Tech Brew Ride Home - Intel Punches Back

Episode Date: October 9, 2025

Intel tries to punch its way back to relevancy with the release of Panther Lake, the first chip built on its 18A process. Is Cursor so successful it’s about to rocket to a $30 billion valuation? Big raise for a US Deep Seek competitor? And are we starting to get the first models moving beyond attention based architecture? Here is Panther Lake, Intel’s 2026 laptop chip with next-gen graphics (The Verge) Intel Debuts New Technology in Make-or-Break Moment for CEO’s Turnaround Bid (Bloomberg) Discord says 70,000 users may have had their government IDs leaked in breach (The Verge) Cursor-Maker Anysphere Considers Investment Offers at $30 Billion Valuation (The Information) Reflection AI, an A.I. Model Start-Up, Raises $2 Billion (NYTimes) Samsung AI researcher's new, open reasoning model TRM outperforms models 10,000X larger — on specific problems (VentureBeat) Microsoft Tries to Catch Up in AI With Healthcare Push, Harvard Deal (WSJ) Learn more about your ad choices. Visit megaphone.fm/adchoices

Ready to soundtrack your summer? With Red Bull Summer All Day Play, you choose a playlist that fits your summer vibe the best. Are you a festival fanatic, a deep end DJ, a road dog, or a trail mixer? Just add a song to your chosen playlist and put your summer on track. Red Bull Summer All Day Play.

Red Bull gives you wings. Visit Red Bull.com slash Bright Summer Ahead to learn more. See you this summer. Welcome to the TechBrew ride home for Thursday, October 9th, 2025. I'm Brian McCullough today. Intel tries to punch its way back to relevancy with the release of Panther Lake, the first chip built on its 18A process, is cursor so successful it's about to rocket to a $30 billion valuation, big raise for a U.S. Deep Seek competitor, and are we starting to get the first

models moving beyond attention-based architecture? Here's what you miss today in the world of tech. When your business evolves, so does the risk of data loss. But with VIM, your data is always on the map. Partner with VIM for coverage that keeps you moving and get protection for workloads of all shapes and sizes, even the ones you haven't created yet, so you can stay resilient as you scale. With VIM, it's all good. Get workload coverage that works for your business at VIM.com. That's VEEAM.com. Big positive news from Intel today for a change. Intel this morning unveiled Panther Lake or Intel Core Ultra Series 3, the first chip built on its 18A process coming to laptops and handhelds in late this year or early next year. Intel says 18A-based Panther Lake chips can deliver 50% more performance

at similar power to Lunar Lake chips. Intel CTO Sachin Kadi said 18A and Panther Lake are, quote, foundational to our future, and manufacturing is quote mission critical, not just for Intel, but for the United States. Quoting the verge. Panther Lake is the most important chip Intel has made in years. It's the one that will tell the world if a smaller, more focused Intel can still meet or beat the competition from Apple, AMD, and Qualcomm. It's the first on Intel's 18A process, the one Intel CEO Pat Gelsinger claimed would

write the ship and make Intel a manufacturing leader before he was unceremoniously shown the door last December. That's a lot writing on one piece of silicon when it arrives in late 2025 and early 2006. But what is Panther Lake, aka Intel Core Ultra Series 3, actually going to bring to a laptop or handheld near you? Well, more battery life, more performance, more gaming graphics, more affordability. That's the idea, anyhow. Oh, and Panther Lake isn't just for thin and light machines. This time Intel has built three different flavors of this chip, so it can replace both the lighter weight Lunar Lake and the heftier Arrow Lake H you find in more powerful laptops right now. Intel,

will sell eight and 16 core CPUs, each with four brand new XE3 graphics cores, as well as a 16 core CPU with 12 XE3 graphics cores and 12 ray tracing units, the most integrated graphics horsepower. It's shipped to date, and a preview of what might come if Intel continues to ship desktop gaming GPUs. With the last generation, we gave you a dilemma, says Intel chief CPU architect Stefan Robinson. You could buy a lunar lake and get fantastic battery life, or you could buy Arrow Lake that had more throughput. Now Intel is trying to solve that dilemma, he tells me. Although Panther Lake abandons the onboard memory that helped make last year's Lunar Lake so efficient and adds the low power e-cores that were holding Meteor Lake's battery life back,

