- AI Rundown
- Posts
- The AI Rundown by Lightscape Partners - 7/20/2026
The AI Rundown by Lightscape Partners - 7/20/2026
Apple takes OpenAI to court, China's Kimi K3 rattles the frontier, and Databricks hits a $188B valuation

Image credit: Moonshot AI
Welcome back to The AI Rundown.
This week had a little bit of everything: a blockbuster trade-secrets lawsuit between two of the most valuable companies in tech, a Chinese open-weights model that has Silicon Valley genuinely rattled, the largest private-market valuation ever for a data-and-AI company, and the first statewide data center moratorium in the U.S. Meanwhile, Anthropic is reportedly shopping for $10 billion of Meta's data center capacity, Thinking Machines finally shipped its first model, and IBM suffered the worst stock crash in its history. Let's get into it.
Apple vs. OpenAI goes nuclear. Apple's trade-secrets lawsuit against OpenAI reads more like a spy novel than a legal filing — allegations of secret moles, stolen laptops, and recruiting-as-espionage, reaching all the way up to OpenAI's chief hardware officer. OpenAI is pushing back hard, calling the suit meritless. With OpenAI's IPO looming and its first hardware device — a screenless, moving smart speaker designed with Jony Ive — at the center of the dispute, the stakes could not be higher.
China's Moonshot throws down the gauntlet. Moonshot AI announced Kimi K3, the world's largest open-weights model at a reported 2-3 trillion parameters, and early expectations are that it closes the gap with Anthropic's Opus 4.8. Combined with Alibaba's preview of Qwen3.8 — which the company claims trails only Claude Fable 5 — the frontier race is no longer a U.S.-only affair. Businesses are already experimenting with cheaper Chinese models as U.S. rivals get more expensive.
Databricks hits $188B. Databricks is finalizing a new funding round at a $188 billion valuation, extending its run as AI's favorite second act. The company has remade itself into an AI company, publishing research on the cost savings of open-weights models for coding and riding enterprise AI adoption to one of the largest private valuations in history.
That's just the start — below you'll find the full rundown across models, infrastructure, funding, policy, and the workforce. If you haven't yet, please support the newsletter by subscribing!
Apple vs. OpenAI
The wildest allegations in Apple's trade secrets lawsuit against OpenAI Link.
Apple's complaint alleges a pattern of misconduct ranging from employees joking about unauthorized access to Apple's systems to job candidates being asked to bring Apple materials with them.
Reported claims include stolen laptops, data breaches, secret moles inside Apple, and recruiting used as a form of espionage.
The alleged misconduct reaches all the way up to OpenAI's chief hardware officer.
The suit lands at a delicate moment, with OpenAI reportedly preparing for an IPO.
OpenAI pushes back on Apple trade secret lawsuit Link.
OpenAI issued a new statement suggesting the lawsuit lacks merit.
The company frames the suit as an attempt by Apple to slow a hardware competitor rather than protect genuine trade secrets.
Analysts are already gaming out how discovery and a potential injunction could disrupt OpenAI's IPO timeline.
Sam Altman and Elon Musk spent the week trading accusations of investor deception as the drama spilled onto social media.
OpenAI's first hardware device is reportedly a screenless speaker that can move Link.
The Jony Ive-designed device would use ChatGPT to answer questions and is intended to feel "human-like" and alive.
Apple claims the concept is a stolen idea — the device sits at the heart of its trade-secrets complaint.
OpenAI also shipped a $230 keyboard for Codex this week, signaling a broader hardware ambition despite the legal battle.
Frontier Models
Thinking Machines amps up its bet against one-size-fits-all AI with its first open model, Inkling Link.
Mira Murati's Thinking Machines Lab released Inkling, its first foundation model, with full open weights available for developers to fine-tune.
It's the company's first public proof point after roughly a year and a half of building AI infrastructure largely out of public view.
The release positions the lab against one-size-fits-all frontier models, betting that customization wins in the enterprise.
China's Moonshot throws down the gauntlet with Kimi K3, the world's largest open-weights model Link.
Kimi K3 is reported to carry a parameter count between 2 trillion and 3 trillion, making it the largest open model out of China.
The FT reports the model is expected to close the gap with Anthropic's Opus 4.8.
Founder Yang Zhilin, a Carnegie Mellon-trained researcher (and Pink Floyd fan), is being credited with giving Claude and ChatGPT their own "DeepSeek moment" — a model just as good at half the price.
Businesses are increasingly experimenting with cheaper Chinese models as U.S. model prices climb.
Alibaba previews Qwen3.8, claims it's second only to Claude Fable 5 Link.
Alibaba previewed the most powerful model in its Qwen family, claiming it trails only Anthropic's Claude Fable 5 among the world's frontier models.
