Subscribe Forward this edition

The Agentic Enterprise.

AK · Monday, November 16, 2026 · 8 min read

Monday, November 16, 2026

The fight over who runs Nvidia

Investors, tech executives and a petition with more than 175,000 names want Jensen Huang back as Nvidia's CEO, days after he resigned and a day before Nvidia reports earnings.

Most enterprise buyers never think about who runs their chip supplier. This week they might. Huang's backers argue that much of Nvidia's business was done on his word. A board investigation is under way and the Treasury Secretary wants an outsider. Interim CEO Colette Kress is expected to pull back the financing Nvidia extended to the AI industry. The outcome will shape what Nvidia sells and on what terms.

The Big Story

Governance

Nvidia's handshake problem

A

llies of Jensen Huang are campaigning to put him back in the CEO's chair he left on November 12, after reports of two affairs, one with a former executive of Megaspeed, a Singapore company US investigators suspect of moving Nvidia chips to China. An open letter asking the board to reinstate him while an independent investigation runs has more than 175,000 signatures, including Elon Musk and Tim Sweeney, though its organizers say they can't verify who signed. Altimeter's Brad Gerstner and Atreides' Gavin Baker urged the board to keep "all options on the table, including the potential return of Jensen Huang." The FT reports Huang has discussed a return with confidants. BBC via Yahoo · TechCrunch · PC Gamer

Washington is split. President Trump said he "would certainly not mind" Huang getting his job back; Treasury Secretary Scott Bessent wants Nvidia to hire from outside. The board has hired Wachtell, Lipton, Rosen & Katz to investigate, Huang has agreed to testify before Congress, and the BBC understands reinstatement is unlikely while the inquiries continue. BBC via Yahoo

Musk's argument matters most to buyers. Many Nvidia deals, he said, "were a handshake with Jensen, not a written contract." If that's even partly true, some of the industry's supply commitments rest on one man's word. His interim successor, longtime finance chief Colette Kress, is already expected to rein in the financing Nvidia extended to customers, the Wall Street Journal reported. Nvidia joined 129 venture deals in the past year, and its latest filing lists $36 billion in off-balance-sheet AI cloud agreements. The Next Web

Whoever the board picks will also set how much of Nvidia's selling runs on paper. Kress takes her first earnings call as interim CEO on Tuesday.

On paper, Nvidia's commitments to you may be smaller than you think.

The Spearhead Take

List every Nvidia commitment your AI roadmap depends on, whether you buy direct or through a cloud: capacity, allocation, pricing, co-investment. Get anything that isn't in a signed contract onto paper this quarter, whoever ends up running the company, and keep a second accelerator path for the workloads you can't afford to stall.

Sources: BBC via Yahoo · PC Gamer · The Next Web · TechCrunch

The Obvious & The Overlooked

What is easy to see, and what is easy to miss.

The Obvious

This is a market-test week for AI. Nvidia's earnings Tuesday and Anthropic's IPO pricing Wednesday will set the tone for AI stocks into year-end. StockAnalysis

Defenders are getting agents too. Exaforce's new ExaU lets any security staff member describe a job in plain language and get a personal agent team, with approval required before actions that touch people or systems. Exaforce has raised $200 million since 2023. SiliconANGLE

Central bankers now list AI beside sovereign debt. ECB Vice President Boris Vujcic put the financing need of major US tech firms at $5.5 trillion through 2030. Reuters via AOL

The Overlooked

Coding-assistant settings are an execution surface. GTIG says a suspected North Korean cluster edited .claude/settings.json in internal repositories so Claude Code fetched a backdoor at launch. Google Cloud

AI noise and AI fraud both cost defenders. After a PHP fix on October 21, hundreds of non-working, likely AI-written proofs of concept appeared, adding noise to defenders' triage. GTIG also saw single-day fraudulent charges above $100,000 in LLMJacking cases. Google Cloud

Agent actions are logged under human names. In Palo Alto Networks' survey of 2,800 security executives and practitioners, 53% record agent activity under a developer's identity or personal access token. SiliconANGLE

Moving Pieces

Google catches an exploit bot, AT&T makes agents earn their freedom, Alibaba prices a long-context flagship, Washington targets rented chips, and the Fed reads your agent's permissions.