Intel claims you'll actually see up to 10% lower power than Lunar Lake across the entire chip. and Robinson tells me that should genuinely mean better battery life than Lunar Lake in real-world use cases, including Microsoft Teams. Diving into theoretical workloads, the chips sound even more impressive. Intel's brand-new Cougar Cove, Performance Cores, P-Cores, and Darkmont E-Cores on the 18A process, let Intel claim 40% lower power at similar single-threaded performance or 50% more multi-threaded performance at similar power, seemingly leaving Lunar Lake in the dust. Meanwhile, Intel's new graphics claim to offer over 50% more GPU power than their predecessors too. But it will depend on which of

Intel's three new kinds of chips you're buying, of course. While the top flight Panther Lake with 12XE3 GPU cores might offer 50% more GPU power than the Lunar Lake chip with 8xE2 GPU cores, it isn't yet clear to me how the new chips with just four XE3 GPU cores might compare or whether one might have a battery life advantage over another. That said, even the 12xE3 core variant shouldn't be too beefy to fit into a handheld like the MSI Claw 8 that surprised me last month. Intel fellow Tom Peterson suggested as much on stage during Intel's architecture event, and when it does, it should have more stable performance because of a new intelligent bias control V3, which sends hints to Windows to offload gaming tasks to E-Cores instead of P-Cores,

so the system can divert that extra power to graphics where the games really need it. We're more heavily relying on our e-cores for gaming because they're beefy ecores, and that frees up more power for the GPU, Robinson tells me, and quote. And Intel really wants you to know that this is being made in the USA. Quoting Bloomberg, Intel executives explain the benefits of their latest offerings in presentations hosted near a new factory at its Osotilo site in Arizona, known as Fab 52. the facility is the first to go into mass production with the 18A technique. Intel also is shouldering the costly burden of trying to update its factories.

Fab 52 alone required more steel than the Eiffel Tower to build and contains machines that cost hundreds of millions of dollars. Over three days of presentations at the event, Intel executives repeatedly asserted that 18A is the most advanced chip production technology developed and deployed in the U.S. That Made in America, Spirit lines up with a Trump administration push to bolster domestic manufacturing and reduce reliance on factories in East Asia. The company can make 18A equipped factories pay for themselves by producing chips for its own needs, Intel said, but the next phase, a technology called 14A, will require outside customers

and a high volume of orders to be cost-effective. Getting the new 18-A chips to market in products that demonstrate better performance for consumers will be the first step toward reestablishing Intel's credibility, according to Kevin O'Buckley, general manager of the company's foundry business. Don't trust us until we can do that, says, said the executive who's tasked with persuading other companies, including rivals to use its factories, we know we have a long way to go to deliver trust for our customers, end quote. Something, something, my threshold for covering breach stories that I always tell you about.

Discord estimates that 70,000 of its users government ID photos may have been exposed in a 1.5 terabyte data breach of third-party customer service provider Zendesk, which happened on September 20th. Quoting the verge, a tweet by VX Underground said that the company was being extorted over a breach of its Zendesk instance by a group claiming to have 1.5 terabytes of age verification related photos, 2,185,151 photos. When we asked about the tweet, Discord spokesperson New Wexler shared this statement. Following last week's announcement about a security incident involving a third-party customer service provider, we want to address inaccurate claims by those responsible that are circulating online. First, as stated in our blog post, this was not a breach of Discord, but rather a third-party service we used to support our customer service efforts. Second, the numbers being shared are incorrect and part of an attempt to extort a payment from Discord. Of the accounts impacted globally, we have identified approximately 70,000 users that may have had government ID photos exposed, which our vendor used to review age-related appeals.

