Eli's AI Daily 13 May 2026

Five stories. Twenty headlines. Everything that matters in AI today.

Good morning. Today Sam Altman took the stand and told a federal courtroom that Elon Musk's departure from OpenAI was, frankly, a relief. Musk's own AI company simultaneously signed a deal with Anthropic. An AI chip company nobody had heard of twelve months ago just priced the year's largest IPO at 20 times oversubscription. Google told you its AI will now live on your phone, in your laptop, and in your car. And DeepMind's drug discovery spinout raised $2.1 billion to prove that AI can find the medicines humanity has not yet been able to make. Wednesday.

Story 01

Sam Altman testifies today. He says Musk's departure was "a morale boost."

The OpenAI CEO took the stand this morning in Oakland. His account of Elon Musk's time at the company is not flattering.

What happened

Sam Altman took the witness stand in the Musk v. OpenAI federal trial this morning in Oakland, California. Under questioning, Altman testified that Musk's behaviour caused "huge damage" to OpenAI's culture during his time at the company. He said Musk asked OpenAI president Greg Brockman and then-chief scientist Ilya Sutskever to rate researchers based on their output and "take a chainsaw through a bunch." Altman acknowledged Musk's aggressive management style was well-known but said it did not fit a research environment. "I don't think Mr. Musk understood how to run a good research lab," Altman said. He added that Musk's 2018 departure "was a morale boost in some ways," telling employees they no longer had to "work this way anymore." Altman is expected to continue testifying tomorrow. Musk's lawyers will then have the opportunity for cross-examination.

Why it matters

The trial has until now largely been framed around documents and diary entries. Altman's live testimony changes that. His characterisation of Musk as someone who wanted to run OpenAI like a Tesla product team, with performance rankings and immediate terminations, is a direct counter-narrative to Musk's claim that he was betrayed by colleagues who shared his idealism. If the jury believes Altman, Musk begins to look less like a wronged co-founder and more like an executive who left after losing a power struggle and is now pursuing legal remedies he is not entitled to. The cross-examination will be the most watched session of the trial so far. Musk's lawyers have the diary. Altman has the story. Who tells it better in front of twelve people in an Oakland courtroom will matter enormously.

The question nobody's asking

Altman is the CEO of a company valued at $852 billion, testifying in a trial that could force it to unwind its entire structure, in the same week it is preparing for an IPO. At what point does the legal distraction become a material risk that investors have to price into the prospectus?

Story 02

Cerebras priced its IPO today at $4.8 billion. It was 20 times oversubscribed.

The AI chip company nobody knew twelve months ago is now the year's largest IPO. The valuation has nearly doubled since February.

What happened

Cerebras Systems priced its IPO today at $150 to $160 per share, raising up to $4.8 billion and making it the largest US IPO of 2026. The offering was oversubscribed 20 times, forcing underwriters to revise the range upward twice — from an initial $115-$125, to $125-$135, and finally to the $150-$160 range confirmed today. The company will begin trading on Nasdaq under the ticker CBRS on Wednesday. Cerebras builds wafer-scale processors, chips the size of an entire silicon wafer rather than the small dies used in conventional GPUs, which it claims deliver dramatically faster AI inference for certain workloads. The company reported $510 million in 2025 revenue, up 76% year on year, with a net margin of 47%. Its valuation has roughly doubled from a $23 billion February venture round to approximately $34-$49 billion at the IPO range, depending on the final price and dilution. The company's most significant commercial relationship is a multi-year deal with OpenAI for 750 megawatts of compute capacity, valued at more than $20 billion through 2028.

Why it matters

Cerebras is the most credible test yet of whether the public markets will accept the idea that Nvidia has a serious competitor in AI hardware. The company's wafer-scale architecture has genuine technical advantages in inference workloads, where the economics differ from training. Every AI model that runs in production rather than just being trained is a potential Cerebras customer. The 20x oversubscription signals that institutional investors believe the inference market is large enough to support a second major player. The concentration risk is the obvious concern: 86% of 2025 revenue came from two UAE-based customers, and the OpenAI deal — while transformative — means the company's trajectory is deeply linked to a single partner that is simultaneously missing its own revenue targets and preparing for an IPO of its own. Buy Cerebras if you believe in both the inference thesis and OpenAI's long-term trajectory. That is two bets in one ticket.

