AI Trends — 15 May 2026

🌐 The Big Story: US and China Move Toward AI Safety Guardrails

In what may be the week's most geopolitically significant development, the US and China have opened formal discussions on shared AI guardrails at the Beijing summit. Treasury Secretary Scott Bessent confirmed that Washington is prepared to engage Beijing on establishing mutual safeguards — a rare note of cooperation between two countries in fierce technological rivalry.

The discussions center on protocols to prevent non-state bad actors from exploiting the most powerful AI models, including potential agreements around disclosure of dangerous model capabilities, restrictions on proliferating advanced AI to rogue states, and shared red-teaming standards. Bessent was careful to frame the US position: "What we don't want to do is stifle innovation. So our responsibility is to come up with the highest performance calculus where we can get the most innovation and the highest level of safety."

Skeptics note that China used similar Biden-era talks primarily to gather intelligence rather than engage seriously — and that this round's delegations will be watched closely for technical depth.

🏥 Healthcare AI: OpenEvidence Hits $12B Valuation

OpenEvidence, the AI-powered medical search engine built for clinicians, has reached a $12 billion valuation following a $250 million funding round — doubling its previous assessment. The numbers behind the milestone are striking: the platform now serves approximately 65% of US doctors and logged nearly 27 million clinical encounters in April alone.

The service helps physicians search vast databases of medical literature to support clinical decisions, medication choices, and licensing exam prep. What started as a niche tool has quietly become a core part of daily clinical workflow across American healthcare. The clinical AI landscape is clearly moving from novelty to standard infrastructure.

⚡ Power Is the New Compute: GridCARE Raises $64M

GridCARE emerged from stealth this week with a $64 million Series A led by Sutter Hill Ventures and John Doerr, with a blunt thesis: power, not compute, is now the defining constraint for AI. The company is positioning itself as "power acceleration for AI" — helping data centers and hyperscalers solve the energy bottleneck that has become the silent limiter on AI scaling.

This framing is increasingly common in infrastructure circles, where the race to build more GPUs has been outpaced by the challenge of actually powering them.

🔴 Cisco Reorganizes Around AI, Cuts 4,000 Jobs

Cisco announced it will cut nearly 4,000 positions (less than 5% of its workforce) while redirecting investment aggressively toward AI, silicon, optics, and security. The move underscores how incumbent enterprise tech companies are reorganizing around AI demand. Cisco has already secured $5.3 billion in AI infrastructure orders from hyperscalers this fiscal year and now projects that figure will reach $9 billion.

🤖 New Models: Chinese Labs, Anthropic, and xAI

The model release pace remains relentless:

Chinese open-weights surge. Four Chinese labs released competitive open-weights coding models within a 12-day window — Z.ai's GLM-5.1, MiniMax M2.7, Moonshot's Kimi K2.6, and DeepSeek V4. None costs more than a third of Claude Opus 4.7 at inference, raising pressure on Western labs to justify their premium pricing.

Anthropic releases Claude Opus 4.7. The latest Opus model brings substantially improved vision capabilities, higher image resolution, and notable gains in creative and professional task quality.

xAI launches Grok 4.3. Elon Musk's AI company launched its latest model at an "aggressively low price," bundling a powerful voice cloning suite and a new "Imagine" agent mode tailored for creative projects.

Google's Gemma 4 family (released May 4) continues to boost the open-source ecosystem, giving developers access to capable models for on-device and edge deployments.

🦾 Physical AI: Neurovia and NVIDIA Push the Frontier

Neurovia AI (a Robo.ai subsidiary) launched the NeuroStream platform this week, targeting the data infrastructure challenge for physical AI. Using a bitmap vectorization algorithm, NeuroStream delivers high-fidelity visual data streams at low bandwidth and power — a key bottleneck for real-world robotics and autonomous systems.

Meanwhile, NVIDIA's Isaac GR00T open models continue to enable robots to understand natural language instructions and perform complex multistep tasks. General availability of Isaac Sim 6.0, Isaac Lab 3.0, and Omniverse NuRec technologies is helping developers validate robotic systems before costly physical deployment.

💊 Pharma Goes All-In: Novo Nordisk + OpenAI

Danish pharmaceutical giant Novo Nordisk announced a sweeping strategic partnership with OpenAI to integrate AI across its entire business — from drug discovery and clinical trials to manufacturing, supply chains, and commercial operations — with full deployment planned by end of 2026. If successful, it will be one of the most comprehensive pharma-AI integrations attempted at enterprise scale.

📊 By the Numbers

17.8% of the world's working-age population now uses AI (up from 16.3% in Q4 2025)
27.5% usage rate in the Global North vs. 15.4% in the Global South — the AI divide is widening
$9B in AI infrastructure orders projected by Cisco this fiscal year
$12B valuation for OpenEvidence after its latest round
27M clinical encounters handled by OpenEvidence in April 2026 alone

Sources: Hipther AI Dispatch May 14 · Daily Signal – US-China AI Guardrails · Axios – US-China AI · SiliconAngle – OpenEvidence · Anthropic – Claude Opus 4.7 · Microsoft – State of Global AI Diffusion · Nathan Benaich – State of AI May 2026