The preview caps a week of aggressive moves by Chinese labs to contest the frontier.
Apple Intelligence was also approved for launch in China this week, powered in part by Alibaba's Qwen — a major distribution win.
Anthropic
Anthropic, Meta reportedly discussing $10B data center leasing deal Link.
The New York Times cited three sources saying Anthropic is seeking to lease Meta data center capacity in a deal worth roughly $10 billion.
The arrangement would be one of the largest compute-leasing deals ever struck between two AI rivals.
It underscores how compute scarcity is pushing labs toward unconventional partnerships.
Anthropic, Blackstone bet the next trillion-dollar AI business is implementation, not models Link.
Anthropic-backed Ode launched to embed forward-deployed engineers inside enterprises.
The bet: the bottleneck in enterprise AI is no longer model quality but implementation capacity.
Blackstone's involvement signals that private equity sees services-led AI adoption as a trillion-dollar category.
Here's why Anthropic is pushing states to regulate AI faster Link.
Anthropic endorsed landmark AI transparency laws in California and New York last year.
The company's head of U.S. state and local policy now says those laws may already be outdated.
Anthropic also began localizing Claude pricing for India — its biggest market after the U.S. — with rupee-denominated plans.
Enterprise AI + Big Tech
Microsoft is reportedly training salespeople to talk down OpenAI and Anthropic Link.
Microsoft is positioning its in-house models as more efficient and cost-effective than partners-turned-rivals OpenAI and Anthropic.
The move marks a new phase in the unraveling of the Microsoft-OpenAI alliance.
Big Tech executives spent the week warning enterprises they may be giving away too much data and leverage to AI labs.
Satya Nadella has issued a shocking warning to companies using AI Link.
In a surprise blog post, Microsoft's CEO warned enterprises about the dangers of over-reliance on proprietary frontier models.
The post dovetails with Microsoft's sales push for its own in-house models.
Meanwhile, Intel said it will lean on Google's Gemini to help automate and accelerate silicon development — a notable cross-camp alliance.
Startup Funding & Valuations
Databricks hits $188B valuation, extending its run as AI's favorite second act Link.
Databricks is finalizing a round at a $188 billion valuation, among the largest private valuations in history.
The company has remade its image into an AI company and published research on the cost savings of open-weights models for coding.
The round extends a remarkable run for a company once known primarily for data lakehouses.
AI infrastructure startup Fireworks closes $1.5B round at $17.5B valuation Link.
Fireworks AI helps developers train and run AI models in production.
Atreides Management, Index Ventures, and TCV jointly led the round.
The raise reflects continued investor appetite for the inference and model-serving layer.
Defense technology startup Helsing raises $1.8B at $18B valuation Link.
The German defense-tech company closed a $1.8 billion Series E from a group of 10 new and returning investors.
Helsing is now among the most valuable defense startups in the world.
European defense AI continues to attract late-stage capital at an accelerating pace.
Indian AI coding startup Emergent becomes a unicorn with $130M Series C Link.
Emergent reached a $120 million annualized revenue run rate with more than 200,000 paying customers.
It's the company's third major round in just 10 months, aimed at giving non-technical users production-grade software tools.
Elsewhere in funding: video-gen startup PixVerse raised $439M past a $2B valuation, and Nous Research is in talks at $1.5B.
Data Centers + Energy
New York State halts construction of all new data centers Link.
Gov. Kathy Hochul signed an executive order pausing state environmental permits for large data centers, making New York the first state to do so.
Hochul argues the AI building boom shouldn't come at the expense of higher electricity costs for residents.
The moratorium is a warning shot for hyperscalers counting on rapid Northeast buildouts.
Meta boosts investment in Hyperion data center campus to $50B+ Link.
Meta significantly expanded the scope of its Hyperion AI hub in Richland Parish, Louisiana.
The $50B+ commitment makes Hyperion one of the largest single-site AI infrastructure projects in the world.
Power companies are increasingly using eminent domain to seize land for data centers — even as 70% of Americans say "not in my backyard."
SoftBank's CEO estimates $5 trillion needed annually to meet AI demand Link.
Masayoshi Son dismissed AI skeptics, saying "those who condemn AI are themselves spitting upward."
Son estimates $5 trillion in annual investment is needed to meet AI compute demand.
The AI boom is increasingly built on debt — just as investor demand for hyperscaler bonds is plunging.
Hardware + Infrastructure
IBM shares plummet 25% on weak preliminary earnings Link.
IBM posted preliminary Q2 revenue of ~$17.2 billion, missing expectations and triggering its worst stock crash of all time.
Analysts say the crash reveals uncomfortable truths about both the AI trade and the legacy-tech bubble.
