Google DeepMind brings computer-use agentic capability to its fast, cost-efficient Gemini 3.5 Flash model. 🔧
The AI Daily
Daily · June 25, 2026
Two hardware stories dominate Thursday: OpenAI and Broadcom have debuted a custom inference chip called Jalapeño, marking OpenAI's first serious move toward silicon independence, while memory giant SK Hynix filed for a blockbuster $29.4B Nasdaq listing riding the AI memory supercycle. Meanwhile, Micron's 84.9% gross margin and Qualcomm's $15B data-center chip forecast signal that the AI infrastructure buildout is minting money at every layer of the stack.
Featured Analysis
The Demand Paradox: More software is being written. IT services companies are crashing.
On June 19, Accenture guided revenues down and fell 18% in a session — its stock at a nine-year low. The Indian IT pack followed on Monday. But GitHub data shows a 63% surge in new bug filings in 2025, and Intercom’s engineering team just cleared 54% of its entire defect backlog using AI. The market is pricing in displacement. It may be pricing in the wrong phase.
Read the analysis →🧠 Foundation models
go deeper →Model activity today ranges from Google extending computer-use capabilities into Gemini 3.5 Flash to Alibaba's surprising result showing that models never trained as agents can still outperform dedicated agent architectures — a finding with real implications for how labs build the next generation of reasoning systems.
Introducing computer use in Gemini 3.5 Flash
Alibaba's model never trained as an agent — and improved agent performance across seven benchmarks
Qwen-AgentWorld beats dedicated agent models on seven benchmarks without any agent-specific training, challenging standard fine-tuning assumptions. 🔧
Mistral launches OCR 4, turning document extraction into a full enterprise AI play
Mistral's fourth-generation OCR model returns structured document representations with bounding boxes and confidence scores, targeting enterprise document intelligence at scale. 📱
AI researchers continue to leave Google for its rivals
Top Google AI researchers Jonas Adler and Alexander Pritzel are joining Anthropic, continuing a talent drain that has already cost Google Noam Shazeer and John Jumper.
🏗️ Infrastructure
go deeper →OpenAI's debut of the Broadcom-designed Jalapeño inference chip is the headline, but the broader story is an AI infrastructure gold rush: Micron quadrupled revenue with an 84.9% gross margin, Qualcomm is forecasting $15B in data-center chip sales by 2029, and SK Hynix filed for a $29.4B Nasdaq listing — together signaling that AI compute demand is reshaping the entire semiconductor landscape.
OpenAI, Broadcom debut custom Jalapeño chip for AI inference
OpenAI's first custom inference chip, co-designed with Broadcom, reduces dependence on Nvidia and signals a new era of vertically integrated AI compute. 🧠
SK Hynix surges 11% after filing for blockbuster Nasdaq listing
The world's largest HBM memory supplier files for a $29.4B Nasdaq IPO, the second-largest listing on record, underscoring insatiable AI memory demand. 💰
Qualcomm stock pops 15% after chipmaker almost doubles projection for 2029 non-handset revenue
Qualcomm forecasts $15B in data-center chip sales by 2029, nearly doubling prior guidance, as it acquires inference-software startup Modular to build out its AI stack. 💰
Micron stock jumps 15% as soaring prices from memory crunch lead to quadrupling of revenue
⚡ Micron's AI-driven memory crunch pushed quarterly revenue to $41.45B and gross margin to a record 84.9%, surpassing Nvidia and Meta on profitability.
Europe is pushing back on Washington's chip war
Europe is resisting US MATCH Act export controls that would restrict ASML's older deep-UV lithography tools to China, highlighting transatlantic semiconductor trade tensions. ⚖️
💰 Funding & deals
go deeper →Qualcomm's acquisition of inference-software startup Modular is the strategic deal of the day, pairing custom chip ambitions with a software stack — while a clutch of seed and Series A rounds for agentic infrastructure players reflects continued investor appetite for the middleware layer between models and enterprise workflows.
Qualcomm inks deal for AI startup Modular to bolster software stack, data center build-out
Qualcomm acquires inference-software startup Modular to pair its upcoming AI chips with a full software stack, directly challenging Nvidia's CUDA ecosystem. 🏗️ 🔧
Cerebras stock plunges after earnings as CEO says margin outlook was misunderstood
Cerebras's first post-IPO earnings report spooked investors with a narrower gross-margin forecast, testing confidence in alternative AI chip business models. 🏗️
Agentic infrastructure startup Seltz raises $12.5M to help AI agents search the web for answers
Seltz raised a $12.5M seed round from Speedinvest and B Capital to build optimized web-search infrastructure specifically for AI agent pipelines. 🔧
Former Infosys chief has a new startup that wants to challenge the IT services world
Ex-Infosys CEO Vishal Sikka's new venture Hang Ten Systems raised $32M backed by Mayfield and Aramco Ventures to disrupt AI-era IT services. 🇮🇳
🔧 Middleware & platforms
go deeper →Enterprise AI middleware is maturing fast, with Shopify's model-agnostic LLM proxy and Xiaomi's self-rewriting agent harness demonstrating that production resilience — not raw model power — is becoming the key differentiator, while Mindstone's Rebel and HelloTwin's Digital Authority show the market for governed, enterprise-grade agent runtimes is heating up.
