in cheaper open-weight models. TechCrunch |
Chips Fractile raised a $720 million Series C led by Thrive Capital, Founders Fund, Lightspeed and Redpoint, with Britain's Sovereign AI Fund also in. The inference chip startup plans to ship its first chips and systems next year. TechCrunch |
Compute AMD's EPYC Venice server CPUs top out at 256 cores for $14,904, with SP7 systems due this quarter. AMD's own tests claim up to 2.8x more AI agents per watt than Nvidia's Vera. Digital Citizen · StorageReview |
Energy Torus launched Foundry, modular units that add 1 to 20 megawatts of inference capacity at existing sites, and secured $500 million in funding. It quotes 12 to 16 weeks from contract to compute. Torus |
Security Bay came out of stealth with $50 million for software that checks each action an AI agent proposes and allows it, blocks it or sends it to a person. It names Micron, Peloton, ClickUp and Wix as customers. The Next Web |
Talent Meta's infrastructure chief Santosh Janardhan, there since 2009, is moving to an advisory role on Meta Compute. Dave Brown, who spent 19 years at Amazon, takes over. TechCrunch |
Health Oura agreed to buy Counsel Health, which pairs an AI medical chat with licensed physicians, for $625 million. Counsel's CEO becomes Oura's chief clinical AI officer. TechCrunch |
|
| Quick HitsThe wider field, one line each. Funding | 01 | Hatz AI raised a $24 million Series A led by Nadia Partners for an AI platform that managed service providers resell to small businesses.SiliconANGLE | | 02 | Plurai raised a $13 million seed led by Team8 to simulate where AI agents fail before customers find out.PR Newswire |
|
Infrastructure and research | 03 | AMD says MEXT software, which lets flash stand in for DRAM, saved its own IT team about $2.78 million, or 82%, on an initial 86 servers.AMD | | 04 | Pew found 60% of US adults would be uncomfortable with a new data center near them.Pew Research Center | | 05 | PSI says it used early access to an unreleased Anthropic model to settle an 18-year-old quantum computing conjecture.TechCrunch | | 06 | MIT researchers trained climate emulators on far less data that still predict unseen scenarios better than older emulators.MIT News |
|
Products | 07 | Adobe released a local MCP server that exposes Photoshop, Illustrator, After Effects and Premiere to ChatGPT, Claude and Gemini.TechCrunch | | 08 | Ring's Rover Cam, a $1,000 patrolling security robot for businesses, ships December 16.TechCrunch | | 09 | Reddit Reviews turns community threads into product summaries for about 10,000 products, for all US users.TechCrunch | | 10 | Meta's Muse agent now runs on Apple Watch and Wear OS.TechCrunch |
|
| From Spearhead · AI Bootcamp Get your leadership team AI-ready, in a day.The AI Bootcamp is a one-day, in-person session designed for your industry, your organization, and your use cases. Your leaders understand AI, build AI agents, envision outcomes, and start building a new work culture. |
| The Number -109% OpenAI's reported adjusted operating margin in the third quarter, the first in which its revenue topped $10 billion. Bloomberg reported the third-quarter figures, citing unnamed sources. Revenue rose about 50% from $6.7 billion in the second quarter, a figure first reported by the Wall Street Journal, and the margin improved from -150%. On that measure OpenAI still spends about $2.09 for every dollar it brings in, which is worth remembering the next time a token price looks permanent. TechCrunch | Counter-Signal Security The AI bill that hurt this fall came from a stolen keyRailway, a developer cloud platform, published its full report this week on an October intrusion. An attacker partially escaped its build isolation, chained stolen credentials and decrypted environment variables in at most 1.85% of workspaces, prioritizing AI provider keys. Those keys ran up an estimated $740,000 in unauthorized AI usage. Railway believes they were resold for model distillation, though it hasn't established who ran the resale. Wiz has documented a similar pattern of stolen AI keys being checked and resold across other platforms, though Railway says that work doesn't attribute its own incident. Railway The money AI teams actually lost this fall left through a credential in a config file. Refunds depended on the provider: some issued them under fraud policies, one enforced a no-refund rule, and Railway is covering the gap. Spend caps per key and alerts on usage spikes would have shortened the bill. Sources: Railway | From the Field Most AI programs we see have their own single point of failure. Usually it's the engineer who wrote the first prompts, built the evaluation set and still keeps the provider keys in a personal password manager. Ask why the agent refuses a certain request, and the answer is a Slack message to that person. Early on, that's how anything gets built. One person holds the context, and decisions are quick because nobody else has an opinion yet. The trouble starts when the system reaches production and the context stays behind. The reason for a guardrail lives in one memory. When that person takes a new job, the team inherits a system it can run but can't explain, and the first incident takes days to diagnose. Nvidia at least has a board and a CFO to fall back on. Most AI teams have a README last touched in the spring. In an AI program, succession planning is mostly documentation and a shared vault for the keys, and almost nobody budgets for either. Let's create value with AI,
AK | Work with Spearhead AI Bootcamp One day to get leaders fluent enough to decide. |
|
| The Agentic Enterprise Know more about AI than 95% of your peers. Every morning. NVIDIA, Anthropic, OpenAI, AWS, Microsoft/Azure, Databricks and Snowflake are Spearhead technology partners; IBM, Google, Meta, AMD and CoreWeave are not. Produced using Anthropic’s Claude with editorial direction from AK. © 2026 Spearhead. All rights reserved. |
|
