• AI Rundown
  • Posts
  • The AI Rundown by Lightscape Partners - 5/4/2026

The AI Rundown by Lightscape Partners - 5/4/2026

Anthropic eyes a $900B valuation in days, the Musk v. Altman trial reshapes the OpenAI origin story, and AWS officially welcomes OpenAI to its cloud as Microsoft rewrites the partnership.

Author

Ted Wagner
May 04, 2026

Image Credits: Krisztian Bocsi/Bloomberg / Getty Images

Good morning, and welcome back to another week of the AI Rundown by Lightscape Partners. This week was a reminder that AI's center of gravity is still moving, fast.

Three stories worth your time at the top:

  • Anthropic could close a $50B round at a $900B valuation in the next two weeks. Pre-emptive offers have come in between $850B–$900B, and investors were given 48 hours to confirm allocations. If it lands, it would put Anthropic on the doorstep of OpenAI's market value — and reset the funding ceiling for every other frontier lab.

  • The Musk v. Altman trial began, and Musk admitted under oath that xAI trained Grok on OpenAI models. The "distillation" admission could reshape how courts and frontier labs treat training data, model weights, and competitive copying. Beyond the testimony, the trial is dredging up the OpenAI founding story — and putting Altman, Musk, and Shivon Zilis under the same hot lights.

  • OpenAI and Microsoft revised their partnership terms, freeing OpenAI to sell on AWS — and AWS immediately rolled out OpenAI models and Codex on Bedrock. Microsoft gets more cash via revenue share and resolved its $50B Amazon-deal legal exposure; OpenAI gets distribution; AWS gets a flagship model partner. The center of cloud AI just got a lot more multipolar.

Let's get into it.

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

Anthropic

Anthropic could raise a new $50B round at a valuation of $900B Link.

  • Multiple pre-emptive term sheets have arrived in the $850B–$900B range, per sources.

  • A close at the high end would make Anthropic the second-most-valuable AI company in the world.

  • Capital is reportedly earmarked for compute, talent, and enterprise distribution as Claude becomes the agentic backbone for both Google Cloud and AWS workloads.

  • Read alongside the next story: investor enthusiasm reflects how dependent the hyperscalers' AI narrative has become on Anthropic's growth.

Anthropic's $900B+ round could happen within two weeks Link.

  • Anthropic asked investors to submit allocations within 48 hours — an unusually compressed timeline for a round this size.

  • Lead investors are not yet public, but several sovereign and crossover funds are reportedly competing.

  • The round would dwarf last year's $13B raise and put Anthropic ahead of nearly every private company globally except OpenAI and SpaceX.

  • Watch for what happens to Anthropic's enterprise pricing if the round closes — the company has been deliberately conservative about commercial monetization.

Half of Google's and Amazon's "blowout AI profits" came from a stake in Anthropic, not from their actual business Link.

  • Both Alphabet and Amazon recorded billions in unrealized gains tied to their Anthropic positions in Q1 earnings.

  • That accounting tail makes hyperscaler results look stronger — and more correlated to private-market AI sentiment — than the underlying cloud businesses suggest.

  • It also raises the stakes if Anthropic's valuation ever moves the wrong way.

The Google–Anthropic partnership built a $30B run rate on trust Link.

  • Google and Anthropic have been quietly building a tightly integrated agentic stack for three years, with safety reviews embedded into the joint roadmap.

  • Combined, they're now powering an AI run rate north of $30B across Vertex, Workspace, and Claude API.

  • Differentiation versus AWS + OpenAI: Google is leaning into governance, Anthropic into model behavior, with both betting that "trust" sells in regulated enterprise.

OpenAI & The Musk v. Altman Trial

Musk v. Altman kicks off and the trial could reshape OpenAI and the AI industry Link.

  • The lawsuit is forcing public testimony about OpenAI's founding intent, charity-to-PBC conversion, and Microsoft entanglement.

  • Outcomes could constrain OpenAI's structure or set precedent on how mission-driven AI labs convert to commercial entities.

  • Cultural sideshow: jury selection has surfaced that several jurors openly dislike Musk — a wrinkle xAI's lawyers are clearly tracking.

Elon Musk testifies that xAI trained Grok on OpenAI models Link.

  • Under oath, Musk acknowledged "distillation" of OpenAI outputs into early xAI training pipelines.

  • Frontier labs have been racing to lock down terms-of-service language and watermark outputs to prevent exactly this; the testimony will be cited in future copying disputes.