How Shopify built an AI stack that doesn't care which models survive
⚡ Shopify's internal LLM proxy gives engineers automatic failover across AI providers, making the company's AI infrastructure resilient to any single model being deprecated. 📱
Xiaomi's HarnessX rewrites its own AI scaffolding mid-task — and smaller models gain the most
Xiaomi's HarnessX framework dynamically rewrites agent scaffolding during task execution, boosting smaller model performance most dramatically. 🧠
Your enterprise AI agents should automatically remember which model is right for which task. Mindstone built the capability with Rebel
Mindstone's Rebel is a local-first agentic OS that learns which AI model performs best per task type, reducing token costs and improving reliability in enterprise deployments.
HelloTwin launches 'Digital Authority' to bring governed AI agents to the enterprise
HelloTwin's patent-pending AI twin platform anchors agent answers to a verified business-context model rather than generative hallucination, targeting enterprise governance needs.
📱 Application solutions
go deeper →AI application stories today cluster around two themes: the surprising resilience of engineering jobs despite AI fears, and deepening AI integration into creative and enterprise workflows — from Grammarly-parent Superhuman acquiring AI detector GPTZero to Google DeepMind's controversial $75M investment in A24 that signals Hollywood's accelerating AI entanglement.
Grammarly parent Superhuman buys AI detector GPTZero
Superhuman acquires GPTZero to add AI-content detection to its writing suite, serving teachers and editors needing to verify human authorship. 💰
A24 Knows You're Mad About the Google AI Collab
⚡ Google DeepMind's $75M investment in indie studio A24 is drawing backlash from film fans, spotlighting AI's deepening and contested influence in Hollywood. 💰
AI was supposed to kill engineering jobs, but new data suggests they're the most resilient
SignalFire data shows engineers are growing as a share of new hires despite AI automation fears, challenging the dominant narrative of tech-job displacement.
Stanford researchers will discuss their agentic 'scientists' that are on course to reshape drug discovery at VB Transform 2026
Stanford's agentic AI systems are automating multi-step drug discovery workflows, targeting the 90–95% project failure rate that plagues pharmaceutical R&D. 🧠
⚖️ Policy & legal
AI policy today is defined by legal confrontation at multiple fronts: a legaltech firm is suing the US government over Anthropic export-control restrictions on foreign nationals, Anthropic itself is accusing Alibaba of the largest-ever 'distillation attack' on its models, and a congresswoman is embroiled in controversy over AI-assisted legislative drafting — all signaling that AI governance is entering a litigious, contentious phase.
Anthropic accuses Alibaba of campaign to 'brazenly' and 'illicitly' extract AI capabilities
Anthropic alleges Alibaba ran the largest known distillation attack on its Claude models, systematically extracting capabilities without authorization — raising urgent IP and security questions. 🧠
Lowdown: Legal tech firm sues US over Anthropic AI access ban for foreign nationals
Legion LegalTech's lawsuit argues that US export-control law cannot legally restrict access to hosted AI services like Claude, creating a potential landmark precedent. 📱
Congresswoman denies staff used AI to write defense funding amendment
Rep. Luna's office is under fire after screenshots suggested Claude was used to draft an amendment summary, spotlighting unresolved norms around AI in legislative drafting.
CISF wants facial recognition cameras at airports linked to NATGRID
India's CISF proposes linking facial recognition feeds from 150,000 CCTVs across airports and infrastructure to NATGRID, raising sweeping surveillance and civil liberties concerns. 🇮🇳
Lowdown: SEBI wants to treat finfluencers and AI avatars as celebrities under the new ad code
India's SEBI proposes classifying AI avatars and virtual influencers as celebrities under its draft ad code, a first for AI persona regulation in financial markets. 🇮🇳
🇮🇳 India lens
India's AI signal today is multi-dimensional: Vishal Sikka's post-Infosys AI venture pulls in $32M in seed funding, while on the regulatory front SEBI is moving to classify AI avatars as celebrities and CISF is proposing mass facial-recognition surveillance linked to NATGRID — together reflecting both the entrepreneurial energy and the governance tensions shaping India's AI moment.
Ex-Infosys CEO Vishal Sikka's New AI Venture Hang Ten Bags $32 Mn
Vishal Sikka's new AI venture Hang Ten Systems raised $32M seed funding backed by Mayfield and Aramco Ventures to challenge legacy IT services with AI-native delivery. 💰
JustAI Bags $17 Mn To Scale Its Agentic Martech Stack
Indian AI-powered martech startup JustAI raised $17M Series A to expand its agentic marketing automation platform across enterprise clients. 💰 📱
📖 Beyond the headlines
How Shopify built an AI stack that doesn't care which models survive — As model churn accelerates — providers shutting down versions, changing APIs, going dark — Shopify's LLM proxy architecture represents the playbook every enterprise should be studying right now. For a senior strategist, this is a rare peek into how a production-scale company has solved AI vendor lock-in in a concrete, replicable way before it became a crisis. Read →