Transcript
SPEAKER_1: Musk testified he was a fool for backing OpenAI, and then admitted xAI partly distilled OpenAI's technology. Let's start there. What happened in that courtroom? SPEAKER_2: Day one set the tone. Musk's lawyer told the jury that Altman made a mockery of OpenAI's public mission. Musk himself testified he sued because it is not okay to steal a charity, and that OpenAI's pivot sets a concerning precedent for philanthropy. SPEAKER_1: And then day two got more complicated for Musk's side. SPEAKER_2: Significantly. Musk called himself a fool for backing OpenAI, accusing Altman and Brockman of manipulating him into donating tens of millions. But internal emails from 2017 showed Musk sought majority voting control over OpenAI's co-founders, withheld promised funding, and tried to poach researchers. SPEAKER_1: So those emails undercut his narrative of being a passive donor who got betrayed. SPEAKER_2: Exactly. And then day three made it worse. When asked whether xAI ever distilled technology from OpenAI, Musk said the claim is partly true. That admission hands OpenAI's legal team exactly the kind of cross-contamination evidence they need to reframe the case from Musk-as-victim to Musk-as-beneficiary. SPEAKER_1: What did the judge have to say about all of this? SPEAKER_2: Both Musk and Altman were told to control their propensity to use social media to make things worse outside the courtroom. Musk had boosted a Ronan Farrow article about Altman's alleged deceptions on X the night before the trial. SPEAKER_1: And the jury situation? SPEAKER_2: Some seated jurors expressed negative feelings about Musk and AI, but assured the court they could be impartial. SPEAKER_1: Now, separately from the trial, there were reports about OpenAI missing some internal targets. What do we know? SPEAKER_2: OpenAI missed an internal goal of reaching one billion weekly active ChatGPT users by the end of 2025, and missed multiple monthly revenue targets earlier in 2026. OpenAI called the report prime clickbait and says its business is firing on all cylinders. SPEAKER_1: There's also a pricing shift happening in their subscriber projections. SPEAKER_2: Right. Internal projections show OpenAI expects its eight-dollar-a-month ChatGPT Go subscribers to grow roughly thirty-six times to one hundred twelve million in 2026, while twenty-dollar-a-month Plus subscribers will fall eighty percent to around nine million. The bet is volume at a lower price replaces premium subscribers who are downgrading. SPEAKER_1: And on the compute side, OpenAI has been moving fast. SPEAKER_2: Very fast. OpenAI signed contracts for ten gigawatts of US AI compute capacity, with over three gigawatts added in just the past ninety days, hitting a goal it once aimed to reach by 2029. Sources say OpenAI has abandoned the Stargate joint venture in favor of large bilateral deals. Its guiding principle remains: build more compute. SPEAKER_1: And there's a new Amazon partnership in the mix too. SPEAKER_2: Yes. Amazon and OpenAI expanded their partnership, making OpenAI's models and Codex available on Amazon Bedrock. That came a day after OpenAI ended Azure exclusivity with Microsoft. SPEAKER_1: Let's move to Anthropic. There's a nine-hundred-billion-dollar valuation being floated, but it's running into some serious headwinds. SPEAKER_2: Anthropic has begun weighing a new funding round at over nine hundred billion dollars, after previously resisting investor proposals at over eight hundred billion. This comes days after Google committed up to forty billion dollars and Amazon committed thirty-three billion total. But the valuation is running into a national security wall. SPEAKER_1: Tell me about the Mythos situation. That seems like the center of a lot of contradictions. SPEAKER_2: The White House opposes Anthropic's plan to expand Mythos access to seventy additional companies and organizations because of security concerns. At the same time, the White House is developing guidance for government agencies to get around Anthropic's supply chain risk designation and onboard models like Mythos anyway. The administration is simultaneously blocking and seeking workarounds for the same model. SPEAKER_1: And the NSA is involved too. SPEAKER_2: Sources say the NSA has been testing Anthropic's Mythos to find vulnerabilities in Microsoft products and widely used software from other companies. Meanwhile, AISI's cybersecurity analysis found GPT-5.5 reaches a similar level of performance as Mythos Preview and is the second model to solve a multi-step cyberattack simulation. Anthropic's competitive moat in national security AI just got thinner. SPEAKER_1: The Pentagon is also diversifying away from single vendors. SPEAKER_2: Pentagon Chief Digital and AI Officer Cameron Stanley confirmed the DOD is expanding its use of Gemini, saying overreliance on one vendor is never a good thing. Sources say the White House is preparing a wide-ranging AI policy memo outlining rules for national security agencies, including avoiding reliance on a single vendor. SPEAKER_1: And Goldman Sachs made a move on Anthropic too. SPEAKER_2: Goldman Sachs stopped its Hong Kong bankers from using Anthropic's models. Anthropic says its models were never officially supported in Hong Kong. SPEAKER_1: What about Google's relationship with the Pentagon? That seems complicated. SPEAKER_2: