Tuesday, November 17, 2026 · 5 updates

AI briefing: November 17, 2026

Nvidia's first post-Huang scorecard lands after the close today, Trump and Xi take AI safety talks to the APEC summit tomorrow, Washington prepares to designate two dozen AI 'national champions', Tencent undercuts the model market again, and Amazon says AI agents ran cybercrime campaigns against AI apps.

The report card

1Nvidia reports after the close: the numbers matter less than the call

Nvidia's first quarter without Jensen Huang gets its verdict after the market close today, when interim CEO Colette Kress presents fiscal third-quarter results. Consensus clusters around $109 billion in revenue - Reuters' survey reads $109.08 billion, the Financial Times' reads $109.5 billion - against the company's own $108 billion guide, with adjusted earnings of $2.47 a share expected and data-center revenue around $101.8 billion. One number matters as much as the print: the fourth-quarter consensus already sits at $124.89 billion, so a merely in-line guide could disappoint. Options markets imply a swing of roughly 8% on the report; the stock closed Monday down nearly 19% for November.

In plain English

Almost doubling last year's $57 billion quarter is the baseline expectation - what moves the stock is what Kress says comes next.

Why it matters

The quarter ended October 25, before Huang's November 12 resignation and before Monday's new chip-export rule - so everything investors actually want to know (Vera Rubin shipments, China exposure, governance) lives in the outlook and the call, Kress's debut as the company's voice. Results land less than two days before Anthropic's IPO prices in the same jittery market.

SourcesINDmoney's consensus breakdown·FX Empire on what is priced in

The summit

2Trump and Xi take AI safety talks to APEC, with Monday's chip rule hanging over the table

Trump and Xi will hold AI safety talks on the sidelines of the APEC summit on November 18-19, attempting to turn the AI crisis hotline they agreed to in September into a working channel, AFP reports. The agenda spans infrastructure hacking, AI-assisted bioweapons, military use, mass surveillance and out-of-control agents - one proposed use for the hotline is exactly the scenario of one country's AI agents touching the other's sensitive systems. Monday's Commerce Department rule restricting advanced AI chips in third-country data centers now hangs over the meeting: Bloomberg reports it forces Xi to weigh a response going into the summit.

In plain English

The two countries building the most powerful AI are trying to set up a phone line for AI emergencies - while actively restricting each other's chip supply.

Why it matters

The talks land in the middle of the Nvidia governance crisis that AFP notes is threatening to steal their thunder - a scandal about a suspected Chinese intelligence asset at the top of America's most strategic chip company. Any hotline agreement would be the first institutional AI-safety channel between the two powers.

SourcesAFP on what to expect from the talks·Bloomberg on Xi's dilemma

Washington

3Catch-up: the administration plans to designate two dozen AI 'national champions'

The Trump administration plans to designate more than two dozen AI companies - reportedly including Nvidia, Amazon, Microsoft, OpenAI and Anthropic - as US 'national champions', a status that would bring expedited chip export licenses, support from the US government in overseas markets and access to federal financing, Reuters first reported Sunday. The aim is to help American AI firms win foreign markets against Chinese competitors; the designation would formalize the state backing that has been building all year, from equity stakes to export deals.

In plain English

Washington is picking its AI team - and membership comes with faster licenses, diplomatic muscle and federal money.

Why it matters

Read together with Monday's export rule and this week's Anthropic IPO, the state and the market are both formally anointing the industry's winners in the same week - and the list's boundaries will matter as much as its contents. Reuters reported the plan Sunday; it is included here as catch-up coverage, and the company list has not been officially confirmed.

SourcesQuartz on the designation plan·Yahoo Finance on the reported company list

The model race

4Catch-up: Tencent's Hy4 Flash undercuts the market at seven cents per million tokens

Tencent launched Hy4 Flash on Monday, a lower-cost sibling of its Hy4 model aimed at everyday work - research, slides, spreadsheets - with a 224-billion-parameter backbone activating 12 billion per token, multimodal inputs and a one-million-token context window. List prices are $0.07 per million input tokens and $0.281 per million output, and Tencent says the model cost one-twentieth as much per task as the full Hy4 on its internal WorkIndex benchmark, while a blind test with 85 users rated it near Claude Haiku 5.5 for satisfaction. It is free in Tencent's WorkBuddy through November 30, at half price on its TokenHub API, and the weights are promised for open release, with OpenRouter availability planned.

In plain English

One day after Alibaba's Qwen4-Max, China's second tech giant cut frontier-class AI prices again - this time to pocket change.

Why it matters

Two Chinese flagships in two days, both claiming near-parity with US models at a fraction of the price, turns the model race into a price war - every figure here is Tencent's own testing, but the API prices are real and published. Launched November 16 and included as catch-up coverage.

SourcesTechNode on the launch and pricing·CASRAI's summary

Agents gone wrong

5Amazon: AI agents ran cybercrime campaigns that hunt other AI apps for their keys

Amazon's threat-intelligence team says it watched cybercrime operators use AI agents to run attack campaigns against AI applications themselves. From October 3 to November 5, the company's MadPot decoy network - servers posing as popular AI services - captured records from agents built on a framework it calls ARTEX: an operator supplies a target and a goal, a planner agent splits it into tasks, and worker agents run security tools while a language model decides each next step. The agents hunted internet-exposed AI gateways and app builders - LiteLLM, Dify, One API, Open WebUI and others - for the provider API keys stored inside, verified the keys worked and pooled them. Amazon assesses, with moderate confidence, that separate China-linked groups were behind the campaigns, and cautions it cannot verify what the agents achieved on systems it does not watch.

In plain English

Criminals are now using AI agents to rob AI apps - and the agents' own chatter, captured by decoys, shows the whole playbook.

Why it matters

The week before Trump and Xi discuss out-of-control agents at APEC, here is the live case study: autonomous attack tooling aimed at the AI supply chain itself. Amazon disclosed its findings November 13; The Hacker News reported them November 17 - included as catch-up coverage.

SourcesAmazon Threat Intelligence's disclosure·The Hacker News' report

Checked against original announcements and named press reports on November 17, 2026, morning. Nvidia's results had not been published at writing; consensus figures are analyst surveys, not results. The national-champions plan was first reported by Reuters on November 15 and the company list remains unconfirmed; Tencent's Hy4 Flash launched November 16 and Amazon's disclosure published November 13 - all three are included as clearly-dated catch-up coverage. Tencent's cost and satisfaction figures are its own tests. Amazon's attribution is its own moderate-confidence assessment. Microsoft holds its Ignite keynote today, with announcements due at its conclusion; Anthropic's IPO is expected to price Wednesday.