AI Daily Briefing — 2026-06-23

SpaceX has entered a significant computing agreement with Reflection AI, an open-source artificial intelligence lab, marking the startup's first major infrastructure deal. Under the terms, Reflection will pay $150 million per month starting July 1, 2026, through 2029, for priority access to Nvidia's latest GB300 AI chips and supporting hardware at SpaceX's Colossus 2 data center near Memphis, Tennessee. The total value of the contract could reach $6.3 billion, though either party can terminate with 90 days' notice after the first three months.

This deal is notably smaller than SpaceX's previous arrangements with Anthropic and Google, which cost $1.25 billion and $920 million per month respectively. Those contracts also run through July 2029, though Elon Musk has emphasized the cancellation clauses. Reflection, founded in 2024 by two former Google DeepMind researchers, is using this agreement to highlight the value of its open-weight AI strategy, which releases trained parameters publicly as an alternative to closed models from labs like Anthropic and OpenAI.

The timing aligns with growing interest in open-source AI following the U.S. government's ban on Anthropic's closed models, Fable and Mythos. A Reflection spokesperson stated that recent events underscore the importance of open source, as more nations and enterprises recognize the risks of relying solely on closed systems. The Colossus data center, originally built by Musk's xAI for internal projects, has been repurposed as SpaceX pivots to renting its valuable chip inventory to leading AI labs.

A Miami-based startup called Subquadratic has emerged from stealth mode with a bold assertion: it has solved a mathematical bottleneck that has limited large language models for nearly a decade. The company introduced SubQ, a new type of LLM that it claims is faster, cheaper, and far more energy-efficient than existing models. According to Subquadratic, SubQ can process up to 12 times more text at once, enabling tasks like analyzing entire code bases or hundreds of documents, while matching the performance of top models from Google DeepMind, OpenAI, and Anthropic on key benchmarks like coding.

Initially, the company faced widespread skepticism, as it provided little evidence beyond self-published test scores and had not made SubQ widely available. AI engineer Dan McAteer captured the sentiment on X, calling it “either the biggest breakthrough since the Transformer … or it’s AI Theranos.” In response, Subquadratic commissioned independent tests from the third-party firm Appen, which evaluated SubQ’s performance. The results appeared to validate many of the startup’s claims, with Appen’s Jeanine Sinanan-Singh calling it a potential “game changer” for speed and efficiency.

Subquadratic acknowledges the initial doubt, with CTO Alex Whedon noting that releasing third-party benchmarks alongside the announcement could have preempted skepticism. While SubQ may not replace top models across the board, it offers significant speed gains at a fraction of the cost for specific tasks. CEO Justin Dangel hopes the breakthrough will usher in a new era of efficiency, predicting that transformers—the foundation of today’s LLMs—may soon be obsolete. The company plans to ensure future results are fully verified before public release.

Amazon is broadening the reach of its advanced conversational AI assistant, Alexa+, by initiating a beta test in India focused on Hindi-speaking users. The company has begun sending email invitations to select customers, asking them to complete a Hindi-language form by June 22 to join the testing program. The email states that participants will help refine Alexa+'s capabilities and will be notified when the Hindi-language trial becomes available to them, though it warns that the beta software may contain bugs or mispronounce local terms.

This move marks a significant step for Amazon in India, where it first launched Alexa with English support in 2017 and added Hindi compatibility in 2019. With over 600 million Hindi speakers in the country, many of whom mix Hindi and English in everyday conversation, Amazon aims to tap into a vast market where voice-based AI tools are expected to play a key role. The company confirmed the testing but declined to provide further details on a full launch timeline.

Alexa+, powered by generative AI, was first announced in 2025 but saw a slow rollout, becoming available to all U.S. users only in February. Since then, Amazon has expanded the assistant to several countries, including the U.K., Canada, Brazil, Mexico, Italy, and Germany, with localized support. In India, Alexa+ is offered for free to Prime members, while non-members can access it through a monthly subscription.

TechCrunch Founder Summit 2026 is set to take place on November 4 in Boston, bringing together over 1,000 founders and investors for a day focused on growth and execution. The event emphasizes peer-to-peer learning and candid discussions, offering practical insights from experienced operators and investors. Early Bird pricing is available until June 26 at 11:59 p.m. PT, providing savings of up to $190 per pass, with additional group discounts of up to 30% for teams of four or more.

The summit is designed to help founders accelerate growth by connecting with peers at similar stages, veteran founders who have scaled teams and revenue, and investors sharing their market insights. Sessions cover critical topics such as raising Series A, building effective pitch decks, reaching $10 million ARR, and preparing for an IPO. Breakout and roundtable discussions aim to deliver actionable strategies for every growth stage.

Past speakers have included Jon McNeill, former Tesla president, and partners from Sapphire Ventures and Index Ventures, alongside leaders from Sequoia Capital, NFX, and Greylock. The 2026 agenda is still being finalized, with additional speakers to be announced. Founders interested in leading a session can submit their speaking topics for a chance to be voted onto the agenda by the TechCrunch audience.

Register before June 26 to secure the lowest rates and build relationships that drive company growth.

The Trump administration has ordered Anthropic to take its two newest AI models, Fable 5 and Mythos 5, offline, citing undisclosed national security concerns. The directive, delivered via letter last Friday, gave no specific details about the alleged risks, leaving Anthropic with little choice but to remove the models entirely. The company noted that it could not guarantee foreign nationals wouldn't access the tools, given that many of its own employees are not U.S. citizens.

Reports suggest the White House acted after Amazon CEO Andy Jassy raised concerns about potential guardrail bypasses discovered by Amazon researchers. This has fueled speculation that the crackdown may be as much about strained relations between Anthropic and the administration as about actual security threats. On TechCrunch's Equity podcast, analysts noted that Anthropic has historically had a more adversarial relationship with the Trump administration compared to other AI labs.

Meanwhile, leading cybersecurity experts have signed an open letter urging the president to revoke the order, arguing that removing these advanced capabilities from U.S. network defenders could be dangerous. Some observers wonder whether the controversy might ultimately benefit Anthropic, framing the company as a defiant underdog in the AI landscape. The situation underscores ongoing tensions around AI regulation, digital sovereignty, and the uneven application of export controls.