AI in the News

AI in the news: week of November 2, 2025

GitHub Universe ships Agent HQ. Big-tech earnings push 2025 AI capex past a quarter trillion. Amazon cuts 14k jobs the same week. OpenAI completes its PBC restructuring. Cursor ships its own model. The week that made the AI capex-vs-jobs trade explicit.

Sid Smith

02 Nov 2025 • 9 min read

What this week actually changed: the AI capex-and-AI-layoffs pairing finally got told as one story. Big-tech earnings landed Wednesday and Thursday with capex numbers that, summed across Microsoft, Google, Meta, and Amazon, clear a quarter trillion dollars for the calendar year. Same week, Amazon announced 14,000 corporate layoffs while citing "our biggest bets", which means AI. Add GitHub Universe announcing Agent HQ, OpenAI completing its PBC restructuring with Microsoft locking in 27%, and Cursor shipping its own coding model, and the week's dominant story is the capex-and-cuts pairing landing on the same page.

GitHub Universe and Agent HQ

October 28 in San Francisco. GitHub used Universe to announce Agent HQ, a platform inside Copilot that lets developers run, orchestrate, and govern agents from multiple vendors (Anthropic, OpenAI, Google, Cognition, and xAI) under a single subscription. The companion piece is Mission Control, a dashboard for assigning work to agents, watching them run, intervening, and reviewing across GitHub, VS Code, mobile, and the CLI. Visual Studio Magazine framed it as "any agent, any way you work," which is the exact pitch.

I want to be careful here because my reflexes pull two directions on this story. The reflex against vendor concentration says: GitHub is owned by Microsoft, the agents are mostly hosted at concentrated frontier labs, and Mission Control is the control plane GitHub gets to design. That's a real concern. The platform layer being owned by one company is the lock-in vector, not the model layer.

But the structurally honest read is that Agent HQ is the most plainly multi-vendor agent posture a major platform has taken to date. The contrast with the alternative, where each lab tries to build the IDE-and-orchestration layer themselves on top of their own model only, is meaningful. Letting Anthropic, OpenAI, Google, Cognition, and xAI agents all run inside the same workflow is the closest we've come to the distributed-orchestration shape being a default rather than a custom build.

The catch, and it's the right catch, is that "multi-vendor inside one orchestrator" is not the same as "distributed." It's polycentric at the model layer and monocentric at the platform layer. The platform layer is where the lock-in lives in the cloud-era pattern. I'll celebrate the multi-vendor model layer and stay skeptical of the single-platform orchestration. Both can be true. What I'll watch: whether the agent-vendor neutrality holds when GitHub and OpenAI agents start competing for the same workload, what the egress story looks like for teams who want to take their Mission Control workflows elsewhere, and whether any non-Microsoft platform ships a credible alternative in 2026.

A quarter trillion dollars of capex in one earnings week

Microsoft, Google, Meta, and Amazon all reported in the same 48-hour window. The numbers are the story. Microsoft Q1 FY26: $77.7B revenue, Azure up 40%, capex growing sequentially with FY26 growth higher than FY25. The OpenAI investment took a $3.1B hit through net income. Alphabet Q3 2025, first $100B revenue quarter in Alphabet history, Cloud up 34% to $15.2B: $155B Cloud backlog, full-year capex guided to $91-93B per Seeking Alpha. Meta Q3 2025: $51.2B revenue, full-year capex raised to $70-72B, with Zuckerberg telling analysts 2026 capex growth will be "notably larger" than 2025 per the Motley Fool transcript. Amazon Q3 2025. AWS up 20% to $33B (best growth in 11 quarters), AWS backlog at $200B, full-year cash capex around $125B with another increase in 2026.

Add it up. Roughly $400B+ of 2025 capex committed across the four, with all four signaling higher in 2026. This is real money, and it is bet on AI infrastructure. GPUs, custom silicon (Trainium2, TPUs), and the datacenters to house them. The Global Data Center Hub piece on Meta is right that this is the moment AI stopped being software and became infrastructure on the cloud-era scale.