Very. Google signed a classified deal allowing the DOD to use Google's AI for any lawful government purpose, and told employees in a memo that it proudly works with the US military. Then separately, Google dropped out of a one-hundred-million-dollar Pentagon drone swarm contest after an internal ethics review. SPEAKER_1: And employees pushed back. SPEAKER_2: More than six hundred Google employees demanded Pichai bar the DOD from using Google's AI for classified work. Google's response was to sign the deal anyway and tell employees it's a point of pride. The gap between what Google tells its employees and what it tells the Pentagon has never been wider. SPEAKER_1: There's also a congressional probe into American companies using Chinese AI models. SPEAKER_2: House Homeland Security and the House China Select Committee are probing Airbnb and Cursor maker Anysphere over their use of Chinese AI models. The investigations reflect growing concern that American companies are building on Chinese-developed technology while the State Department warns about distillation in the other direction. SPEAKER_1: Now let's talk about China's crackdown on AI deals. Beijing blocked Meta's acquisition of Manus. SPEAKER_2: Beijing blocked Meta's two-billion-dollar acquisition of Manus in a fifty-four-character decree, reviewing whether it violated investment rules and ordering both sides to cancel. Meta is preparing to unwind the deal, and Manus investors have already received their returns. SPEAKER_1: And Bloomberg says this killed a broader template for Chinese AI startups. SPEAKER_2: Right. Bloomberg says the move killed the Manus template for Chinese AI startups hoping to build in China and sell to American buyers. Manus had moved to Singapore in 2025, but that wasn't enough. SPEAKER_1: China still has leverage over Meta though, even after blocking the deal. SPEAKER_2: Sources tell the FT that over ten percent of Meta's revenue comes from Chinese advertisers, and Goertek, a key manufacturer of Meta AI glasses, sits in China's supply chain. Chinese billionaire Chen Tianqiao says his startup MiroMind is adding strict firewalls between its Chinese and US businesses in response to the Manus precedent. At least two China-based funds are using parallel fund structures to continue fundraising from US investors. SPEAKER_1: China also pulled back on its own autonomous vehicle rollout. SPEAKER_2: China suspended new Level four autonomous vehicle licenses after over one hundred of Baidu's Apollo Go robotaxis stalled and disrupted traffic in Wuhan in March. Even China's own AI deployments aren't immune to crackdowns. SPEAKER_1: And despite all the blocking, Chinese companies are still using American AI tools. SPEAKER_2: Tencent employees used Claude Code to evaluate and fine-tune the company's Hy3 model. The US Commerce Department ordered chip equipment companies to halt some shipments to Hua Hong, China's second-largest chipmaker. But Beijing's GPU maker Moore Threads posted a four-point-three-million-dollar profit on one hundred seven-point-nine million dollars in Q1 revenue, up one hundred fifty-five percent year over year. SPEAKER_1: And there was a notable IPO out of China too. SPEAKER_2: Lightelligence, the first mainland Chinese photonics chipmaker to IPO in Hong Kong, jumped nearly four hundred percent on its debut. SPEAKER_1: Let's get into the big capex numbers. The FT is reporting combined 2026 spending for the four major hyperscalers is on track to hit seven hundred twenty-five billion dollars. SPEAKER_2: That's up seventy-seven percent from four hundred ten billion in 2025. And the earnings this week showed why. Alphabet reported Q1 revenue up twenty-two percent to one hundred nine-point-nine billion. Google Cloud was up sixty-three percent to twenty billion. Capex was raised to between one hundred eighty and one hundred ninety billion for 2026, with expectations to significantly increase in 2027. The stock jumped over six percent after hours. SPEAKER_1: Microsoft also had strong numbers. SPEAKER_2: Microsoft Q3 revenue was up eighteen percent to eighty-two-point-nine billion. Its AI business hit a thirty-seven-billion-dollar annual run rate, up one hundred twenty-three percent. Azure was up forty percent. Copilot now has over twenty million seats. SPEAKER_1: Amazon's results stood out too, especially AWS. SPEAKER_2: Amazon Q1 revenue was one hundred eighty-one-point-five billion, up seventeen percent. Net income was up seventy-seven percent to thirty-point-three billion. AWS was up twenty-eight percent to thirty-seven-point-six billion, its fastest growth in fifteen quarters. And capex was forty-four-point-two billion in Q1 alone. SPEAKER_1: Meta had a more mixed reaction from the market. SPEAKER_2: Meta Q1 revenue was up thirty-three percent to fifty-six-point-three billion. Net income was up sixty-one percent to twenty-six-point-eight billion. But capex was raised to between one hundred twenty-five and one hundred forty-five billion, exceeding estimates. The stock dropped over six percent after hours. Reality Labs lost over four billion, and total losses now exceed eighty billion since late 2020. SPEAKER_1: Samsung also reported, and their memory warning is significant. SPEAKER_2: Samsung Q1 revenue was up sixty-nine percent to around ninety-point-two billion. Operating profit was up seven hundred fifty-six percent