Third, we will not reward those responsible for their illegal actions. All affected users globally have been contacted, and we continue to work. work closely with law enforcement, data protection authorities, and external security experts, end quote. Peak pollination season, and my business is scaling fast. To keep the nectar flowing, I need a phone plan with top priority data speeds. That's why I chose GoogleFi Wireless. My connections stay strong even when the hive is buzzing.

Plus, unlimited plans started $35 a month. Now, that's a deal that doesn't stay. Explore GoogleFi Wireless plans today. Plus taxes and government fees. Google Fiwifilis is not subject. to data traffic deprioritization during times of high network usage. Study and play. Come together on a Windows 11 PC.

And for a limited time, college students get the best of both worlds. Get the Unreal College deal, everything you need, to study and play with select Windows 11 PCs. Eligible students get a year of Microsoft 365 premium and a year of Xbox GamePass Ultimate with a custom color Xbox wireless controller. Learn more at Windows.com slash student offer.

While supplies last, ends June 30th, terms at AKA.m.m.m.m. Ambition comes in all shapes and sizes. At First Citizens Bank, we roll with your goals because we're built for what you're building. Fit for your ambition for Citizens Bank. A couple of interesting raises for you here. The information says any sphere, the company behind Vibe Coding Sensation Cursor, is considering investment offers at around a $30 billion valuation. Cursor apparently generates $500 million in ARR as of June, the third highest for any AI app. Quote, the funding discussions signal a remarkable run-up in any sphere's price from the start of this year when investors, including Thrive Capital and Andreessen Horowitz, valued the San Francisco startup at around $2.5 billion. The company then raised money again in June at a $9.9 billion valuation. Over the summer,

the startup also received an investment offer at a valuation of between $18 to $22 billion, but declined to accept it, according to people involved in the conversations. The company, which employs more than 150 people, previously raised more than $1 billion from investors, including Andresen Horowitz, Thrive Capital, a Cell, and DST Global. After Open AI and Anthropic, Cursor is generating more revenue than any other AI-native application generating $500 million in annual recurring revenue as of June, up 10 times from its revenue pace in November, according to the information's generative AI database. It's expected to generate $1 billion in annual recurring revenue by the end of this year, Bloomberg has reported.

AnySphere is increasingly competing with the biggest AI companies that have also developed their own coding tools. Google, which has several coding assistants earlier this year, agreed to pay $2.4 billion to hire the founder and some senior staff of coding tool developer Windsurf, AI coding startup cognition, bought WinSurf's remaining assets. Phanisphere has been developing its own AI models, which could cut the costs it pays Anthropic and OpenAI for their models and has been trying to focus more on large corporate customers rather than individual users. Its corporate customers include Figma and Stripe, according to Cursor's website. On Wednesday, NVIDIA CEO Jensen Huang said on CNBC that CERC is his favorite

enterprise AI service and that every Nvidia engineer is assisted by AI coders. The chip giant company also uses several coding assistance, including cursor, according to a person with knowledge of the matter. In recent months, EniSphere held preliminary discussions with AI coding model developers, including XAI, OpenAI, and Anthropic about potentially licensing the data it gets from customers, such as what coding suggestions they approve or what are most common requests from developers. Besides paying for models to run its assistant, the company has expanded to multiple offices in San Francisco and has opened an office in Manhattan, according to job postings end quote. And Reflection AI is developing an open source AI model to compete with Deepseek,

and it has raised $2 billion led by Nvidia, valuing the company at $8 billion up from $545 million just back in March. Quoting the Times. The move shows that investors are willing to invest not only in the proprietary software sold by Open AI and Google, but also in open source models which are free to share and modify. Skeptics of of open source AI say the approach could cause significant harm. Yet Reflection AI argues that the United States needs an equivalent to deep-seek an open-source model developer whose software can compete with top-closed source models to maintain technological superiority around the world. Existing Western open-source AI models are underperforming DeepSeek and other Chinese rivals,