  • Expect Anthropic, OpenAI, and Google to tighten enterprise API ToS and add stronger lineage tracking to their models in response.

OpenAI ends Microsoft legal peril over its $50B Amazon deal Link.

  • OpenAI extracted concessions allowing it to sell products on AWS while Microsoft picks up additional revenue-share economics.

  • The agreement neutralizes Microsoft's contractual claim that could have blocked the AWS partnership.

  • It's the cleanest signal yet that OpenAI's growth model is moving from "Microsoft-exclusive" to multi-cloud.

OpenAI and Microsoft revise the terms of their AI partnership Link.

  • Third revision since early 2025 — each one has loosened OpenAI's exclusivity in exchange for cash flowing back to Microsoft.

  • Satya Nadella publicly framed the new deal as something Microsoft is ready to "exploit" via Azure, M365, and Copilot.

  • Strategic read: Microsoft has shifted from "owning the model" to "owning the distribution" of every major model.

AWS brings OpenAI's models and Codex programming assistant to its cloud Link.

  • OpenAI models now live in Amazon Bedrock, with Codex available as a developer assistant.

  • AWS also launched Bedrock Managed Agents to make OpenAI-powered agents easier to deploy.

  • Customers finally get a single procurement path to GPT, Claude, and Nova within one cloud — reducing one of the bigger frictions in enterprise multi-model adoption.

OpenAI announces new advanced security mode for ChatGPT accounts, including a Yubico partnership Link.

  • Hardware-key MFA, session monitoring, and device attestation now available to "at-risk" accounts.

  • Targets enterprise admins, journalists, public figures, and developers handling sensitive prompts and tools.

  • Expect the rest of the frontier labs to fast-follow with similar tiers.

Frontier Models & AI Labs

DeepMind's David Silver raised $1.1B to build an AI that learns without human data Link.

  • Silver, the architect behind AlphaGo, is now at Ineffable Intelligence and chasing reinforcement learning at frontier scale.

  • The bet: pre-training on the entire internet has hit diminishing returns, and the next leap comes from self-play and synthetic environments.

  • Backers include Lightspeed, Sequoia, Nvidia, Google, the U.K. Sovereign AI Fund, DST, and Index.

Ineffable Intelligence raises $1.1B at a $5.1B valuation to build an AI "superlearner" Link.

  • Same round, deeper read: Ineffable is hiring aggressively in London and Mountain View, with a stated focus on agents that improve from their own trajectories.

  • Lightspeed and Sequoia co-led — a bet that the post-LLM paradigm will be RL-native, not transformer-stack incremental.

The Man Behind AlphaGo Thinks AI Is Taking the Wrong Path Link.

  • Silver argues current frontier labs are over-invested in next-token prediction at the expense of grounded learning loops.

  • His bet diverges sharply from the OpenAI/Anthropic playbook — and aligns with where Demis Hassabis has been steering Google DeepMind.

Standard Intelligence raises $75M to develop efficient computer-use models Link.

  • Six-person team, Sequoia and Spark co-lead, Karpathy as angel.

  • FDM-1 is purpose-built for computer-use agents — a category where every frontier lab has been throwing big models at small interface tasks.

  • Efficiency-first thesis: compete on cost-per-action rather than benchmark scores.

Parallel Web Systems hits $2B valuation five months after its last big raise Link.

  • Parag Agrawal's startup is building a parallel web specifically optimized for AI agents (think: structured, machine-readable, agent-native).

  • Sequoia led another $100M round on top of last year's $100M.

  • Bet: the web as humans see it is too noisy for agents — agents need their own substrate.

Big Tech Earnings & Cloud

Alphabet's stock climbs as Google Cloud revenue grows 63% Link.

  • EPS came in at $5.11 vs. $2.63 expected; Cloud revenue surged 63% YoY.

  • Capex guidance is up sharply; management cited Anthropic + Vertex demand as primary drivers.

  • Search remains profitable, but Cloud is now the marginal-dollar story for the stock.

Google Cloud surpasses $20B but says growth was capacity-constrained Link.

  • Translation: Google could have grown faster if it had more GPUs and power.

  • That's a familiar refrain from AWS and Azure — Google saying it out loud is what justifies the next leg of capex.

  • Watch for Google to lean harder on TPUs to relieve the supply bottleneck.

Amazon's cloud business is surging — and so is its capital spending Link.

  • AWS beat on revenue; capital expenditures are running well above prior guidance.

  • Andy Jassy reaffirmed that the spend will continue, framing it as a multi-year buildout.

