• AI Rundown
  • Posts
  • The AI Rundown by Lightscape Partners - 7/13/2026

The AI Rundown by Lightscape Partners - 7/13/2026

Apple sues OpenAI over trade secrets, OpenAI ships GPT-5.6 and ChatGPT Work, and SK Hynix pulls off the biggest foreign IPO in U.S. history

Author

Ted Wagner
July 13, 2026

Image credits: Kevin Dietsch / Getty Images


Welcome back to The AI Rundown.

This week had a little bit of everything: a blockbuster lawsuit between two of the most valuable companies in tech, a major frontier model launch paired with an aggressive agentic push into the enterprise, and Wall Street's biggest-ever foreign IPO courtesy of the AI memory boom. Add in Anthropic's $19B data center lease, Meta's new coding model, and a venture market that has already eclipsed all of 2025, and you have one of the busiest weeks of the year. Let's get into it.

  • Databento raises $97M to modernize market data. Databento closed a significantly oversubscribed Series B led by NEA, drawing more than $300M in demand from investors. The company provides self-service access to institutional-grade real-time and historical market data, replacing the lengthy contracts, opaque pricing, and legacy workflows that have defined the category. With revenue growing 6.65x year over year, 97% enterprise retention, and profitability, Databento is emerging as a leading challenger to the legacy providers powering global financial markets.

  • Apple sues OpenAI over alleged IP theft. Apple filed suit in the Northern District of California accusing OpenAI and former Apple employees of stealing hardware trade secrets. The dispute traces back to OpenAI's acquisition of io Products, the device startup founded by former Apple executives, which put the two companies on a collision course in consumer hardware. It's the highest-profile legal clash yet between Big Tech and the frontier labs.

  • OpenAI launches GPT-5.6 and ChatGPT Work. OpenAI rolled out its newest family of models, promising improvements across reasoning, coding, and cybersecurity. Alongside the launch, the company debuted ChatGPT Work, an agentic mode that autonomously executes tasks across a user's connected apps, files, and recurring workflows. The launch lands amid leadership churn, with No. 2 executive Fidji Simo stepping down the same week.

  • SK Hynix raises $26.5B in the biggest foreign IPO in U.S. history. The Nvidia memory supplier jumped nearly 13% in its Wall Street debut, capping the largest U.S. initial share sale ever by a foreign company. The listing is being read as a bellwether for investor appetite for AI infrastructure names. U.S. officials are already urging SK Hynix and Samsung to build new domestic fabs.

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

OpenAI

OpenAI launches its new family of models with GPT-5.6. Link

  • The GPT-5.6 family promises improvements across a range of areas, including cybersecurity, reasoning, and coding.

  • The release marks OpenAI's most advanced model family to date and anchors its broader enterprise push this week.

  • The launch also comes as the company navigates a possible IPO and intensifying competition with Anthropic in the enterprise market.

OpenAI debuts ChatGPT Work, an agentic tool for automating business workflows. Link

  • ChatGPT Work is a new mode within ChatGPT designed to perform actions autonomously across a user's connected applications, files, web tools, desktops, and recurring workflows.

  • It launched alongside the global rollout of GPT-5.6, positioning OpenAI squarely against agentic offerings from Anthropic and Microsoft.

  • The tool is aimed at automating repeatable business processes rather than one-off chat interactions.

Fidji Simo steps down from OpenAI's no. 2 role. Link

  • Simo, OpenAI's CEO of Applications and No. 2 executive, is stepping back from her full-time role after a longer-than-expected medical leave.

  • The departure creates a leadership vacuum at a tricky time, as OpenAI eyes a possible IPO and races to catch Anthropic in the enterprise market.

  • Her exit adds to a string of senior departures this week, including OpenAI's head of safety and its chief futurist.

Apple v. OpenAI + Legal

Apple sues OpenAI, former employees over alleged intellectual property theft. Link

  • Apple filed the complaint in the U.S. District Court for the Northern District of California, alleging theft of IP related to its consumer devices.

  • The suit targets OpenAI and former Apple employees who joined via io Products, the hardware startup founded by ex-Apple executives that OpenAI acquired last year.

  • The case sets up a landmark fight over talent mobility and trade secrets as frontier labs push into consumer hardware.

New York Times says OpenAI hid evidence in ChatGPT copyright trial. Link

  • News publishers allege OpenAI concealed tools and datasets that could identify copyrighted journalism in ChatGPT outputs.