My take is unchanged from last week, but the evidence got heavier. The capex bet only pencils if the productivity gains are large, broad, and durable, and "large, broad, and durable" is doing a lot of work in that sentence. The early returns on enterprise AI are uneven enough that "we will spend $125B per company per year on this" is a forward bet on a future that hasn't arrived. The companies will be fine. The bet itself is heavier than the demos justify. The piece I want flagged: capex this large changes the political-economic conversation. When four companies spend $400B+ in a year on AI infrastructure, the cost of being wrong gets externalized. Datacenter power demand, chip supply allocation, the compute concentration question, these become public-policy issues, not just shareholder issues. SB 53 looks even more prescient than it did three weeks ago.

The matching layoffs: Amazon cuts 14,000 corporate jobs

October 28. Same day as Agent HQ. Amazon announced layoffs of 14,000 corporate workers, with reporting from GeekWire and others suggesting up to 30,000 across the full program. Amazon's CEO Andy Jassy framed it as "reducing bureaucracy, removing layers" while investing in "our biggest bets", and per NPR. AI is the load-bearing one of those bets.

I want to be straight about my read on this. The displacement is real and it's accelerating faster than I expected. I always knew it was coming, automation is what I've spent my career on, and work that can be automated eventually is. What I keep coming back to is the pace, and the driver behind the pace. The financial incentives are doing a lot of work here. AI is in a rush phase right now, driven by what the markets reward. Companies aren't cutting at this speed because the AI is ready to cleanly cover the work. They're cutting because the AI narrative is convenient and the markets reward the cuts.

The capex pairing makes the driver explicit. Amazon is committing $125B to AI infrastructure and cutting 14,000 corporate roles in the same week, and pitching the cuts as the operational complement to the bet. The two announcements are designed to be read together, "we are spending big on AI, here's the headcount story that justifies it." That's the analyst-call story. Whether the AI agents actually cover 14,000 jobs of work is a separate question, and the answer is "not yet." The cuts are being made on the forward assumption, and the gap between announcement and execution is where the operational pain accrues. I wrote about this pattern at length, and I'm going to keep writing about it.

The sustainable shape is human+AI collaboration. Companies that build for human-and-AI teams will outperform companies that just cut. To be clear about what I mean: the headcount still shrinks under collaboration. It just shrinks slower, more deliberately, and the surviving humans aren't doing two-and-a-half jobs each while the AI catches up. The 2027 correction cost is lower under collaboration than under racing-to-cut. The Threads observation from a watcher in the field, that Amazon's PR began counter-positioning against the "AI-replaces-jobs" narrative once the optics turned bad, is the early signal that the cut-fast play is going to need a lot of cleanup. I'd rather be wrong about how fast this moves than be caught off guard.

OpenAI's PBC restructure; Microsoft locks in 27%

October 28. Also that day. OpenAI completed its restructuring into a public benefit corporation, with the nonprofit retaining control. Microsoft's stake is now 27%, valued at $135B per the Fortune writeup. OpenAI commits to purchasing $250B more of Azure capacity. Microsoft can pursue AGI independently or with third parties; OpenAI can build with non-Microsoft cloud partners. AGI declarations require an independent expert panel verification.

Two reads. The neutral read: the partnership is now formalized in a structure that satisfies regulators in Delaware and California (both issued no-objection statements per CNBC) and gives both companies a legal-and-financial foundation to keep going. That's a real outcome. The skeptical read: "public benefit corporation controlled by a nonprofit" is a structure designed to look like governance without being constraining. The nonprofit can set "mission", but the PBC has $500B of for-profit upside on the line, Microsoft has $135B of equity in it, and the gravity of those numbers is not symmetric with the gravity of the nonprofit's mission language. The PBC structure is real, but so is the financial engineering around it. I'll watch for the first non-trivial decision the nonprofit makes that costs the PBC money. That's when the structure gets tested.

The $250B Azure commitment is the part with the most directly material consequence. It locks OpenAI into Microsoft cloud at a scale that effectively means OpenAI's compute is on Microsoft's terms for years. The first-right-of-refusal language went away, but the contractual gravity replaces it. That's the lock-in pattern in its most expensive form.

Cursor 2.0 and Composer

October 29. Cursor shipped 2.0 with Composer, its first in-house frontier model, claiming 4x speed at similar quality to other frontier models on agentic coding tasks. Multi-agent interface that can run up to eight agents in parallel from a single prompt. Sandboxed shell execution. Frontend testing via direct DOM access. Simon Willison's writeup is the calmest summary I've read.

The technically interesting part is that Cursor (an IDE company) felt the strategic pressure to ship its own model. The implicit thesis is that the IDE layer and the model layer will increasingly be the same product, and that being a customer of frontier labs leaves margin and control on the table. Same thesis as GitHub Agent HQ, played from the other direction: GitHub is bringing many models into one platform; Cursor is bringing one model into one platform, on the bet that vertical integration wins.

I'm watching whether Composer is actually competitive. Early reports are positive but the model-eval space is noisy. The strategic move matters more than the model quality at launch. Cursor has now signaled that the vibe-coding-IDE category is going to be a model-plus-IDE category, not a wrapper category. That changes what "build your own coding agent" means for everyone else. For my own MCP-only orchestration preference this matters less than it sounds. The architecture I run treats the model as a swappable backend behind an MCP-shaped interface, and Composer is just another endpoint. The day Cursor exposes Composer over an open API at competitive pricing, I'll try it. If they don't, the IDE itself is the lock-in vector and I'll skip.

Smaller items

A few smaller pieces worth a line. Anthropic opened its first APAC office in Tokyo on October 29. CEO Dario Amodei met Japan's PM Sanae Takaichi and signed a Memorandum of Cooperation with the Japan AI Safety Institute. The international expansion matters for the governance story, frontier AI labs operating in jurisdictions with their own AI-safety institute frameworks creates a federated safety conversation rather than a US-only one. Apple Q4 FY25 earnings, $102.5B revenue, all-time record, with Cook framing the next growth cycle as "AI-driven and services-centric." A Siri rebuild on LLM foundations is committed for the 2026 iOS update. Apple's AI strategy continues to be the slowest of the megacaps, which is either patient or behind depending on which quarter you ask. Nvidia GTC DC keynote. Jensen Huang keynoted in Washington on October 28 with announcements on Blackwell production at the Arizona TSMC fab, the NVQLink quantum interconnect, the ARC telecom platform with Nokia, and a Uber autonomous-vehicle partnership for 100,000 cars by 2027. Nvidia continues to be the company that benefits the most when everyone else spends $400B on AI capex. Claude Sonnet 3.5 retirement. Anthropic deprecated Sonnet 3.5 on October 28 with migration recommendations to 4.5. Worth noting for the cadence, labs retiring older models on a roughly six-month cycle is becoming standard practice, and teams running long-running production systems need to plan around it.

What this week tells me

Three things. The AI-capex-and-AI-layoffs pairing is now load-bearing in big-tech narratives. Amazon spending $125B on AI infrastructure and cutting 14,000 corporate roles in the same week is not a coincidence of timing. It's the executive story being told as a single story. The displacement is real and running faster than I expected; the pace is the problem, and short-term incentives are driving it. Companies that build human+AI collaboration will outperform the cut-fast firms, and to be clear, the headcount still shrinks under collaboration, just slower and better.

Multi-vendor agent orchestration is becoming the platform play. GitHub Agent HQ, Cursor 2.0's multi-agent interface, the OpenAI-Microsoft restructuring that plainly allows third-party collaboration on both sides, the structural direction is toward orchestrators that route work across multiple model vendors rather than verticals locked to one. This is mostly good news for the distributed-AI position at the model layer, and a new lock-in question at the platform layer. Both the gain and the new risk are real.

And the OpenAI-Microsoft PBC structure will be a governance test case. A nonprofit controlling a for-profit with a $500B valuation and a $135B Microsoft stake is a governance structure with high-gravity financial counterweights. It's also the first clean test of whether mission-led structure survives the scale where the money matters. The next two years of PBC governance decisions are the data we'll judge it on, and the governance framework conversation needs this case study. Next Sunday: GitHub Agent HQ developer reception, the Amazon layoff-to-execution tracking, whatever the cadence brings.