Security

Google catches a machine reading the release notes

Google's threat team found a framework, exploit_agent, on an exposed staging server. It compares popular web applications' code before and after each patch, drafts an exploit and runs it without anyone approving a step. GTIG calls it the first fully automated discovery and exploitation it has seen from attackers. Disclosures and exploits nearly doubled from January to August, and after two PaperCut flaws in August, researchers had working exploits within a day. The patch itself now points at the hole. Plan to patch internet-facing apps within days.

Sources: Google Cloud · The Register

Deployment

AT&T makes its agents earn autonomy

AT&T runs nearly 500 agentic workflows in production, handling more than 1 million transactions a month and about 45 billion tokens a day, and plans to roughly triple the workflow count next year. Every workflow starts under human oversight. An agent runs alone only after its accuracy reaches the mid-to-high 90s, with drift monitoring after that. Customer-facing and business-critical work goes through a central Business Agent Foundry and its security, privacy and compliance reviews. The figures are company-reported. The accuracy gate is the part worth copying, since it's a policy an auditor can read.

Sources: SiliconANGLE

Models

Alibaba prices a flagship for long documents

Alibaba released Qwen4-Max, a closed model with a 984,000-token context window, at $1.50 per million input tokens and $4.50 per million output. Artificial Analysis scores it 43 on its Intelligence Index, 38th of the 81 models it ranks, so it isn't a frontier leader on that measure. Procurement and data-residency rules will keep it out of most US production stacks, though its price card sets the floor US vendors get compared against when long-context work goes to tender.

Sources: Artificial Analysis

Compute

Washington goes after rented chips

The administration issued rules Monday requiring export licenses for advanced chips sent to data centers in Singapore, Thailand, Malaysia, Indonesia and Brazil, the BBC reports, and Commerce said the rule keeps existing restrictions from being circumvented through remote computing at third-country data centers, per Reuters. It lands days after the Nvidia affair; TechCrunch describes Megaspeed as a Singapore data center company US investigators suspect of funneling Nvidia chips to China. Detailed rule text is still thin. If you rent GPU capacity in those countries, expect customer-verification questions, and check who carries that compliance cost.

Sources: Reuters via Yahoo · BBC via Yahoo · TechCrunch

Policy

The Fed reads your agent's permissions

Fed Governor Michael Barr spent part of a Monday speech in Milan on agents acting inside customer accounts. He warned they could take unauthorized actions and "may accelerate deposit outflows during stress periods." His guardrails are specific: strict authentication, explicit consent for specific actions, and technical limits on what agents can do. He also flagged circular financing among AI firms and their off-balance-sheet obligations. If you build agents that move money, treat those three guardrails as a preview of the exam.

Sources: Federal Reserve

On the Radar

Eight signals worth tracking, one line each.

Deals

Dreamteam raised $53 million in combined seed and Series A funding led by Accel for an AI-native CRM whose agents research accounts, prepare meetings and watch pipeline risk. SiliconANGLE

Security

Palo Alto Networks found 40% of organizations can't reconstruct what an AI agent did, while 73% let agents change code, pipelines or production. SiliconANGLE

Security

Detectify's Agentic Tooling lets a company's own agents test whether suspected flaws are exploitable in live apps, with a vendor-written ledger recording each request and result. SiliconANGLE

Health

Apple opened a US study app with partner OpenEvidence to learn how people use AI for health questions, aiming for up to 200,000 participants over more than a year. TechCrunch

Deals

Blue J raised a $176 million Series E led by JMI Equity, valuing the AI tax-research firm above $700 million with about 6,500 customers. BetaKit

Data

Databento raised $100 million at a $1.1 billion valuation and says it has enterprise agreements with five of the six largest frontier AI labs. FinTech Global

Policy

The administration plans to name Nvidia and more than two dozen AI companies "national champions," the WSJ reported, as the Washington Post estimated from reported trades that Trump held $13 million to $62 million in Nvidia stock at the end of August, a figure Reuters couldn't verify. Reuters via Yahoo Finance

Health

Charta Health raised a $55 million Series B led by Insight Partners for chart review that serves providers and payers from one AI engine. The SaaS News

Quick Hits

The wider field, one line each. Funding figures and product claims are the companies' own.

