• AI Rundown
  • Posts
  • The AI Rundown by Lightscape Partners - 5/18/2026

The AI Rundown by Lightscape Partners - 5/18/2026

Cerebras stages the year's biggest IPO, Anthropic edges past OpenAI in business customers, and Sam Altman finally takes the stand in Musk v. OpenAI.

Author

Ted Wagner
May 18, 2026

Image credit: Bloomberg | Getty Images

It was a week where the money, the courtroom, and the energy grid all collided. Cerebras priced, twice, and still doubled on day one, dragging the rest of 2026's IPO pipeline behind it. Anthropic, the perennial #2, slipped past OpenAI in verified business customers even as it broadened from the Fortune 500 down to SMBs and law firms. And in San Francisco, Sam Altman spent the week explaining to a jury, and to history, exactly how he wrestled OpenAI away from Elon Musk. Underneath it all, the data center build-out kept brushing up against physical reality: gas turbines in Mississippi, transformer shortages in Tahoe, and a $100B plan that has Utah residents in open revolt.

  • Cerebras prints money. The Sunnyvale chip company raised $5.5B in one of the largest U.S. tech IPOs in years and saw shares double on debut, with CEO Andrew Feldman calling AI compute demand "not speculative." Wall Street took the cue: 2026's AI IPO window is officially open.

  • Anthropic passes OpenAI in business count. New Ramp data shows Anthropic now has more verified enterprise customers than OpenAI for the first time — and Anthropic isn't slowing down, with PwC committing to train 30,000 staff on Claude, a new Claude for Small Business tier, and a legal-services suite aimed squarely at the Clio/Harvey corridor.

  • Altman takes the stand. In the most consequential tech trial of the year, Sam Altman testified that Musk repeatedly tried to take control of OpenAI — including a "hair-raising" suggestion that he hand it to his own children. Ilya Sutskever, returning to defend his former employer, said he still stands by the 2023 board ouster: "I didn't want it to be destroyed."

If you haven't yet, please support the newsletter by subscribing!

Anthropic

Anthropic passes OpenAI in business customers, per Ramp Link.

  • Ramp's monthly AI Index shows Anthropic edging ahead of OpenAI in verified business customers for the first time.

  • The shift is being driven by Claude Code adoption among engineering teams and a series of new vertical pushes.

  • It's the clearest data point yet that the enterprise AI market is no longer winner-take-all.

  • The lead is narrow, but the trajectory matters: a year ago Anthropic was barely on the chart.

PwC expands Anthropic alliance, will train 30,000 staff on Claude Link.

  • PwC will roll out Claude Code and Claude Cowork across its U.S. workforce.

  • 30,000 professionals will be trained directly on the Claude model family.

  • The deal is one of the largest enterprise Claude commitments to date and a real Big Four endorsement.

  • Anthropic is increasingly leaning on consulting partners — PwC, Deloitte, Accenture — as a distribution channel.

Anthropic launches Claude for Small Business Link.

  • A new SMB tier bundles Claude with prebuilt automation workflows for everyday operations.

  • It's the fifth market-specific Claude bundle Anthropic has shipped this year.

  • Pitch: take the Claude Code/agentic toolchain and translate it for non-engineers running 5–100 person companies.

  • A clear shot at OpenAI's ChatGPT Business and Microsoft's Copilot SMB push.

Anthropic moves into legal services as Clio crosses $500M ARR Link.

  • Anthropic is launching a dedicated legal-services suite to compete with Harvey, Clio, Eve, and Spellbook.

  • The move comes just as Clio reports $500M in ARR, underscoring how fast legal tech is being repriced by AI.

  • Anthropic's pitch: Claude's longer-context, more cautious behavior is a natural fit for legal review.

  • Expect a wave of incumbent legal-tech vendors to pick a model partner or get squeezed.

OpenAI

Sam Altman testifies that Musk wanted control of OpenAI Link.

