Global Top 5 (Nov 6): AI Capex Repriced, AI Traffic Tiny

On November 6, 2026, global technology and business sent two opposing signals on the same day: capital markets repriced the payback period of AI infrastructure in a historic sell-off, while models, tooling and agentic entry points kept getting cheaper and spreading wider — with merchants already moving first. Below are five same-day stories that matter most to AI, digital marketing and cross-border business, each with the facts and a practical takeaway. Data as of 14:00 Beijing Time, November 6.

1. Tech stocks shed about $2.5 trillion in a day as AI capex is repriced

The Nasdaq fell 5.52% to 24,314.997, down 11.9% from its October 6 closing high and into technical correction. The Philadelphia Semiconductor Index dropped 12.25%, its worst single day since March 2020; the S&P 500 lost 4.36%; the VIX jumped 69.48% to 28.15. Roughly $2.5 trillion in technology market value evaporated in one session, about $1 trillion of it from the seven largest tech names. Memory stocks were hit hardest, with SanDisk down more than 22%.

What is being repriced: not AI demand itself, but data-center approval requirements, power costs, and how quickly massive capital expenditure converts into cash flow. A related signal: Apple tightened employee AI budgets, cutting daily model-spend caps for some engineering teams from $300 to $200 and considering $100, while steering teams toward cheaper or locally run models.

Takeaway for exporters: the unlimited-AI narrative is giving way to unit economics. Batch content generation, multilingual rewriting and AI customer service should be budgeted against acquisition return per asset, not against token volume.

2. Google commits $17B to its first own data center as Microsoft cancels a $3B lease

Infrastructure told both stories at once. Google announced a $17 billion investment in its first self-built data center in Wyoming, paired with 702 MW of fuel cells and carbon capture, and committed to covering 100% of the facility electricity costs. At the same time, Microsoft withdrew from a $3 billion UK lease with Nscale and dropped a Texas expansion option, as questions grew over the debt-heavy operating model of AI infrastructure.

Takeaway for exporters: even hyperscalers are switching from securing capacity first to settling the ten-year math first. The same applies when you build your own website, knowledge base and agents: prefer portable, measurable architectures, keep content assets and customer data in your own hands, and avoid locking long-term assets into a single vendor.

3. OpenAI ships GPT-6 Cyber, with stronger security skills but vetted access only

OpenAI released GPT-6 Cyber, a research model for cybersecurity. Public reporting says its exploit-benchmark score rose 51% over the previous generation and that it has helped uncover hundreds of high-severity vulnerabilities. The model is not generally available; access is limited to vetted security teams. Stronger capability paired with a narrower release is the most notable part of this launch.

Takeaway for exporters: when agents are connected to your website, knowledge base, enquiry forms or order systems, permission tiers, action logs and audit trails must mature alongside capability. For overseas buyers, an explainable data boundary is itself a trust asset.

4. The AI traffic reality check: 4 Uber rides versus 40,000 restaurants signing up

This is the day most useful comparison for marketers. Uber Q3 results show that out of roughly 320 million weekly orders, third-party AI agents placed just 4 ride orders and about 500 food-delivery orders — a sharp contrast with the expectation that AI shopping would already be reshaping commerce. On the same day, ordering-technology company Gett launched First Party Everywhere, a merchant-direct AI ordering network where orders, payments and customer data all run through the restaurant existing first-party system. More than 40,000 restaurants have signed up. Third-party figures cited by Gett also suggest that AI-referred customers in September converted about 90% better and spent about 83% more per order.

Takeaway for exporters: demand is still a rounding error while supply is already sprinting. That is not a contradiction — it means the current window is about being found by AI, not waiting for AI to bring orders. Three things are worth doing now: turn business details, product specs, pricing and multilingual metadata into machine-readable structured content (schema, hreflang, llms.txt) so generative engines can read and cite you accurately; make sure enquiry and checkout entry points can be reached from external channels; and decide upfront who owns the customer data an AI channel produces. That is exactly what GEO, Generative Engine Optimization, solves: first become citable, then become orderable.

5. Model and tooling prices keep falling as AI revenue shifts to enterprises

• Falling costs: Google launched Gemini Omni 1.2 Flash for video generation, cutting 720p pricing by 30% to about $0.07 per second, with native clips up to 15 seconds, conversational multi-turn editing and stronger reference consistency. A 15-second 720p clip now costs roughly $1.
• Wider use cases: Anthropic is testing personal finance features in Claude that connect to bank accounts and credit cards to query spending and answer money questions.
• Revenue mix: OpenAI president Greg Brockman said most of the company revenue now comes from enterprise sales, and that it expects to be ready for an IPO in 2027.

Takeaway for exporters: the marginal cost of short video, product imagery and multilingual landing pages keeps dropping, which turns an 18-language content matrix from unaffordable into overdue. The real dividing line, however, is not generation speed — it is whether that content is structured enough to be reliably cited by search engines and generative engines.

Conclusion: keep two separate ledgers

Capital is cooling while applications are heating up. Exporters do not need to rebuild their business around imagined AI traffic, but they should not wait until orders become visible either. The safer sequence is: make brand and product information readable to AI, let enquiry and transaction entry points be reachable by external agents, and clarify data ownership. The traffic will arrive; get the information and the interfaces aligned first.

Sources: Wallstreetcn, Jiemian, 21st Century Business Herald, Cailianshe, ifanr, Phoenix Technology and other public reports; Uber figures from its 2026 Q3 results, Gett and third-party figures from its official release (facts only). This is an industry roundup based on public reporting and is not investment advice.