In partnership with
In Today’s Issue:
📉 Microsoft is lowering sales targets
🧑💻 A new & more powerful coding model is now available for free
💰 Nvidia has so much cash that antitrust restrictions are blocking it
🏆 The Gemini Deep Think model is crushing brutal tests
⚔️ A new bipartisan bill seeks to block the export of Nvidia's most powerful chips to China
✨ And more AI goodness…
Dear Readers,
What happens when your AI stops blurting out the first answer and actually sits down to think? Today’s main story dives into Google’s new Gemini 3 Deep Think mode – a parallel, multi-hypothesis reasoning engine that behaves more like a lab team arguing at a whiteboard than a chat bot, with big implications for research agents, coding workflows and how we judge “real” AI intelligence. Around it, we look at Microsoft’s AI growing pains as Azure Foundry hits the reality wall, Cursor’s free rollout of OpenAI’s new Codex-style model for autonomous coding, and Nvidia’s very 2025 problem of having too much cash in an AI-saturated market. Plus: a sharp look at US attempts to slow China’s AI rise via fresh chip export rules, a video pick on whether today’s AI spending spree is sustainable, and a rumor section that keeps you close to the frontier—if you want the signal behind the AI noise, this issue is for you.
All the best,
Kim Isenberg
Microsoft Navigates AI Growing Pains
Microsoft shares experienced a dip following reports that sales teams missed aggressive growth targets for the new Azure AI Foundry platform, sparking rumors that the tech giant is quietly lowering its internal quotas. While Microsoft firmly denied these cuts, the situation highlights an exciting industry shift: we are moving from the "hype cycle" into the "value phase," where enterprises are demanding tangible ROI and safety before scaling their AI adoption!
Cursor Launches OpenAIs New Codex Model for free
Cursor has officially integrated the new Codex model, collaborating directly with OpenAI to optimize it specifically for their agentic coding harness. This update promises to significantly enhance how the AI handles complex, multi-step development tasks, making the coding experience smoother and more autonomous for engineers. The new model is available to use for free until December 11th, offering a crucial window for developers to benchmark its performance without any cost.
Nvidia Confronts Massive Cash Surplus
Nvidia’s unprecedented financial growth has created a unique strategic challenge, with cash reserves swelling to $60.6 billion and nearly $600 billion in free cash flow projected over the next three years. Restricted by antitrust regulators from pursuing large corporate acquisitions, the company is pivoting its capital allocation strategy toward aggressive shareholder returns and strategic minority investments.
Anthropic C.E.O.: Massive A.I. Spending Could Haunt Some Companies
Gemini 3 Deep Think: The AI That Finally Pauses to Ponder!
The Takeaway
👉 Deep Think gives Gemini 3 a parallel, multi-hypothesis reasoning mode that’s better suited for hard math, science and logic than traditional one-shot prompting.
👉 For AI builders, the upgrade means more stable long-horizon agents and fewer broken chains in complex workflows like research, planning and code refactoring.
👉 Ultra subscribers effectively get early access to a “reasoning lab assistant” that can stress-test ideas instead of just generating fluent text.
👉 If this architecture generalizes, expect other labs to copy the pattern, and for “how well can it really think?” to become the new frontier benchmark.
Google just flipped the switch on Gemini 3 Deep Think, a new reasoning mode that turns the Gemini app into something closer to a patient, multi-step problem-solving partner than a chatty assistant. Available to Google AI Ultra subscribers, Deep Think spins up multiple “lines of thought” in parallel, testing different hypotheses before committing to an answer - more like a lab team arguing over a whiteboard than a single model guessing token by token.
In internal benchmarks, it pushes past standard Gemini 3 Pro, scoring 41% on Humanity’s Last Exam, 93.8% on GPQA Diamond and 45.1% on ARC-AGI-2 with code execution, all on notoriously brutal reasoning and science tests. For builders, that means more reliable agents for complex workflows; for researchers, a tool that can actually grind through intricate math, proofs or experimental designs instead of hallucinating halfway through. The real story isn’t just better scores, but a visible shift toward AI systems that reflect, revise and justify their reasoning, the core behavior we’ll need for trustworthy, expert-level AI.
Why it matters: Deep Think is one of the clearest signs yet that frontier models are moving from “smart autocomplete” toward structured, debate-like reasoning that can survive contact with real-world complexity. If it works as advertised, it could become a template for the next generation of research copilots, autonomous agents and safety-critical AI tools.
Sources:
🔗 https://blog.google/products/gemini/gemini-3-deep-think
Find customers on Roku this holiday season
Now through the end of the year is prime streaming time on Roku, with viewers spending 3.5 hours each day streaming content and shopping online. Roku Ads Manager simplifies campaign setup, lets you segment audiences, and provides real-time reporting. And, you can test creative variants and run shoppable ads to drive purchases directly on-screen.
Bonus: we’re gifting you $5K in ad credits when you spend your first $5K on Roku Ads Manager. Just sign up and use code GET5K. Terms apply.
US Rewrites AI Chip Rules in order to slow down China
Washington just put a big question mark over who gets to train the next generation of frontier models. A bipartisan group of US senators has introduced the “Secure and Feasible Exports Chips Act” to block Nvidia from selling its most advanced AI accelerators - including H200 and Blackwell-class chips - to China for 30 months. The move comes after signals that the White House might allow certain exports, triggering a backlash from lawmakers who see cutting-edge GPUs as strategic infrastructure, not just silicon.
At its core, this is a clash between two visions of AI: one where the US leans into free-market competition, and one where AI chips are treated like dual-use tech that must be tightly controlled. Nvidia’s Jensen Huang has argued that China won’t accept crippled chips and that shutting US vendors out simply accelerates local champions and costs American jobs, while senators frame the bill as drawing a hard line in an AI “values contest” with an authoritarian rival. For the AI ecosystem, it’s a live test of how much geopolitical friction the compute supply chain can absorb before innovation patterns fundamentally shift.
Despite all the embargoes, China recently managed to catch up in the race for the best AI; despite this, discussions about liberalizing chip exports continued. This decision has further hampered that progress.