  • The OpenAI-on-Bedrock announcement amplifies AWS's customer pull at the exact moment it's lighting up new capacity.

Microsoft says it has over 20M paid Copilot users — and they're really using it Link.

  • 20M+ paid Copilot seats, with engagement metrics improving — a counter to the persistent "nobody actually uses Copilot" narrative.

  • Microsoft also revealed Accenture's 700K+ user rollout, which is now the largest enterprise Copilot deployment on record.

  • For CFOs evaluating Copilot, the question shifts from "is anyone using this?" to "how do we measure ROI?"

Big Tech will spend nearly $700B on AI this year — no one knows where the buildout ends Link.

  • Combined hyperscaler capex projections cleared $700B for 2026.

  • Investor sentiment is genuinely split: bulls see a multi-decade buildout, bears worry about ROI timing.

  • Inputs (chips, real estate, power, water) are all bottlenecked — meaning capex inflation is also a story.

Enterprise AI & Agentic Software

Stripe introduces Link, a digital wallet that autonomous AI agents can use too Link.

  • Link lets users link cards, banks, and subscriptions, then issue scoped, agent-specific spending authorizations.

  • This is the cleanest "agent-as-payments-principal" UX shipped to date.

  • Pairs naturally with OpenAI's agent stack and Anthropic's tool-use APIs.

Salesforce introduces Agentforce Operations to automate outdated back-office tasks Link.

  • Agentforce Operations targets the unsexy middle of enterprise: invoice routing, vendor onboarding, license reconciliation.

  • Strategy: extend Salesforce's record-of-truth into agentic execution rather than letting horizontal agent platforms own the workflow.

  • A direct shot at Workday and ServiceNow.

Microsoft reveals how Accenture deployed Copilot to 700,000+ users Link.

  • Largest enterprise AI rollout ever — bigger than the population of Denver.

  • Microsoft published the playbook (training, role-based prompts, governance) — a tacit recognition that Copilot adoption depends as much on change management as on the model.

The Bloomberg Terminal is getting an AI makeover, like it or not Link.

  • Bloomberg's CTO walked WIRED through a chatbot-style interface coming to the Terminal — the iconic financial data product is trying to graft an LLM front door without alienating muscle-memory users.

  • High-stakes case study for any incumbent retrofitting a power-user tool with AI.

Startup Funding & Valuations

Rogo raises $160M to speed up financial analysis with AI agents Link.

  • Series D led by Kleiner Perkins; Sequoia, Khosla, and Morgan Growth Equity also in.

  • Rogo automates the recurring grunt work of financial analysts — pitch books, comps, earnings prep.

  • One of the most competitive enterprise AI verticals: see also Hebbia, Brightwave, and Bloomberg's own AI push above.

Legal AI startup Legora hits $5.6B valuation as its battle with Harvey gets hotter Link.

  • Legora and Harvey are now running dueling ad campaigns, pushing into each other's home turf (US vs. EMEA).

  • Both companies have raised massive rounds; legal AI is consolidating into a two-horse race at the top.

Nvidia's NVentures backs $150M round for medical AI startup Aidoc Link.

  • Goldman Sachs led; SoftBank Investment Advisors and General Catalyst joined alongside Nvidia.

  • Aidoc's diagnostic AI runs across hundreds of hospitals — the round is fuel for international expansion.

Coby Adcock's Scout AI raises $100M to train its models for war Link.

  • Defense AI startup is building agents that let individual soldiers control fleets of autonomous vehicles.

  • Defense tech keeps capturing growth-stage capital faster than commercial agentic startups.

Firestorm Labs raises $82M to take drone factories into the field Link.

  • Mobile drone manufacturing — drones produced where they're flown.

  • Pairs with the broader Pentagon AI push: distributed manufacturing is a national-security thesis as much as a commercial one.

JuliaHub lands $65M to automate the design and testing of industrial products Link.

  • Agentic engineering for hard-tech: simulation, optimization, design verification.

  • Industrial AI quietly becoming a real category — see also Standard Bots, Formic, Articul8 in the portfolio jobs section below.

Healthcare + Life Sciences

In a Harvard study, AI offered more accurate diagnoses than emergency-room doctors Link.

  • LLMs outperformed clinicians on a meaningful subset of real ER cases.

  • The study won't change clinical practice tomorrow, but it strengthens the case for AI as a default second-read in triage.

  • Pair with Reid Hoffman's WIRED op-ed this week arguing doctors should ask AI for second opinions — a position that's quickly moving from edgy to mainstream.