  • Altman told the jury that Musk repeatedly tried to take majority control of OpenAI before walking away in 2018.

  • The trial centers on whether Altman and Brockman breached founding agreements when OpenAI converted to a capped-profit entity.

  • Altman testified that Musk floated "hair-raising" ideas, including handing OpenAI's IP to his own children.

  • Ilya Sutskever, testifying for OpenAI, said he still stands by the 2023 board ouster of Altman: he didn't want OpenAI "to be destroyed."

OpenAI launches a $4B professional-services business Link.

  • The new OpenAI Deployment Company will help enterprises adopt OpenAI's models with hands-on integration support.

  • The unit launches with $4B in funding and a long list of partner firms.

  • It's an explicit shot at the IBM/Accenture/Deloitte playbook — and at Anthropic's PwC alliance.

  • A clear admission that selling raw API access isn't enough to win the enterprise.

ChatGPT goes after your bank account Link.

  • ChatGPT Pro users in the U.S. can now connect bank accounts and get personalized financial guidance from GPT.

  • The feature is in limited preview but signals OpenAI's move into the Intuit/Plaid/Rocket Money territory.

  • Combined with shopping, scheduling, and Codex, ChatGPT is steadily morphing into a personal operating system.

  • Regulators will have opinions; expect a CFPB conversation within weeks.

Codex comes to mobile Link.

  • OpenAI's Codex coding assistant is now available in ChatGPT's iOS and Android apps.

  • The release lands about eight months after Anthropic shipped Claude Code on mobile.

  • New customization features let developers tune Codex to their stack and style.

  • Asynchronous, phone-first coding workflows are quickly becoming the norm.

Frontier Models

Thinking Machines drops a real-time, "always listening" model Link.

  • Mira Murati's Thinking Machines Lab released a research preview of a model that processes input and generates a response simultaneously.

  • The architecture aims to kill the awkward "turn-based" rhythm of today's chatbots.

  • Murati's positioning: a more human-feeling assistant that "keeps humans in the loop."

  • Murati's broader product strategy emphasizes safety and interruptibility — a clear differentiation from the OpenAI/Anthropic frontier race.

Recursive Superintelligence comes out of stealth with $650M Link.

  • The new lab launched today with $650M from GV and Greycroft, with Nvidia and AMD as strategic backers.

  • The pitch is literally on the tin: self-improving AI models.

  • Adds to a growing pile of well-funded frontier labs (Thinking Machines, SSI, xAI, Perplexity) chasing post-LLM architectures.

  • Expect more "frontier-only" labs in 2026 as the gap between best-in-class and good-enough widens again.

Hardware + Infrastructure

Cerebras raises $5.5B in a blockbuster IPO Link.

  • Cerebras raised its offering price twice and still saw shares rise 68% on day one, closing roughly double the IPO price.

  • It's one of the largest U.S. tech IPOs in years and the de facto stress test for the AI listing pipeline.

  • CEO Andrew Feldman: chip demand is "not speculative" — Cerebras is sold out through 2027.

  • Bankers are now reportedly accelerating timelines for Databricks, Anthropic-backed CoreWeave peers, and a string of model-layer companies.

Cisco's AI orders forecast hits $9B; stock surges Link.

  • Cisco posted record quarterly revenue and forecast $9B in AI-related orders, well above Street expectations.

  • The stock hit an all-time high on the print.

  • Cisco also announced ~4,000 job cuts to redirect spend into AI engineering and silicon photonics.

  • The story is now clearly that the networking incumbents — Cisco, Arista, Juniper — are AI plays, not telecom plays.

CUDA is why Nvidia is really a software company Link.

  • Wired's deep dive on CUDA argues Nvidia's real moat is its software stack, not the silicon.

  • Two decades of CUDA tooling, libraries, and developer relationships are the real switching cost.

  • AMD and Intel can match flops, but matching the developer ecosystem is a multi-year project at best.

