US and China Back Open-Source AI at APEC as OpenAI and Anthropic Race Voice Agents to the Desktop — July 24, 2026

⚡ Top Story

US and China Back Open-Source AI at APEC's Chengdu Statement — First Ministerial-Level Cooperation on Open Models

Twenty-one APEC economies, including the United States and China, adopted the "Chengdu Statement" on July 23 at the APEC Digital and AI Ministerial Meeting, backing continued development of open-source AI models built with "strong security assurance" while calling for respect of "security, data protection and intellectual property rights." China's Minister of Industry and Information Technology Li Lecheng, who chaired the meeting, called it the first APEC AI statement to secure open-source cooperation at the ministerial level.

Why it matters: It's a rare moment of explicit US-China alignment on AI policy even as the two governments simultaneously escalate bilaterally — Washington threatening sanctions over alleged Chinese model "distillation" this same week, and Beijing reportedly weighing its own export controls on model weights. Multilateral cooperation and bilateral containment are running on separate tracks, not in sequence.

Sources: CNBC: U.S., other nations back open-source AI with 'strong security' at China summit · Xinhua: APEC ministerial meeting adopts Chengdu Statement on digital technologies, AI · APEC: 2026 APEC Digital and AI Ministerial Statement

🔬 Research & Papers

Nothing independently verified as newly published in the last 24 hours met the bar for inclusion. A sweep of arXiv (cs.AI, cs.LG, cs.CL, cs.CV), Papers With Code, and major lab research pages turned up no new results dated July 23–24 that weren't either already covered in prior briefings or too preliminary to independently verify.

🏢 Industry & Startups

AI Inference-Chip Startup Etched Doubles Its Valuation to $10.3 Billion in Seven Months

Etched, a startup building chips dedicated to AI inference rather than general-purpose GPU training, raised $300 million in Series C funding on July 23 at a $10.3 billion valuation — the highest valuation ever for a Sequoia-led Series C. The round included a16z, Jane Street, Diffusion, and SK hynix. Etched was valued at $5 billion in December 2025 and says it already has $1 billion in customer pre-orders.

Why it matters: A concrete signal that investors see durable demand for inference-specialized silicon as a distinct category from Nvidia's general-purpose training GPUs, not just a niche bet.

Sources: GlobeNewswire: Etched raises $300M at a $10.3B Valuation · TechCrunch: AI chip startup Etched defies skeptics, hits $10.3B valuation

Devin-Maker Cognition Acquires Poke Developer Interaction

Cognition, maker of the autonomous coding agent Devin, announced on July 23 it has acquired The Interaction Company of California, maker of Poke — a personal AI agent that lives inside a user's text messages and has exchanged more than 100 million messages with users in the three months before the deal. Terms weren't disclosed.

Why it matters: A consolidation move between two "always-on cloud agent" products built for different surfaces — coding vs. personal messaging — betting on the same underlying architecture; worth watching whether Poke keeps its own identity or gets folded into Devin.

Source: Cognition: Welcoming The Interaction Company

🛠️ Tools & Releases

OpenAI and Anthropic Both Ship Multi-Model Voice Agent Upgrades on the Same Day

OpenAI brought full-duplex ChatGPT Voice (powered by GPT-Live) to its Mac and Windows desktop apps on July 23, letting users control their computer and direct multiple agents in ChatGPT Work or Codex by voice. Separately, OpenAI made ChatGPT Health — which connects Apple Health data and supported medical records to a dashboard with AI-generated insights — available to all logged-in US users 18+, after testing it since January. The same day, Anthropic upgraded Claude's voice mode to let paid users choose between Haiku, Sonnet, and Opus mid-conversation (previously Haiku-only), added voice-triggered automation across Gmail, Google Calendar, Slack, Canva, and Notion, and expanded language support to ten languages.

Why it matters: Two competing labs shipped functionally similar voice-agent upgrades — multi-model selection, cross-app/computer action-taking — on the same day, a sign voice-as-agent-interface is becoming a competitive baseline rather than a differentiator.

Sources: OpenAI (via X): ChatGPT Voice is now in the desktop app · VentureBeat: GPT-Live's full duplex voice control comes to Codex and ChatGPT desktop · OpenAI: Introducing ChatGPT Health · TechCrunch: OpenAI makes ChatGPT Health available to all US users · TechCrunch: Anthropic updates Claude voice mode with more capable models

🌏 Global AI & Geopolitics

China Sends Its Vice Foreign Minister to Washington to Scope Out September's AI Dialogue

Chinese Vice Foreign Minister Ma Zhaoxu visited Washington this week to sound out US agencies on the scope and likely outcomes of the countries' first official AI dialogue under the Trump administration, planned for September ahead of Xi Jinping's planned US visit on September 24. China says it wants technical engagement rather than political talks; the visit follows Treasury Secretary Bessent's threat earlier this week to sanction Chinese AI labs over alleged model "distillation" of US IP.