  • The publishers escalated the long-running copyright lawsuit with a new motion for sanctions.

  • The dispute raises the stakes of the most closely watched AI copyright case in the country.

Anthropic

Anthropic wants you to pay up for Claude Fable 5. Link

  • Claude subscribers must soon pay usage-based fees to access Anthropic's best consumer AI model.

  • Wired frames the move as a sign that the golden era of flat-rate AI subscriptions is ending as frontier inference costs climb.

  • The shift is a test of whether consumers will pay premium, metered pricing for top-tier model capability.

Claude Cowork expands to mobile and web. Link

  • Users can now start a task at their desk, get status updates on their phone, and pick up finished output later — even with their laptop closed.

  • The expansion takes Anthropic's agentic desktop product to wherever users are, extending the coding-agent playbook to general office work.

  • It's a notable step in the race to own the agentic knowledge-work layer, not just the model underneath.

Anthropic inks $19B AI data center lease with TeraWulf. Link

  • Anthropic will pay $19 billion to lease a TeraWulf data center for 20 years.

  • The facility is being built on the site of a former industrial property, and TeraWulf shares rose 4.8% on the announcement.

  • The deal underscores how frontier labs are locking up long-duration compute capacity years in advance.

Why the rise of open source AI isn't hurting Anthropic … yet. Link

  • Open source models' success isn't coming at the expense of frontier labs, TechCrunch argues.

  • Instead, open and closed models appear to capture two different phases of the same workload life cycle — frontier models for cutting-edge tasks, open models for cost-optimized scale.

  • The dynamic echoes the Amazon CTO's observation this week that companies are shifting routine workloads to cheaper open-source models.

Frontier Models

Meta enters the crowded AI coding battle with Muse Spark 1.1. Link

  • Meta's new flagship Muse Spark 1.1 goes up against coding products from Anthropic and OpenAI, with multi-agent upgrades.

  • The release is part of Meta's accelerated AI push under Alexandr Wang.

  • Meta also rolled out Muse, a new AI image generator, and confirmed its custom AI chips begin production in September — a full-stack offensive.

SpaceXAI releases Grok 4.5, which Elon describes as an 'Opus-class model.' Link

  • Grok 4.5 is pitched as a cheaper, more efficient alternative to other powerful AI models, dramatically undercutting Anthropic and OpenAI on price.

  • Musk described the model as "Opus-class," explicitly benchmarking against Anthropic's tier naming.

  • The release came the same week Musk publicly praised Anthropic's Mythos/Fable models and promised not to "cut off" the company.

Refiant goes where rivals only promised with a 10 million-token AI model. Link

  • AI optimization startup Refiant launched Protea, a suite of long-context models led by a 10 million-token context window — among the largest ever made publicly available.

  • Most leading models handle far smaller working memories, making ultra-long context a real differentiator for document- and codebase-scale tasks.

  • The launch pressures frontier labs that have promised, but not shipped, comparable context lengths.

Hardware + Infrastructure

SK Hynix raises $26.5B in the biggest foreign IPO in US history, is urged to build new US fabs. Link

  • The AI chip boom produced its biggest Wall Street moment yet, with the Nvidia memory supplier jumping nearly 13% on debut.

  • It is the largest U.S. initial share sale ever by a foreign company, following a nearly 800% run-up in the stock at home.

  • U.S. officials are now urging SK Hynix and Samsung to build new American fabs.

AI chip maker SambaNova raises $1B at $11B valuation, 5 months after last mega round. Link

  • SambaNova closed the first tranche of its Series F at an $11B valuation, just months after Intel was rumored to be exploring an acquisition at roughly $1.6 billion.

  • The dramatic markup reflects surging demand for inference-focused alternatives to Nvidia GPUs.

  • It's the latest sign that AI chip challengers can still raise mega rounds at steep valuations.

Meta's new AI chips will begin production in September. Link

  • Meta is taking a modular approach to its custom silicon, anticipating that its needs will change as AI evolves rapidly.

  • In-house chips would reduce Meta's dependence on Nvidia and lower the cost of its aggressive AI roadmap.

  • Meta joins OpenAI and reportedly DeepSeek in developing custom inference silicon.

Amazon launches $25B bond sale to fund AI infrastructure. Link

  • Amazon returned to the bond market to raise at least $25 billion for its AI buildout, per Bloomberg.

  • The offering is split into eight parts, with senior unsecured notes maturing anywhere from three to 40 years.