  • Worth a careful read for anyone shorting Nvidia on "the AMD MI400 will catch up" — that's necessary but not sufficient.

Wall Street is calling memory AI's golden ticket; one Harvard prof says watch out Link.

  • HBM and DRAM stocks have ripped on AI training and inference demand.

  • Harvard's Willy Shih: "curves that just go to the sky with no end never continue forever."

  • DRAM shortages have already pushed prices to multi-year highs; the cycle's late-innings risk is real.

  • A reminder that the AI infrastructure trade has cyclical-commodity exposure baked in.

Data Centers + Energy

Cowboy Space raises $275M to put data centers in orbit Link.

  • Cowboy Space's Series B values the company at $2B; Index, NEA, and IVP led.

  • The thesis: orbit gets you free cooling, near-unlimited solar, and no NIMBYs.

  • The catch — there aren't enough rockets, so Cowboy is also building its own launch capacity.

  • Cool-factor aside, the unit economics still trail terrestrial data centers by several multiples.

Google and SpaceX in talks for orbital data centers Link.

  • Reported talks between Google Cloud and SpaceX would put racks of TPUs in low-earth orbit.

  • SpaceX would handle launch and Starlink-style downlink to ground.

  • It's a leading indicator that hyperscalers are serious about decoupling from the U.S. grid.

  • Combined with the Cowboy Space round, "space data center" is no longer a meme category.

xAI runs 50 unchecked gas turbines in Mississippi Link.

  • xAI's Colossus 2 site is using nearly 50 "mobile" gas turbines as primary power, the subject of an active lawsuit.

  • Air-quality regulators argue the turbines are functioning as a power plant and should be permitted as one.

  • Musk's compute build-out is sprinting ahead of every interconnection and emissions queue.

  • A bellwether for how AI's energy hunger collides with existing environmental law.

Kevin O'Leary's $100B Utah data center sparks resident revolt Link.

  • O'Leary is pitching a 200-acre, 1.5GW campus in rural Utah; locals are organizing against it.

  • He's countered that Chinese-funded misinformation is fueling the opposition.

  • Whatever the truth, the bigger story is that data-center NIMBY-ism is now a national political issue.

  • A new Fortune survey found Americans would rather live next to a nuclear plant than a data center.

Enterprise AI / Cloud

Red Hat unveils a new enterprise AI platform Link.

  • Red Hat's AI Summit announcements span new inference, automation, and sovereignty features.

  • The strategy: lean on open source to beat hyperscalers on AI economics and control.

  • Red Hat also pushed a sovereign-AI stack aimed at EU/government buyers.

  • Combined with its AMD partnership, Red Hat is positioning as the "agnostic AI substrate."

IBM bets on trust and control as the production test Link.

  • IBM Think laid out a governed-AI strategy: bring automation, trusted data, and operational control to enterprise workloads.

  • Big push on watsonx, Granite models, and a new Agentic Studio.

  • IBM is leaning into the regulated-industry story (healthcare, finance, government) where governance is the gating constraint.

  • Don't sleep on IBM — Project Cube and the consulting pull-through are real.

Notion turns its workspace into an agent hub Link.

  • Notion's new developer platform lets teams plug agents, external data, and custom code directly into pages.

  • The bet is that agentic work happens where the documents already live.

  • Notion now sits in the same conversation as Asana, Linear, and ClickUp on "the agentic work surface."

  • Watch how this interacts with Anthropic's new Cowork product and Microsoft's Copilot Studio.

Startup Funding & Valuations

Vapi hits a $500M valuation after winning Amazon Ring Link.

  • Voice-AI infrastructure startup Vapi raised $50M and hit a $500M valuation.

  • Vapi won Amazon Ring's customer-call AI mandate over 40 competing vendors.

  • Vapi says enterprise revenue is up 10x since early 2025 as call centers swap humans for agents.