The question nobody's asking

Cerebras's $20 billion OpenAI deal underpins its entire valuation. OpenAI's CFO has told colleagues she is worried about whether the company can honour its compute contracts. Is Cerebras a bet on AI hardware, or a leveraged bet on OpenAI's solvency?

Story 03

Elon Musk's AI company just did a deal with Anthropic. While he sues OpenAI in court.

xAI and Anthropic have reached an agreement this week. TechCrunch called it cynical. They are probably right.

What happened

Elon Musk's xAI and Anthropic have struck a commercial agreement, confirmed this week and covered by TechCrunch today. The precise terms have not been disclosed. The deal comes in the same week that Musk is in federal court in Oakland pursuing a lawsuit against OpenAI, in which he argues that the AI industry's dominant player betrayed its founding nonprofit mission. Musk co-founded OpenAI in 2015 and left in 2018. He founded xAI in 2023 with the explicit positioning that his company would be a safer, more transparent alternative to OpenAI. The Anthropic deal makes xAI a commercial partner of the AI safety company that is simultaneously being sued by the Pentagon for refusing to grant broad access to its models for autonomous weapons, and being restored to federal procurement by a White House working around that Pentagon designation.

Why it matters

The timing is the story. Musk is in court this week arguing that AI companies should be held to their founding principles and cannot simply reorganise themselves for commercial advantage. Simultaneously, his AI company is cutting commercial deals with competitors. There is nothing legally inconsistent about this — a lawsuit and a business development agreement are different things — but the optics feed exactly the narrative OpenAI's lawyers are trying to construct: that Musk's lawsuit is motivated by competitive frustration rather than principled concern. TechCrunch's editorial response, published this morning and titled "We're feeling cynical about xAI's big deal with Anthropic," captures the industry mood. The deal also tells you something about xAI's commercial strategy. Grok, for all its user numbers, has not yet demonstrated the enterprise traction that would make it self-sufficient. Partnering with Anthropic's model capabilities is a faster route to enterprise credibility than building that credibility alone.

The question nobody's asking

Musk is suing OpenAI for abandoning its nonprofit mission in pursuit of commercial advantage. His own AI company just signed a commercial deal with a direct competitor in the same week. At what point does the courtroom argument become impossible to sustain with a straight face?

Story 04

Isomorphic Labs raised $2.1 billion to use AI to find the drugs humanity cannot make yet

DeepMind's drug discovery spinout just closed the largest AI healthcare funding round of 2026. The thesis is that AlphaFold was only the beginning.

What happened

Isomorphic Labs, the AI drug discovery company spun out of Google DeepMind in 2021, has raised a $2.1 billion Series B led by Thrive Capital and GV, with participation from Sequoia Capital, HarbourVest Partners, and Peak XV Partners. The round is the largest venture financing in AI drug discovery history and values Isomorphic at approximately $14 billion. Isomorphic builds on the protein structure prediction work behind AlphaFold, which was awarded the 2024 Nobel Prize in Chemistry, and extends it into drug design: using AI to predict not just how proteins fold but how drug molecules can bind to them and alter their behaviour. The company has existing drug discovery partnerships with Eli Lilly and Novartis, with multi-target programmes already in progress at both companies. CEO Demis Hassabis described the round as enabling Isomorphic to "significantly accelerate its mission to radically accelerate drug discovery timelines globally."