Why it matters: It's the first concrete sign of how China intends to respond to the sanctions threat — diplomatically, by trying to shape the agenda of the upcoming talks, rather than through immediate retaliation.

Source: Bloomberg: China Seeks Clarity From US on AI Talks as Sanction Threats Loom

⚡ Energy, Infrastructure & Chips

Trump Administration Rolls Out "Ratepayer Protection Pledge" to Shift AI Power Costs Onto Tech Companies

President Trump on July 23 announced new commitments from power utilities and data-center developers under a "Ratepayer Protection Pledge," aimed at ensuring technology companies — not household electricity customers — bear the cost of new grid infrastructure built to serve AI data centers.

Why it matters: It's the clearest federal policy response yet to rising consumer anxiety over AI-driven electricity price increases, directly tying White House messaging to the hyperscaler capex numbers in yesterday's briefing.

Source: Bloomberg: AI Data Center Growth Drives Power Cost Commitments From US Tech Giants

South Korea's LS Electric and LG Uplus Sign MoU on AI Data Center DC Power Infrastructure

The two South Korean firms agreed July 23 to jointly develop and standardize 800V DC power infrastructure for AI data centers designed around Nvidia's upcoming Vera Rubin platform.

Why it matters: A concrete, non-US data point on the DC-power-infrastructure race built specifically around next-generation Nvidia hardware, ahead of Vera Rubin's actual launch.

Source: Data Center Dynamics: LS Electric, LG Uplus ink MoU

🤖 AI Agents & Autonomy

HubSpot Launches Breeze Agent Hub and Agent Builder in Public Beta

HubSpot rolled out Agent Hub and Agent Builder in public beta on July 23 for Professional and Enterprise customers — a central dashboard for configuring, monitoring, and managing both HubSpot's prebuilt AI agents (support, sales prospecting, content) and custom customer-built agents.

Why it matters: A mainstream CRM vendor formalizing agent management and governance tooling for non-technical go-to-market teams, echoing the "governed agent platform" pitch OpenAI made with Presence earlier this week.

Source: CMSWire: HubSpot Launches Agent Hub and Builder in Public Beta

🔒 Safety, Alignment & Ethics

Nothing independently verified as newly published in the last 24 hours from validated safety orgs or labs with a confirmed July 23–24 date.

📊 Numbers & Signals

21 — APEC economies (including the US and China) that signed the Chengdu Statement on open-source AI
$300 million / $10.3 billion — Etched's new Series C raise / resulting valuation, up from $5B in December 2025
$1 billion — Etched's existing customer pre-orders
100 million+ — messages exchanged with Poke (Interaction) in the three months before its acquisition by Cognition
10 — languages now supported in Claude's upgraded voice mode

🧠 Worth Thinking About

Today's stories cluster around interfaces and infrastructure quietly becoming the real competitive battleground, not raw model capability. OpenAI and Anthropic shipped strikingly similar upgrades on the same day — multi-model voice, cross-app action-taking — suggesting voice-as-agent-interface has become table stakes rather than a differentiator between the two labs. Meanwhile, the APEC Chengdu Statement shows the US and China finding room to cooperate on open-source AI standards even as the same two governments circle each other over export controls and IP-theft sanctions threats — multilateral cooperation and bilateral containment running on separate tracks, not resolving into one coherent policy. And underneath both, a chip startup doubling its valuation in seven months and a White House event about who pays for AI's electricity bill are two symptoms of the same fact: the infrastructure buildout from yesterday's Alphabet earnings hasn't slowed — it's just generating steadier, less dramatic news around it now.

🏛️ Government & Regulation

No new formal US, EU, or other regulations, legislation, or executive orders were independently verified as enacted in the last 24 hours. See Energy, Infrastructure & Chips above for the Trump administration's "Ratepayer Protection Pledge," a policy commitment rather than binding regulation.

🔭 Frontier Lab Dispatch

OpenAI — July 23: Shipped full-duplex ChatGPT Voice for desktop with computer/multi-agent control, and rolled out ChatGPT Health to all US users (see Tools & Releases).

Anthropic — July 23: Upgraded Claude's voice mode with Opus/Sonnet model choice and cross-app automation (see Tools & Releases).

🔗 Quick Links

Tier 1 — Official / Primary Sources

Tier 3 — Tech & AI News Media