to around thirty-eight-point-five billion, driven by AI memory demand. Samsung projects the memory shortage will worsen in 2027 as customers are already placing orders for next year. SPEAKER_1: And Nvidia's B300 servers are getting very expensive in China. SPEAKER_2: Nvidia B300 server prices have nearly doubled to around one million dollars each in China as a crackdown on chip smuggling dries up black market supply. SPEAKER_1: The cloud revenue picture is strong across the board. SPEAKER_2: Google Cloud grew sixty-three percent, the fastest of any hyperscaler. AWS grew twenty-eight percent, its fastest in fifteen quarters. Microsoft's AI business surpassed a thirty-seven-billion-dollar run rate, more than doubling in a year. Google's paid subscriptions hit three hundred fifty million, up twenty-five million quarter over quarter. Gemini Enterprise paid monthly active users grew forty percent quarter over quarter. SPEAKER_1: Digital advertising is also booming. SPEAKER_2: Digital advertising is booming for Google and Meta, helped by AI tools for translation, real-time ad copy, and targeting. Google's ad revenue hit seventy-seven-point-two-five billion, up from sixty-six-point-eight-nine billion in Q1 2025. Amazon ad revenue rose twenty-four percent to seventeen-point-two-four billion. SPEAKER_1: The market is clearly rewarding AI revenue but punishing spending without clear returns. SPEAKER_2: Alphabet jumped six percent on cloud growth. Meta dropped six percent after raising capex again and reporting that family daily active people fell twenty million quarter over quarter to three-point-five-six billion. Qualcomm jumped over twelve percent despite weak smartphone revenue because a top hyperscaler will start using its chips. Seagate jumped over fifteen percent on AI storage demand. SPEAKER_1: There were also some notable chip market moves. SPEAKER_2: TSMC sold its remaining Arm shares for two hundred thirty-one million, having invested around one hundred million at Arm's 2023 IPO. NXP jumped over twenty percent on a stronger automotive chip forecast. The EU is drafting a revised Chips Act II, set for late May, that would let the Commission invest directly in large, cross-border fab projects. And South Korea's stock market surpassed the UK's at four-point-zero-four trillion, fueled by Samsung and SK Hynix. SPEAKER_1: Let's run through the product launches quickly. There were quite a few this week. SPEAKER_2: OpenAI launched Symphony, an open-source spec for agent orchestration that turns a project board like Linear into a control plane for coding agents. Anthropic launched Creative Connectors, which are Claude integrations for Blender, Autodesk, Adobe, and Ableton, embedding AI directly into professional creative workflows. SPEAKER_1: Gemini also added file generation. SPEAKER_2: The Gemini app can now generate PDF, Word, Excel, Google Docs, Sheets, plain text, and Markdown files. Poolside debuted Laguna XS.2, a thirty-three-billion open-weight model, and Laguna M.1, a two-hundred-twenty-five-billion proprietary model, both for agentic coding. SPEAKER_1: Nvidia had a new model release too. SPEAKER_2: Nvidia launched Nemotron 3 Nano Omni, an open multimodal model with a thirty-billion-A3B hybrid mixture-of-experts architecture. The Nemotron 3 family saw over fifty million downloads in the past year. SPEAKER_1: And Snap is doing something interesting with conversational ads. SPEAKER_2: Snap launched AI Sponsored Snaps, which are conversational ads in Snapchat's Chat tab where users talk to brand-specific AI agents for product recommendations. Bloomberg Terminal is also getting a chatbot-style interface overhaul called ASKB, currently in beta for around one hundred twenty-five thousand users. SPEAKER_1: A few more to cover. SPEAKER_2: Lovable launched its mobile app on iOS and Android, letting users code via voice or text prompts and switch between PC and mobile. Google launched Ask YouTube, a conversational AI search that generates pages with videos and text summaries, for Premium users in the US aged eighteen and up. AWS launched Quick AI Desktop, a desktop app for Amazon's Quick AI assistant that connects tools and local files to build custom apps and dashboards. SPEAKER_1: And Anthropic has a new security tool. SPEAKER_2: Anthropic launched Claude Security, an Opus 4.7-powered code vulnerability scanner, formerly called Claude Code Security, now in public beta for enterprise users. SPEAKER_1: And finally, there's a KKR infrastructure deal worth noting. SPEAKER_2: KKR secured over ten billion dollars to launch Helix Digital Infrastructure. The AI data center company will be led by ex-AWS CEO Adam Selipsky. SPEAKER_1: Let me bring it all together. Musk's own emails showed he sought control of OpenAI and withheld funding, then admitted xAI partly distilled OpenAI's tech. OpenAI missed growth targets but is signing ten gigawatts of compute deals and projecting one hundred twelve million subscribers at eight dollars a month. The NSA is using Mythos to find vulnerabilities in Microsoft products while the White House tells agencies to diversify. China killed the Manus template but Tencent is still fine-tuning models with Claude Code. And combined 2026 capex for the four major hyperscalers is on track to hit seven hundred twenty-five billion, with Samsung warning the memory shortage gets worse in 2027.