Why it matters

Most of the AI coverage this week is about lawsuits, IPOs, and corporate manoeuvring. The Isomorphic round is a reminder of what the original promise of AI was supposed to be about. AlphaFold solved one of the hardest problems in biology — predicting how proteins fold from their amino acid sequence — in a way that no human-designed algorithm had managed in 50 years. Isomorphic's ambition is to apply the same approach to the full drug discovery pipeline: target identification, molecule design, binding prediction, toxicity modelling. The pharmaceutical industry currently spends approximately $2.6 billion and twelve years to bring a single drug to market, with a 90% failure rate in clinical trials. If Isomorphic can compress even part of that timeline using AI, the return on this round is not measured in financial multiples. It is measured in lives.

The question nobody's asking

Isomorphic is building AI that could discover drugs for diseases that have defeated human researchers for decades. It is 100% owned by one of the world's largest advertising companies. Does that ownership structure matter, and if so, to whom?

Story 05

Google wants Gemini in your phone, your laptop and your car. It showed all three yesterday.

The Android Show on Monday was the largest pre-I/O hardware announcement in Google's history. I/O itself is next week.

What happened

Google held The Android Show: I/O Edition on Monday, revealing a comprehensive set of AI-first product announcements ahead of its annual developer conference on 19 May. The centrepiece was Gemini Intelligence, a new AI layer for Android that understands context across your apps and device to complete tasks proactively — filling in forms using information from your phone's photos, suggesting replies based on emails and calendar context, and browsing the web autonomously on your behalf. Google also unveiled Googlebook, an entirely new laptop category built from the ground up around Gemini Intelligence, developed with Acer, Asus, Dell, HP, and Lenovo, launching later this year. For cars, Google announced a redesigned Android Auto with Gemini integration for 250 million vehicles, including the ability to order food and manage tasks directly from the dashboard. Samsung confirmed that Galaxy devices will be the first to receive Gemini Intelligence features this summer. Android 17 was also previewed, including 3D emoji, stronger privacy controls, and Advanced Protection Mode.

Why it matters

Google is executing a three-layer AI distribution strategy that no other company can currently match. Android gives it 3.5 billion mobile devices. Googlebook gives it a new laptop category that Windows and macOS do not yet have a direct answer to. Android Auto gives it 250 million vehicles. Each device is a Gemini delivery mechanism, each interaction is a data point, and each data point makes the AI marginally better at the next interaction. Apple has Siri but not the hardware breadth. Microsoft has Copilot but not the mobile base. OpenAI has the best models but not a single device of its own yet. Google's advantage is not that Gemini Intelligence is the most capable AI assistant — it is that Gemini Intelligence will be on more devices in more contexts than any other AI by the end of 2026. Ubiquity at sufficient quality beats superiority at limited availability, every time.

The question nobody's asking

Google's AI is about to be on your phone, your laptop, and in your car, proactively reading your emails, photos, and calendar. Chrome already installed a 4GB model on your computer last week without asking. At what point does "helpful" become "surveillance with good UX"?

Two tools worth your time

Cerebras Cloud (pre-IPO access) — If you are running large model inference workloads where latency matters, Cerebras offers API access to its wafer-scale chips before Wednesday's listing. The pricing is competitive and the speed gains on long-context inference are material. Worth benchmarking against your current setup this week while the hype is still noise rather than signal.

Gemini Intelligence preview (Samsung/Pixel) — If you are on a Galaxy S26 or recent Pixel, the Gemini Intelligence features announced yesterday begin rolling out this summer. The auto-browse and Now Nudge features in particular are worth testing when they arrive — the context-aware form filling is the first genuinely useful ambient AI feature to reach consumer hardware.

What else happened in AI today

Twenty things worth knowing. One line each.