  • Hyperscalers are increasingly turning to debt markets to finance the capex race, a trend worth watching for AI-bubble skeptics.

Enterprise AI

Microsoft joins AI cost-cutting trend by relying more on its own models. Link

  • Microsoft is reportedly ditching OpenAI's and Anthropic's models in favor of its own in-house models to cut costs.

  • The move adds to breakup chatter with OpenAI, even as OpenAI says GPT-5.6 remains the "preferred model" for Microsoft Copilot 365.

  • Microsoft's Frontier program aims to turn its enormous AI spending into measurable returns.

Starbucks taps AI to cut reliance on Microsoft, IBM software. Link

  • "There's clear opportunities to reduce the spend in software," said Starbucks CTO Anand Varadarajan.

  • The coffee giant is an early, prominent example of enterprises using AI to replace incumbent enterprise software rather than merely augment it.

  • If the pattern spreads, AI may pressure traditional software vendors' seat-based revenue models.

Companies are shifting toward cheaper open-source AI models to rein in costs, Amazon CTO says. Link

  • Stories of runaway AI bills are making executives skittish about AI spending, says Amazon CTO Werner Vogels.

  • Enterprises are increasingly routing routine workloads to cheaper open-source models while reserving frontier models for high-value tasks.

  • The comments align with this week's Microsoft and Starbucks cost-cutting moves — enterprise AI budgets are maturing fast.

Startup Funding & Valuations

Open-source AI model developer MiniMax raises $2B in funding. Link

  • The Shanghai-based AI developer is raising $2 billion, with more than half from newly issued shares and the rest via convertible bonds, per Bloomberg.

  • MiniMax is one of the leading Chinese open-source model labs riding the global open-weights wave.

  • The raise underscores how much capital is flowing to open-source model developers, not just closed frontier labs.

Mercor is in talks for a $20B valuation. Link

  • A new $20 billion valuation would double the $10 billion mark the AI recruiting and data-labeling company reached in October.

  • Mercor also acquired Deeptune this week to build training environments for AI agents.

  • The talks reflect insatiable investor demand for companies powering frontier-lab training pipelines.

Lovable reportedly in talks to double its valuation to $13.2B. Link

  • The vibe-coding startup is discussing a $300 million round expected to be led by Menlo Ventures, per Sifted.

  • The deal would double Lovable's valuation, making it one of Europe's most valuable AI startups.

  • AI app-building tools continue to command some of the fastest markups in venture.

Prime Intellect raises $130M Series A to help enterprises build their own AI agents. Link

  • The round values the two-year-old company at $1 billion.

  • Prime Intellect's goal is to let organizations train their own agentic systems without relying on frontier AI labs.

  • The bet: enterprises will want sovereign control over their agents, not just API access.

Popular open source AI developer tool Ollama raises $65M, grows to nearly 9M users. Link

  • Benchmark-backed Ollama has amassed 176,000 GitHub stars and nearly 17,000 forks by making it easy to run AI models locally.

  • The tool has grown to nearly 9 million users as local, private inference gains traction.

  • The raise is another data point in the open-source AI momentum running through this week's news.

Policy + Regulation

How did the government decide OpenAI's frontier model was safe to release? Link

  • TechCrunch examines the opaque process by which the U.S. government cleared OpenAI's frontier model for release.

  • "Exactly what that dialog looked like between the government and Anthropic and OpenAI is unclear."

  • Sam Altman says OpenAI made "many changes" during its talks with the U.S. — but the standards applied remain undisclosed.

OpenAI's latest AI model likely has similar cyber vulnerabilities to the one that led to U.S. export controls on Anthropic's Fable, British agency says. Link

  • A U.K. AI agency found "universal jailbreaks" that unlocked dangerous cyber capabilities in OpenAI's GPT-5.6.

  • The flaws resemble the security issues that led the U.S. government to force restrictions on Anthropic's Fable 5.

  • The findings sharpen questions about whether frontier-model safety reviews are keeping pace with releases.

Google will now disclose which ads are made with AI. Link

  • A new feature will indicate when advertisers used generative AI tools to create or edit their ads.

  • The move comes as AI-generated content floods ad platforms and deepfake concerns mount.

  • It sets a transparency precedent other ad platforms will likely be pressed to follow.

Lightscape Portfolio is Hiring

Our portfolio companies have 680+ open roles across 20 companies right now. A few featured opportunities this week: