NVIDIA's Vera CPU Hits First Public Benchmarks — Outpacing Rivals on Agentic AI Workloads

Phoronix has published the first public benchmark results for NVIDIA's in-house CPU, Vera. It packs 88 custom Armv9.2-compatible Olympus cores and delivers 1.2TB/s of memory bandwidth. Optimized for agentic AI workloads in AI factories, the chip reportedly showed strong performance on workloads that have traditionally bottlenecked conventional CPUs, such as branch-heavy runtimes and sandboxed code execution.

NVIDIA's move beyond GPUs into serious CPU territory could reshape how AI infrastructure is designed altogether. These benchmark results symbolize how the CPU's role is being redefined as agentic AI goes mainstream.

DuckDuckGo Installs Surge 30% as Users Push Back on Google's AI Search

Backlash is growing against Google's decision to rebuild Search entirely around AI agents at I/O 2026. According to TechCrunch, DuckDuckGo's app installs have jumped 30%. Users are seeking alternatives out of resistance to being forced into AI search.

The end of the blue links era and the shift to a design where AI mediates the search experience mark a turning point that fundamentally calls into question what a search engine should be. The fact that users are actively seeking out alternatives on their own shows that an AI-first approach isn't necessarily winning over every segment of users.

Uber Burns Through Its Quarterly AI Budget — Corporate AI Spending Under Scrutiny

According to Fortune, Uber blew through its quarterly AI budget in a single quarter. The company's COO says usage of AI tools such as Claude Code vastly exceeded expectations.

As enterprise adoption of AI coding tools accelerates, this is a case that highlights just how difficult cost management can be. Usage-based, per-token pricing models can squeeze budgets in no time once developer usage expands beyond forecasts. How to evaluate the ROI of AI usage and keep costs under control is becoming a new management challenge in enterprise AI adoption.

SK hynix Unveils Self-Cooling iHBM Chip — Taking on AI's Overheating Problem

SK hynix has announced a self-cooling 'iHBM' chip designed to tackle overheating during AI inference. According to The Korea Times, integrating cooling functionality directly into the chip is drawing attention as an approach that could fundamentally solve heat problems under AI workloads.

As AI semiconductors grow more powerful, their heat output keeps climbing, and cooling solutions are becoming a factor that shapes the design of entire data centers. Chip-level cooling integration could open up new options for designing high-density AI servers.

ByteDance Offers Special Shares to AI Teams — The Talent War Intensifies

According to the Financial Times, ByteDance is offering special equity incentives to employees in its AI division, stepping up efforts to prevent poaching by competitors. As the fight over AI talent intensifies, Chinese tech companies are starting to fight back on compensation.

Since the rise of DeepSeek, talent mobility in China's AI industry has been on the upswing, and companies are rolling out measures to keep employees from jumping ship to rivals or startups. It's a move that suggests the main battleground of the AI race is shifting from model development to securing talent.

---