From six months to four days: how AI slashed wait times for essential autism care Link.

  • Acclaim Autism used AI workflow automation (Appian) to compress intake and prior-auth from 6 months to 4 days.

  • A clean example of AI's near-term value: not replacing clinicians, but eliminating administrative bottlenecks that were starving access.

FDA approval, fundraising, and the reality of building in healthcare — a BioticsAI founder's take Link.

  • Healthcare AI startups are running into a familiar wall: model performance is no longer the bottleneck — regulatory clearance and hospital procurement are.

  • Worth reading for any founder weighing healthcare versus other agentic verticals.

Workforce + Labor

A Chinese court ruled firms can't lay off workers on AI grounds Link.

  • An eastern Chinese tech firm illegally fired an employee who refused a demotion after his job was automated.

  • First major court ruling globally to set a "AI-replacement is not just cause" precedent.

  • Expect plaintiffs in the US and EU to cite this as labor disputes accelerate.

Sam Altman confirms some companies are "AI-washing" by blaming unrelated layoffs on the technology Link.

  • Altman publicly conceded what economists have been saying for months: there's no clean signal of AI-driven displacement in the labor data yet.

  • He still expects displacement to land — but is calling out execs who use "AI" as cover for cost cuts unrelated to the technology.

Microsoft researchers reveal the 40 jobs most exposed to AI — and even teachers make the list Link.

  • The list is heavily weighted toward knowledge work: writers, analysts, customer support, developers, and yes, teachers.

  • Expect this to become required reading inside HR organizations and B-school career offices.

Snap CEO praises AI for writing two-thirds of the company's code — but warns of "societal pushback" Link.

  • Spiegel said only 26% of Americans have a favorable view of AI; tech leaders are underestimating the political tail risk.

  • Tension worth watching: at the same time productivity benefits accrue inside companies, public trust in AI is moving the wrong direction.

Hardware, Data Centers, & Energy

SoftBank reportedly plans to spin off a $100B data-center construction venture Link.

  • The new entity, "Roze," would IPO at ~$100B by year-end if reports hold.

  • Translation: Masa is creating a public vehicle for the AI infrastructure trade — a way for retail to participate in hyperscaler buildout without owning the hyperscalers.

As Tim Cook steps down, Apple hits record sales — but a chip shortage looms Link.

  • Apple posted $111B in revenue and is handing the keys to John Ternus.

  • Mac demand surprised Apple itself — buyers are choosing Macs as on-device AI workstations.

  • Mac Mini's starting price was bumped to $799 as the chip supply tightens.

Meta inks a deal for solar power at night, beamed from space Link.

  • A small but symbolic contract with Overview Energy for space-based solar.

  • Hyperscalers are now experimenting with energy sources that didn't exist commercially five years ago — a sign of how acute the AI power crunch has become.

Apple raises Mac Mini's starting price to $799 after AI frenzy drains supply Link.

  • Apple is openly framing the Mac Mini as a local-AI development workstation.

  • The price hike is the first M-series price increase in Apple's history.

Policy + Regulation + Geopolitics

Pentagon inks AI procurement deals with seven companies — Anthropic notably absent Link.

  • AWS, Google, Microsoft, Nvidia, OpenAI, SpaceX, and Reflection AI are in.

  • Anthropic was excluded after refusing to allow domestic mass surveillance and autonomous-weapons use.

  • A clean values-vs-revenue divergence between Anthropic and the rest of the frontier.

Google expands Pentagon's access to its AI after Anthropic's refusal Link.

  • Google moved to fill the gap Anthropic left, signing a new contract with the DoD.

  • Reopens the internal debate that ended Project Maven in 2018.

Hundreds of Google employees sign letter urging CEO to reject US military AI use Link.

  • ~600 employees signed; the letter focuses on classified-setting deployments.

  • Google's response will set a tone for how AI labs handle internal dissent on defense work going forward.

China vetoes Meta's $2B Manus deal after a months-long probe Link.

  • Beijing ordered Meta to unwind the Manus acquisition, blocking a marquee AI-agent deal.

  • Pattern: cross-border AI M&A is now subject to bilateral political risk that didn't exist 18 months ago.

AI-generated actors and scripts are now ineligible for Oscars Link.

  • Academy formalized the rule after the "Tilly Norwood" debate last fall.

  • Symbolic but consequential — sets a creative-industry precedent that other awards bodies will follow.

Lightscape Portfolio is Hiring

Our portfolio companies have 573+ open roles across 18 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].