The selloff rippled across enterprise IT names as investors reassessed AI-driven earnings claims.
Nvidia launches Cosmos 3 Edge model and expands its physical AI push in Japan Link.
Cosmos 3 Edge is a compact world model built to run vision reasoning and robot control directly on edge devices.
Nvidia announced a wave of Japanese partnerships as Jensen Huang courts robotics for Japan's aging society.
The launch extends Nvidia's physical AI strategy beyond the data center.
Inference cloud operator General Compute raises $400M in debt financing Link.
Upper90 is underwriting the $400 million debt round for the AI infrastructure startup.
Notably, the deal is backed by inference chips rather than Nvidia GPUs — a first for the GPU-financing world.
A secondary market for used Nvidia H100s and A100s also opened this week via Compute Exchange, signaling a maturing hardware market.
Healthcare + Life Sciences
Chai Discovery nabs $400M Series C as AI-designed antibodies reach Big Pharma Link.
Chai builds AI models that predict interactions between biochemical molecules.
The $400M Series C nearly doubles its prior valuation as its AI-designed antibodies reach Big Pharma partnerships.
The raise is one of the largest ever for an AI-native biotech platform.
OpenAI researcher Miles Wang in talks to launch AI drug discovery startup valued at $2B Link.
The funding discussions point to intense investor interest in applying frontier AI to life sciences.
A $2B valuation before launch would rank among the richest pre-product rounds ever.
Meanwhile, a decade after Geoffrey Hinton predicted radiologists were obsolete, their salaries have climbed to as much as $571K amid growing demand.
Policy + Regulation
DeepMind CEO calls for an independent standards body to regulate frontier AI Link.
Demis Hassabis proposed an AI "standards body" modeled after FINRA to test frontier models and develop release best practices.
The proposal comes as prominent economists and tech executives published a public letter calling for a new regulatory approach to AI.
Industry self-regulation proposals are gaining steam as federal action stalls.
EU orders Google to share search data with rivals by 2027 Link.
Brussels wants rival AI agents to be able to book restaurants on Android just like Gemini does.
Google is pushing back, arguing the mandate means handing user data to "unfamiliar companies."
The order is one of the most aggressive interventions yet in the AI-agent platform wars.
Xi offers AI olive branch to the world, calling for 'symphony of global cooperation' Link.
"The development of artificial intelligence should not be a solo performance by any single country," Xi said in Shanghai.
The remarks accompany a proposed global AI cooperation organization championed by Beijing.
The charm offensive lands as Chinese open-weights models surge on frontier leaderboards.
OpenAI staffers are funding a rival super PAC to take on their boss Link.
OpenAI employees have donated more than $215,000 to a political effort opposing Leading the Future, the super PAC backed by company president Greg Brockman.
The split highlights deep internal disagreement over AI policy and safety within the world's most valuable startup.
Employee political activism in AI labs is becoming a story to watch heading into the midterms.
Workforce + Labor
'Dr. Doom' Nouriel Roubini says we're headed for universal basic income or 'some form of socialism' Link.
Roubini argues the AI revolution will force governments to take stakes in big tech firms and redistribute the gains.
"Essentially, the government is going to take over some fraction of the big tech firms," he said.
The comments add a prominent macro voice to the AI-and-labor debate.
Goldman economist offers a reality check on AI adoption Link.
Elsie Peng's research draws a parallel to the PC revolution: productivity gains took 15 years to materialize.
The research also flags quiet worker resistance — including sabotage — as a real drag on AI rollouts.
The findings temper expectations of near-term, economy-wide AI productivity gains.
Meta's Adam Mosseri says AI token budgets could soon be capped per engineer Link.
Mosseri predicts companies will manage AI token spending like payroll or other operating expenses.
Engineers could soon face individual token caps as AI becomes a top-line cost center.
Ramp also expanded token spend tracking this week, targeting AI's fastest-growing cost line.
In education, a survey of 435 educators found 84% of students use AI for homework while only 3 in 10 schools have rules for it.
Lightscape Portfolio is Hiring
Our portfolio companies have 687+ open roles across 20 companies right now. A few featured opportunities this week:
Shield AI (Series F · 393 roles) — Principal Engineer, Software (R5285) — United States · Browse all
Etched (Series A · 101 roles) — Mechanical Operations Lab Lead — San Jose · Browse all
Parloa (Series D · 54 roles) — Staff/Principal Software Engineer (Frontend) — Berlin Office · Browse all
Formic (Series B · 25 roles) — Chief of Staff — San Francisco || Oakland, CA · Browse all
Standard Bots (Series B · 24 roles) — Manufacturing and Process Engineering Manager — Glen Cove, NY · Browse all