Agentic Brew Daily

Cerebras priced its Nasdaq debut at $185 a share — above the $150-$160 marketed range — and sold 30M Class A shares for $5.55B. The ticker CBRS started trading May 14 at a fully diluted valuation of about $56.4B, more than double the $26.6B implied just five weeks earlier. Demand exceeded shares by more than 20x, making it the largest US tech IPO since Snowflake's $3.8B debut in 2020. The whole pitch is anchored on a $20B+ master relationship agreement with OpenAI for 750 MW of inference capacity, scaling toward 2 GW by 2030.

Why it matters: This is the opening salvo of an AI IPO wave (OpenAI and SpaceX are rumored next). If CBRS trades cleanly, the cost of capital drops across AI infra; if it breaks issue, the pipeline thins out fast. The wafer-scale chip story is a real shot at "end of GPU homogeneity" in inference — but P/S north of 110x means the market is paying tomorrow's price today.

Ramp's April 2026 AI Index put Anthropic at 34.4% of paying businesses versus OpenAI's 32.3% — the first crossover ever. Anthropic gained 3.8 points month-over-month while OpenAI lost 2.9, and Anthropic has quadrupled in a year while OpenAI added 0.3 points. OpenAI's chief application officer Fidji Simo told staff the company is in "code red" mode. Claude Code is the wedge — Anthropic's fastest-growing product ever — and a freshly expanded PwC alliance will certify 30,000 professionals on Claude.

Why it matters: This is a year-long slope, not a one-month blip. Claude Code is the Trojan horse into finance (Citadel, BNY, FIS, Mizuho), legal (Freshfields, Quinn Emanuel, Holland & Knight), and consulting (PwC). The enterprise AI competitive map has been quietly re-rated.

The Musk-vs-Altman trial finished its liability phase in Oakland after 11 days. Musk is seeking $134B in damages, the removal of Sam Altman and Greg Brockman, and an unwind of OpenAI's for-profit conversion. The witness list reads like an AI history podcast — Altman, Brockman, Sutskever, Murati, Helen Toner, Shivon Zilis, Joshua Achiam, and Satya Nadella. Sutskever, Murati, and Toner all testified about a "consistent pattern of lying" memo that preceded Altman's brief 2023 ouster. Judge Yvonne Gonzalez Rogers retains discretion to overturn the jury verdict.

Why it matters: If the charitable-trust framing sticks, every mission-driven nonprofit-turned-for-profit suddenly has legal exposure. Nadella's testimony also confirmed Microsoft is rewiring its AI dependence toward multi-vendor independence — Azure has been hosting xAI since 2024. Sworn allegations of dishonesty are an obvious overhang on any near-term OpenAI IPO.

Cisco reported Q3 FY2026 revenue of $15.84B (+12% YoY) and non-GAAP EPS of $1.06. The headline: it doubled its full-year AI infrastructure order forecast from $5B to $9B, with $5.3B already booked YTD and networking product orders up over 50% YoY. The company is cutting fewer than 4,000 jobs (<5% of workforce) with up to $1B in pre-tax restructuring charges, and the stock surged ~14% — its best single-day move in more than 20 years.

Why it matters: Investors rewarded the layoffs because the math told a coherent silicon-thesis story: every freed-up dollar is being redeployed into the bucket whose order book just doubled. Cisco's Silicon One inside Nvidia Spectrum-X is the wedge into hyperscale AI Ethernet long held by Broadcom and Arista. As CEO Chuck Robbins put it: "If you don't have silicon you're going to struggle to be relevant to the hyperscalers."

Notion launched its Developer Platform on May 13: Workers (a hosted runtime on Vercel Sandbox for deploying custom code with zero server provisioning), Database Sync in beta, an External Agents API in alpha that lets Claude Code, Cursor, Codex, and Decagon operate inside Notion as native workspace participants, and a new CLI called ntn. Workers are free during public beta, with credit-based billing kicking in August 11. Customers have already built more than a million agents in the three months since Custom Agents shipped in February.

Why it matters: This reframes the workspace as the AI agent orchestration layer and collapses the Zapier/Pipedream middleware tax into one sandbox deploy. It also lowers switching costs across coding agents while raising the stakes for glue vendors. If Notion executes, it goes head-to-head with Microsoft Power Platform.

Meta launched Incognito Chat for WhatsApp and the standalone Meta AI app on May 13, calling it "the first major AI product where there is no log of conversations stored on servers." Inference runs inside a Trusted Execution Environment built on AMD SEV-SNP confidential VMs and NVIDIA H100s in confidential computing mode, fronted by Oblivious HTTP relays with remote attestation. Conversations vanish when the app closes or the phone locks. Text-only at launch, with voice, image, and a branching "Side Chat" feature signaled for later.

Why it matters: This shifts AI privacy from retention policy to hardware guarantee — a structural threat to OpenAI, Gemini, and Claude's policy-based "temporary chat." It also lands during OpenAI lawsuits where preserved logs are being used as evidence. The cheapest log to defend is the one that never existed.