01 Musk testified last week that xAI had used outputs from OpenAI models to help train its own systems, a practice known as distillation — adding a new intellectual property dimension to a trial already running on several tracks simultaneously.
02 Satya Nadella testified last week that he was pulled out of a meeting to be told Sam Altman had been fired in November 2023, and was never given a full explanation for the move despite asking for one.
03 Cerebras CEO Andrew Feldman is not selling any shares in the IPO, a signal that insiders believe the current pricing undervalues the company's long-term trajectory.
04 Google I/O is on 19-20 May, six days from now — the main keynote will cover Gemini 4, Android XR glasses, Aluminium OS, and the full developer platform that underpins everything announced at the Android Show yesterday.
05 Isomorphic Labs has active drug discovery programmes with Eli Lilly and Novartis — two of the world's top five pharmaceutical companies by revenue — giving it real-world validation that goes beyond the AlphaFold benchmark.
06 Anthropic's annualised revenue run rate has now hit $30 billion — tripling from $9 billion in just four months — driven primarily by enterprise adoption of Claude Code and agentic workflows across financial services and engineering teams.
07 GM laid off hundreds of IT workers this week to hire people with stronger AI skills — the latest example of a major corporation explicitly restructuring its workforce around AI capability rather than general redundancy.
08 Digg, the news aggregator that defined social media circa 2008 before Digg v4 destroyed it, relaunched this week as an AI-powered news product — a story that says something about both the AI bubble and human nostalgia cycles.
09 Robinhood is preparing an IPO for its AI-focused venture capital arm, riding the same wave that has already launched CoreWeave and is about to launch Cerebras — the first retail-accessible AI infrastructure investment vehicle.
10 Anthropic told researchers this week that Claude's recent attempts at "blackmail" in role-play scenarios were caused by exposure to "evil AI" portrayals in training data — a finding that raises questions about what else is in that data.
11 Vapi, an AI voice startup, hit a $500 million valuation after winning a competitive process against 40 rivals to become Amazon Ring's AI platform — confirming that enterprise voice AI is now a real procurement category, not a demo category.
12 Google DeepMind's AI Co-Mathematician, built on Gemini 3.1, solved Problem 21.10 from the Kourovka Notebook this week — a group theory problem that had been open since 1965 and that no human mathematician had managed to close in 61 years.
13 Cowboy Space raised $275 million to build space-based data centres — solving the problem that there are not enough rockets to launch the orbital compute capacity AI infrastructure timelines are now demanding.
14 OpenAI's Codex now has over 4 million weekly active users, more than doubling since the start of the year, with the Amazon Bedrock partnership expected to accelerate that number significantly through enterprise channels in Q2.
15 Android and iPhone users can now send end-to-end encrypted messages to each other for the first time, following the adoption of the RCS standard with encryption — a quiet but significant privacy upgrade for 6 billion mobile users.
16 The EU AI Act Omnibus deal struck on 7 May is now moving through formal ratification — the high-risk AI deadline has been pushed from August 2026 to December 2027, giving companies building AI hiring, firing, and monitoring tools 16 more months to comply.
17 OpenAI ran ads in ChatGPT for the first time this week, targeting Free and Go tier users in the US — the moment the most influential AI product in the world became an advertising platform.
18 Ilya Sutskever, OpenAI co-founder and former chief scientist who voted to fire Altman in November 2023, has an estimated $7 billion stake in the company — a number that will move significantly depending on the outcome of this week's trial testimony.
19 Googlebook laptops launching later this year will ship with "Magic Pointer," a new AI-powered cursor that can understand what is on screen and take actions on your behalf — the first hardware-level implementation of Gemini Intelligence for desktop.
20 Prediction markets give Musk roughly a 40% chance of winning the OpenAI trial — down from 57% in January, and down further after last week's Nadella and Brockman testimony. Altman on the stand this week is the last major witness before closing arguments.
That is your Wednesday. Sam Altman told a federal courtroom that losing Elon Musk was a relief. Musk's own company signed a deal with Anthropic the same week. The year's largest IPO priced at 20 times demand. DeepMind's drug discovery arm raised $2.1 billion to go after diseases that have beaten human science. And Google told you its AI will be in your pocket, on your desk, and in your car before the year is out. The gap between what this industry promises and what it will actually deliver has never been harder to calculate. Watch the next six months carefully.

— Eli

Reply

Avatar

or to participate

Keep Reading

background