  • The voice-agents tier is consolidating fast around Vapi, ElevenLabs, and Hume.

White Circle raises $11M to keep AI models from going rogue Link.

  • Paris-based White Circle (legally Pumpkin Intelligence) builds real-time guardrails for production AI.

  • Backers include leadership from OpenAI, Anthropic, DeepMind, Mistral, and Hugging Face.

  • The founder's calling card: a public demo breaking widely deployed enterprise guardrails.

  • AI safety as a B2B SaaS category is starting to look real.

Origin Lab raises $8M to monetize game-engine data Link.

  • Origin Lab is building a marketplace where game studios sell training data to world-model labs.

  • Bridges two structurally adjacent markets: high-fidelity simulated worlds and frontier model training.

  • Picks and shovels for the next wave of world models (Genie 3, V-JEPA, etc.).

  • Expect more "licensed data" startups as labs run out of clean public data.

Workforce + Labor

GM lays off hundreds of IT workers to hire stronger AI talent Link.

  • The cuts target traditional IT and infrastructure roles.

  • New hires will focus on AI-native development, agent and model development, and prompt/workflow engineering.

  • A clear example of the "rebalance, not reduce" pattern playing out at every large enterprise.

  • Expect more Fortune 500 reorgs framed the same way through Q3.

Microsoft AI chief: 18 months until all white-collar work is automated Link.

  • Mustafa Suleyman, leading Microsoft AI, predicts an 18-month window for full automation of white-collar tasks.

  • The framing is provocative and obviously aspirational — but it puts Microsoft's product roadmap in plain view.

  • Expect Copilot, Windows agents, and Microsoft 365 Chat to push hard on agentic task execution all year.

  • A nice contrast with Goldman's Blankfein this week, who explicitly said Goldman doesn't trust AI agents yet.

Policy, Safety + Cybersecurity

Google catches criminals using AI to build a working zero-day Link.

  • Google's Threat Intelligence Group reported the first confirmed case of attackers using AI to develop a working zero-day exploit.

  • The targets included enterprise software vulnerabilities discovered and weaponized end-to-end with LLMs.

  • Defenders' window of advantage just got shorter.

  • Expect a wave of "AI for offensive security" funding rounds — and a corresponding wave of detection startups.

Yoshua Bengio: humanity risks extinction within a decade Link.

  • Bengio warned that hyperintelligent systems with their own "preservation goals" could pose existential risk within 10 years.

  • His specific concern: AI's growing ability to persuade and manipulate humans at scale.

  • Bengio is the most-cited AI researcher alive; this is not a fringe view among the founding generation.

  • He continues to push for international compute-governance frameworks.

Wells Fargo calls AI a "euphoric" bubble — and a buy anyway Link.

  • Wells Fargo equity strategist Ohsung Kwon called AI an "euphoric" bubble that investors should ride.

  • The argument: in a market this concentrated, sitting out the AI trade is the bigger career risk.

  • Echoes recent Goldman, Morgan Stanley, and JPM notes that have all flagged elevated valuations but recommend staying long.

  • The classic "we know it ends badly, but not yet" setup.

Healthcare + Life Sciences

Medicare's new payment model is built for AI agents Link.

  • CMS's new ACCESS model creates the first reimbursement path for AI-driven patient monitoring, outreach, and care coordination.

  • This is a massive structural change: until now, there was no way to bill for an agent between visits.

  • Early winners will be care-coordination platforms, RPM vendors, and AI-native primary care.

  • Tech investors are largely sleeping on it — a structural tailwind hiding in plain sight for the digital health stack.

Lightscape Portfolio is Hiring

Our portfolio companies have 581+ open roles across 19 companies right now. A few featured opportunities this week:

Know someone who'd be a great fit? Forward this newsletter or send them straight to the role.

Thank you for reading The AI Rundown by Lightscape Partners. Please send any questions, comments, or suggestions to [email protected].