Nvidia weighs a stake in Perplexity, Hugging Face tests the market for a sale, agent protocols consolidate under one foundation, and the security industry starts buying agent red-teamers. The week’s signal, minus the noise.
The last few days delivered a clear theme: the agent economy is consolidating. Money is consolidating around the winners, protocols are consolidating under neutral governance, and security vendors are consolidating agent-defense talent before the rest of the market wakes up. Here are the six stories worth your attention, each picked for what it means to people building and securing systems for a living. Tap through the log files below for the quick version:
[2026-08-20 protocols/a2a]
Google’s Agent2Agent (A2A) protocol formally joined the Linux Foundation-backed Agentic AI Foundation, landing next to Anthropic’s MCP. The AAIF has grown past 250 members including AWS, Google, Microsoft, and OpenAI.
Why it matters: the agent-to-agent layer and the agent-to-tool layer now sit under one neutral roof. Fewer protocol silos, saner security patching, easier enterprise sign-off. Forbes
[2026-08-23 funding/perplexity]
Nvidia is in talks to invest in Perplexity at a valuation above $30 billion, per The Information. Annualized revenue has reportedly passed $750 million, up from under $250 million at the start of the year.
Why it matters: the growth is credited largely to Perplexity Computer, its agent for automating professional computer work. Agents, not search, are carrying the valuation. US News
[2026-08-24 ma/huggingface]
Hugging Face has retained a bank to test buyer interest in a sale at $13 billion or more, nearly triple its 2023 valuation. No buyer identified, no deal reached, early stages.
Why it matters: the open-model commons changing hands would touch nearly every ML pipeline that pulls weights from the Hub. Worth watching who bids. TechCrunch
[2026-08-21 security/fortinet]
Fortinet acquired Virtue AI, folding AI-agent red-teaming and runtime attack protection into its platform. The pitch: find agent vulnerabilities before attackers exploit them.
Why it matters: agent security just became an acquisition category. Expect every major security vendor to buy or build an equivalent within a year. AI Agents Directory
[2026-08-24 robotics/xpeng]
XPeng’s robotics unit raised over $900 million at a $6.3 billion valuation, with Alibaba and Tencent among backers. Its IRON humanoid targets mass production by end of 2026.
Why it matters: China’s largest embodied-AI round to date. Physical AI is pulling serious capital, and fleet ops for robots looks a lot like DevOps. Electrek
[2026-08-21 appsec/veracode]
Veracode’s latest testing found AI-generated code passed security checks only 56 percent of the time, with 44 percent of samples containing OWASP Top 10 class flaws. Java fared worst at a 30 percent pass rate.
Why it matters: AI-assisted velocity without AI-aware review is negative progress. Scanning and review gates matter more now, not less. eSecurity Planet
A2A joins the Agentic AI Foundation
On August 20, Google formally moved its Agent2Agent protocol into the Agentic AI Foundation, the Linux Foundation-directed body that already governs Anthropic’s Model Context Protocol. The AAIF has grown from 49 to more than 250 members in under a year, with AWS, Google, Microsoft, and OpenAI among its platinum signatories.
Why it matters: MCP standardizes how an agent reaches tools and data. A2A standardizes how agents delegate work to each other. With both under neutral governance, the interoperability bet gets much safer for enterprises, and security teams get a single upstream to track for protocol patches instead of a vendor-by-vendor mess. Source: Forbes.
Nvidia weighs a Perplexity stake above $30 billion
The Information reported over the weekend that Nvidia is discussing an investment in Perplexity as part of a round that would value the company north of $30 billion, a jump of more than 50 percent over its last round. Annualized revenue has reportedly climbed past $750 million.
Why it matters: the revenue surge is being attributed to Perplexity Computer, its cloud agent that automates computer-based work for professionals. The market is now pricing agents that do work above chat that answers questions, which is worth internalizing if you are deciding where to build skills. Source: US News.
Hugging Face explores a sale at $13 billion or more
Hugging Face has reportedly retained a bank to gauge acquirer interest at a valuation of $13 billion or higher, nearly triple where it raised in 2023. No buyer has been named and the process is early.
Why it matters: an enormous share of production ML quietly depends on the Hub for model weights, datasets, and libraries. Whoever ends up owning that chokepoint will shape licensing, availability, and supply-chain trust for the whole ecosystem. If your pipelines pull from the Hub, this is infrastructure news, not finance news. Source: TechCrunch.
Fortinet buys Virtue AI, and agent security becomes an M&A category
Fortinet acquired Virtue AI this week, adding AI-agent red-teaming and runtime protection to its portfolio. The goal is to catch vulnerabilities in deployed agents before attackers do.
Why it matters: for cybersecurity professionals this is a hiring signal as much as a product one. Agent red-teaming just went from research niche to line item at a major vendor, and the skills to attack and defend tool-calling systems are about to be in demand everywhere. If you want structured practice on the defense side, our cybersecurity and DevOps courses cover the fundamentals these roles are built on. Source: AI Agents Directory.
XPeng’s robotics arm raises over $900 million
XPeng’s humanoid robotics unit closed a round of more than $900 million at a post-money valuation above $6.3 billion, with IDG Capital leading and Alibaba and Tencent participating. Its IRON humanoid is slated for mass production by the end of 2026.
Why it matters: physical AI is absorbing capital at software-AI rates, and the operational stack behind robot fleets, telemetry, OTA updates, incident response, looks strikingly like the DevOps toolchain. That overlap is a career adjacency worth noting early. Source: Electrek.
Veracode: AI-generated code passes security checks just 56 percent of the time
Research highlighted in this week’s security roundups found AI-generated code cleared security testing only 56 percent of the time, with 44 percent of samples containing OWASP Top 10 vulnerabilities. Results varied sharply by language, with Python at 63 percent and Java at just 30.
Why it matters: teams shipping AI-written code at AI speed without adjusting their review gates are accumulating vulnerabilities at AI speed too. The uncomfortable but useful takeaway: static analysis, dependency scanning, and human review are appreciating assets. Source: eSecurity Planet.