Agents for specific industries

01Numa raised a $45 million Series C led by Blue Cloud Ventures for customer-service agents at more than 1,500 car dealerships. Fortune
02Lumbra raised a $25 million Series A co-led by Point72 Ventures and Ansa Capital for an agent platform built for defense work. Tectonic
03Nidus has raised $30 million across a Costanoa-led pre-seed and a Kleiner Perkins-led seed for robotics AI in defense manufacturing, starting with wire harnesses. Upstarts Media

Security, compute and chips

04Bay emerged with $50 million, including a $37 million Series A led by Battery Ventures, to evaluate each proposed agent action and allow, refer, redirect or block it before it runs. SecurityWeek
05Mount Thor raised $45 million, with CRV and Conviction co-leading its Series A, to rent Apple-silicon Macs to AI agents by the day. 9to5Mac
06Tendrils raised $30 million led by Benchmark for a many-core processor built on interaction nets; no hardware benchmarks yet. Tech.eu
07US AI data centers deliver about 2.5 times more AI compute per unit of power than Huawei's Horinger site, by Epoch AI's estimate.Epoch AI

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.

Plan a Bootcamp for your team

The Number

18.4 days

The average time for a code security fix to reach production.

That's from Palo Alto Networks' survey of 2,800 security executives and practitioners in 10 countries, and it sits next to the PaperCut exploits researchers built within a day. Business pressure to ship (39%) and unclear risk-acceptance rules (30%) were the top reasons fixes waited. SiliconANGLE

Counter-Signal

Governance

Nvidia's board never named a successor

Whoever runs Nvidia next answers to a board built around its founder. Nvidia has 11 directors; Huang and two others have served since the company's founding in 1993, and every director but Huang is classed as independent. As of last week it had no board chair and had not publicly identified a successor; the BBC now refers to a new chair, Harvey Jones. Santa Clara University's Jo-Ellen Pozner called it a "captive" board and said of its members, "They've become billionaires because of their affiliation and ownership of this company."

A $3.87 trillion supplier without a public successor is a key-person risk inside your own AI roadmap, whichever way the CEO question goes.

Sources: Fortune · BBC via Yahoo

From the Field

Every enterprise AI program we review has a few relationships that run on one person.

Usually they sit on the vendor side. An account executive found you capacity during the shortage, or a partner manager waived a minimum commitment that never came off the order form. Those favors are real, and they're often why a project shipped on time. They also tend to leave with the person who granted them.

Nvidia's week is that pattern at the scale of an industry. Whether or not Huang returns, whoever runs the company is likely to honor what's written down and look harder at the rest. Any new management team would. It's also the moment a buyer learns which commitments were contracts and which were goodwill.

Most of that goodwill can be turned into paper at little cost, usually a side letter or an amended order form signed by someone with authority. It's far easier while the people who offered it still answer the phone.

Let's create value with AI,
AK

The Agentic Enterprise

Know more about AI than 95% of your peers. Every morning.

Anthropic, OpenAI, Google, Microsoft/Azure, AWS, NVIDIA, Databricks and Snowflake are Spearhead technology partners; this edition covers several of them, including the fight over Nvidia’s leadership, Google’s own threat research and misuse of an Anthropic coding tool, and applies the same scrutiny to each. This edition was produced using Anthropic’s Claude with editorial direction from AK. Sources are linked inline.

© 2026 Spearhead. All rights reserved.