Nvidia is reportedly in talks to invest in Perplexity at a valuation north of $30 billion — a signal that AI’s money is consolidating around the compute-and-inference layer even faster than around the models themselves. From OpenAI switching on ads across Europe to a leadership earthquake at Google DeepMind, today was about who controls the rails, not just who ships the smartest model.
Perplexity
What happened
The Information reports that Nvidia is in talks to take a stake in Perplexity at a valuation above $30 billion — up more than 50% from its roughly $20 billion round a year ago. Perplexity’s annualized revenue has climbed past $750 million, up from under $250 million at the start of 2026, and it recently committed to Nvidia’s Vera CPUs for agent workloads.
What it means for your agentic build
An AI answer engine is now being capitalized as core infrastructure, not a search novelty. If you are evaluating Perplexity’s Agent, Search, and Embeddings APIs as a build layer, note that its roadmap is increasingly tied to Nvidia’s silicon — a bet on availability and cost, but also on a single supplier’s stack.
OpenAI
What happened
ChatGPT ads went live in Spain and expanded to 31 European countries today, appearing for free and ChatGPT Go users while Plus, Pro, Business, Enterprise, and Education plans stay ad-free. The rollout follows OpenAI’s “AI Futures” announcement a day earlier, and new data shows OpenAI gaining ground on Anthropic among business users.
What it means for your agentic build
The ad tier signals a consumer-monetization pivot that does not touch paid enterprise plans, but it changes the data and attention economics beneath the free surface your employees may already use. If you standardize on ChatGPT, keep staff on managed paid tiers where ads and ambiguous data terms do not apply.
Anthropic
What happened
Anthropic is reportedly preparing an IPO that could rival SpaceX’s record debut and may file as soon as month-end. It also announced an Oracle Cloud Infrastructure partnership and growing financial-services uptake of Claude 3.1 Guardian, a compliance-tuned model, while its computer use, browser use, Skills, and Files APIs reached general availability.
What it means for your agentic build
Guardian, OCI, and generally available agent tooling make Claude materially more deployable inside regulated workflows. A pending IPO adds financial transparency and reference-customer momentum, but also public-market pressure — factor vendor stability into any multi-year platform commitment.
Google DeepMind
What happened
Demis Hassabis has moved to chairman of Google DeepMind and chief scientist of Alphabet, with former CTO Koray Kavukcuoglu now running day-to-day operations. Separately, chief scientist Jeff Dean and longtime colleagues Sanjay Ghemawat, Quoc Le, and Oriol Vinyals departed to found Discovery Loop, a public-benefit company automating scientific research that Google will back.
What it means for your agentic build
Leadership churn and reported model delays raise fair questions about Gemini’s near-term delivery cadence. Google’s distribution and cloud remain formidable, but if your roadmap depends on Gemini timelines, build in contingency and watch shipping velocity over the next two quarters.
Meta AI
What happened
Meta is pressing its open-weight strategy, opening the weights of Muse Glimmer and, soon, a version of Muse Spark 1.2, following Mark Zuckerberg’s “The Future is for Everyone” manifesto. At the same time, its AI infrastructure spend is pushing free cash flow toward negative territory this quarter, and a child-safety case has grown to a $942 million liability.
What it means for your agentic build
Open weights lower the cost and control barriers to self-hosting frontier-class models — worth evaluating for on-prem or data-sensitive builds. But weigh Meta’s cash-flow strain and legal exposure as governance signals about how durably it will fund and support the program.
xAI
What happened
xAI’s Grok 4.6 — a 500K-context model with configurable reasoning and $2/$6 per-million-token pricing — is now available through Google’s Enterprise Agent Platform and Model Garden. The company also launched Grok Bot, its “AI teammates you can give real work to,” and pushed Grok Build to every plan, while Grok 5 remains in training.
What it means for your agentic build
Grok 4.6’s presence in Google’s Model Garden gives enterprises a governed, procurement-friendly way to trial it for long-running agents without leaving their cloud. The teammate framing is aimed squarely at delegating real work, so pilot it against concrete tasks rather than benchmarks.
