Good morning. It's Thursday, March 19, and we're covering NVIDIA’s Groq-powered inference push, Anthropic’s enterprise takeover, Midjourney’s V8 model upgrade, and more.

Midjourney Launches Faster, Smarter V8 Model
Midjourney is testing its V8 image model with users. It delivers 5x faster generation, stronger prompt accuracy, and improved text rendering. The update adds native 2K output and enhanced personalization via style references and moodboards. Early access is designed to gather feedback and refine performance ahead of a broader release.

Meta Brings AI Agents To Desktops
Meta-owned Manus launched a desktop app that moves its AI agent from cloud-only to local devices. The “My Computer” feature lets it access files, apps, and perform tasks directly on users’ machines. The shift mirrors rising interest in local agents like OpenClaw. It expands capabilities but raises security concerns over device access.

Coinbase Pushes Stablecoin for AI Payments (Paywall)
Coinbase is pursuing a deal with Cloudflare to issue a stablecoin for AI agent payments. It is also building wallets, marketplaces, and its x402 protocol to support agent transactions as trading revenue declines. Micropayments averaging 31 cents point to a new use case, though competition from Stripe and others is increasing.

Gamma Launches AI Design and Workflow Updates
Gamma introduced new features including Gamma Imagine, connectors, and AI-native templates. The tools let users generate visuals, integrate with apps like ChatGPT, and turn prompts into decks. The update expands automation across design and workflows. Features are available now, with expanded access for 30 days.

Amazon CEO Andy Jassy told employees that Amazon Web Services (AWS) could reach $600 billion in annual revenue within a decade, doubling his prior $300 billion projection, according to remarks reported March 18, 2026. The revised outlook reflects Amazon’s growing confidence that AI-driven demand will significantly expand cloud usage.

To support that growth, the company plans roughly $200 billion in capital expenditures this year, largely for AI infrastructure such as data centers and chips—despite investor concerns that have weighed on its stock. AWS generated $128.7 billion in 2025, meaning Jassy’s target would require sustained annual growth of about 17% amid strong competition from Microsoft and Google. → Continue reading here.

NVIDIA's annual GTC conference revealed a significant strategic move: NVIDIA is integrating technology from AI chip startup Groq into its GPU systems to handle specialized inference tasks like coding. This is a notable admission that GPUs alone aren't sufficient for some of today's most demanding AI workloads — a striking reversal given Jensen Huang's dismissive comments about Groq just months ago. Potential customers for the combined system include OpenAI, Anthropic, Meta, xAI, and cloud partners like Google, Oracle, and Amazon. The deal made sense for both sides: NVIDIA wanted to lock up Groq's talent and IP before competitors like AMD or OpenAI could, while Groq needed NVIDIA's massive-memory GPUs to stay relevant as language models keep growing.

GTC also surfaced the ongoing tension between NVIDIA and the big cloud providers — Microsoft, Google, and Amazon — who collectively account for 60% of NVIDIA's revenue but are also building competing chip architectures like Google's TPUs. Huang framed NVIDIA as the key orchestrator deciding which clouds GPU customers land on, but Google was simultaneously running TPU customer meetings a few blocks from the conference to stay within event rules. Separately, OpenAI is pivoting back toward enterprise after Anthropic's coding tools helped close the revenue gap between the two companies, and Meta committed $27B to Nebius data centers as part of a broader wave of AI infrastructure deals.→ Read the full article here. (Paywall)

Anthropic now accounts for over 73% of first-time AI tool spending by companies, according to Ramp data cited by Axios. The shift marks a sharp reversal from just 10 weeks earlier, when spending was evenly split with OpenAI—and from early December, when OpenAI led 60/40. The data suggests the competitive edge in AI is moving from model performance to enterprise monetization, where Anthropic is gaining traction.

Meanwhile, reports indicate OpenAI may pivot toward enterprise customers as it continues subsidizing consumer usage despite projecting $25 billion in annual revenue, compared with Anthropic’s $19 billion. Executives remain cautious, with many large firms opting to use multiple models amid rapid technological change. → Read the full article here.

Robot “dogs” from Boston Dynamics and Ghost Robotics are increasingly being deployed to patrol and inspect AI data centers, as operators scale infrastructure to meet surging demand. In a March 16, 2026 report, executives said interest has spiked alongside a global buildout that includes roughly 35 gigawatts of capacity under construction in North America alone.

The robots—priced between about $165,000 and $300,000—can monitor equipment, detect anomalies, and patrol perimeters, with companies reporting payback periods of roughly 18–24 months. Rather than replacing guards, firms position them as force multipliers that provide continuous, mobile surveillance across massive sites. → Read the full article here.

Mysterious AI Model Fuels Speculation: “Hunter Alpha” surfaced March 11 with 1T parameters and 1M-token context, resembling DeepSeek’s rumored V4 and hinting at a stealth prelaunch test.

Tencent Revenue Tops Forecasts: 2025 revenue hit 751.8B yuan, beating estimates, as Tencent boosted AI spending to 18B yuan and plans to double it, signaling deeper investment in cloud and AI growth.

Vercel Launches AI Agent Plugin: New CLI tool adds 47+ skills, sub-agents, and dynamic context management, helping coding agents like Claude Code and Cursor build, deploy, and optimize apps on Vercel.

Patreon CEO Slams AI Fair Use: Jack Conte said AI firms’ training claims are “bogus,” arguing creators deserve pay as companies strike major licensing deals yet use broader content unpaid.

NVIDIA Wins China Chip Approval: Beijing cleared H200 AI chip sales, enabling NVIDIA to resume production and orders, while adapting Groq chips for China to regain a market once driving 13% of revenue.

Thanks for reading. See you next time!

— Matthew Berman, Nick Wentz & the Forward Future Team