DeepSeek
What happened
DeepSeek’s V4-Pro reached general availability — an agent-focused model with a 1 million-token context window and 384,000-token outputs. The company is raising close to $8 billion at roughly a $74 billion valuation and has begun raising prices, adopting peak and off-peak rates ahead of a possible IPO, though its pricing still undercuts Western rivals.
What it means for your agentic build
Cheap, capable agentic models keep downward pressure on incumbents’ prices, and V4-Pro’s long context suits document-heavy automation. But price increases, IPO prep, and data-governance questions around a China-based provider should shape where and whether you deploy it — favor low-sensitivity pilots first.
Mistral AI
What happened
Mistral introduced Agentic Search, a retrieval layer that reads and verifies complex documents with fewer turns, lower token use, and reduced latency. It is also building a 10 MW inference data center at Les Ulis and leading a coalition to secure long-term European compute, while extending its Vibe agent across inbox, calendar, research, and coding.
What it means for your agentic build
For European organizations with data-residency requirements, Mistral pairs credible agentic tooling with regional inference. Agentic Search specifically targets the accuracy and latency problems that undermine document-heavy retrieval systems, making it worth benchmarking against your current stack.
Cohere and Aleph Alpha
What happened
Cohere released Command A+ as open source — a 111-billion-parameter mixture-of-experts model with a 256,000-token context that runs on two A100 or H100 GPUs and delivers roughly 150% higher throughput than its predecessor. The move builds on Cohere’s April merger with Germany’s Aleph Alpha and its private, on-premise revenue base ahead of a planned IPO.
What it means for your agentic build
Command A+ is an efficient, open-weight model designed for private and regulated deployments — a strong candidate where data cannot leave your environment. Combined with Aleph Alpha, Cohere is consolidating a sovereign, multilingual enterprise stack aimed directly at European and public-sector buyers.
This Week’s Structural Trends
Capital is consolidating around compute, not just models. Nvidia’s reported $30 billion-plus Perplexity stake, DeepSeek’s $8 billion raise, and IPO preparations at Anthropic and Cohere show investors financing the inference-and-compute layer as the durable moat — with Nvidia increasingly taking equity in the very companies that buy its chips.
The agentic pivot is now the product. DeepSeek V4-Pro, Mistral’s Agentic Search and Vibe, xAI’s Grok Bot, Anthropic’s computer- and browser-use GA, and Cohere’s North platform all ship tools that act rather than merely chat. Evaluate models on tool use, long-context reliability, and workflow delegation — not benchmark scores.
Monetization and sovereignty are splitting the market. OpenAI’s European ads and DeepSeek’s price hikes sit opposite open-weight and sovereign pushes from Meta, Mistral, and Cohere–Aleph Alpha. Enterprises now face a real fork between hyperscale proprietary platforms and open, self-hosted, region-controlled stacks.
Sources
https://www.benzinga.com/markets/tech/26/08/61374579/nvidia-perplexity-30-billion-valuation-750-million-revenue-bezos
https://guslok.com/en/anuncios-chatgpt-espana-25-agosto/
https://techcrunch.com/2026/08/20/openai-is-gaining-on-anthropic-with-business-users-new-data-indicates/
https://techstartups.com/2026/08/21/top-tech-news-today-august-21-2026-anthropic-apple-broadcom-google-nvidia-openai-tesla-more/
https://www.axios.com/2026/08/05/google-deepmind-demis-hassabis-ai
https://www.washingtonpost.com/technology/2026/08/10/zuckerberg-manifesto-says-meta-ai-will-make-future-everyone/
https://financebuzz.com/news/meta-stock-things-to-know-august-24-2026
https://codersera.com/blog/grok-4-6-launch-guide-2026/
https://qz.com/deepseek-v4-pro-official-launch-081326
https://www.pymnts.com/news/artificial-intelligence/2026/deepseek-resumes-funding-round-to-raise-8-billion/
https://mistral.ai/news/agentic-search/
https://betakit.com/cohere-releases-its-most-powerful-ai-model-as-open-source/
https://www.cnbc.com/2026/04/24/cohere-aleph-alpha-germany-ai-europe-expansion.html