which may lead to greater adoption of Chinese-created models, according to Misha Laskin, reflection AI's co-founder and chief executive. Among the Western companies making open source models are meta, mistral AI of France, and even Open AI, though that company is primarily focused on proprietary software. There's a deep seek-shaped hole in the U.S., which I think is what makes it critical for a lab like ours to exist, Mr. Laskin said in an interview, comparing the situation to the Cold War space race. But developing any AI model, open source or not, requires ever higher amounts of computing power and research talent, in short, money. That is why Reflection AI raised $2 billion just seven months after raising $130 million in March. In recent months,

Reflection AI held talks with potential investors and closed the round within a matter of weeks. Mr. Laskin said, Nvidia wrote the biggest check, $800 million, and had several engineers work with Reflection AI to optimize its most recent generation of AI chips, according to a person with knowledge of the matter, who spoke on the condition of anonymity because they were not authorized to discuss the deal's details. Other participants in the round included existing backers like Lightspeed Venture Partners and Sequoia Capital. New investors included DST, the former Google Chief Executive Eric Schmidt, and 1789 Capital, the investment firm that counts Donald Trump Jr. as a partner, end quote. Well, speaking of smaller models, Samsung has introduced the tiny

recursion model, a 7 million, and I said million, not billion, 7 million parameter model, that can outperform LLM's 10,000 times larger, like Gemini 2.5 Pro or O3 Mini on specific problems. Quoting Venture Beat. The goal is to show that very highly performant new AI models can be created affordably without massive investments in the graphics processing units and power needed to train the larger multi-trillion parameter flagship models powering many LLM chatbots today. The results were described in a research paper published in Open Accessarchive.org entitled, Less is More, Recursive Reasoning with Tiny Networks. The idea that one must rely on

massive foundational models trained for millions of dollars by some big corporation in order to solve hard tasks is a trap, wrote Alexia, Jolikour-Martineau, senior AI researcher at Samsung's Advanced Institute of Technology on the Social Network X. Currently, there is too much focus on exploiting LLMs rather than devising and expanding new lines of direction, end quote. Jolacure Martinot also added, With recursive reasoning, it turns out that less is more, a tiny model pre-trained from scratch, recursing on itself and updating its answers over time can achieve a lot without breaking the bank. TRM's code is available now on GitHub under an enterprise-friendly, commercially viable MIT license,

meaning anyone from researchers to companies can take, modify it, or deploy it for their own purposes, even commercial applications, end quote. TRM is apparently a radical simplification of HRM or the technique called Hierarchical Reasoning Model. Instead of two-cooperating networks and fixed-point math, it uses a single two-layer model that recursively updates a latent state and its answer until stable, plus a lightweight halting rule. Iteration substitutes for depth and size, delivering high accuracy with a tiny parameter count, rivaling DeepSeek R1, Gemini 2.5 Pro, and O3Mini. The minimalist design generalizes better, adding layers often overfits small data sets for small-fix context. Sudoku, for example, a simple MLP apparently outperforms self-attention. For larger grids, attention still helps, however.

Finally today, the journal says Microsoft is planning a major health care push for co-pilot in partnership with Harvard Medical School as it seeks to reduce its dependence on OpenAI. I quote. Mustafa Suleiman, chief executive of Microsoft AI, has made healthcare a focus of the division's efforts as he has increased staffing at an internal AI lab that competes with OpenAI. In June, the company which employees clinicians said an AI tool it developed can diagnose disease at a rate four times more accurate than a group of doctors and do so at a fraction of the cost. Despite a tentative agreement announced last month to extend the partnership between Microsoft and OpenAI, there remains urgency within Microsoft. on building up a measure of technological independence from OpenAI, the people familiar with the matter said. Last week, Microsoft chief executive Sachin Adela said he would hand off some duties to a deputy so he could concentrate on the company's biggest AI bets. Microsoft, which created its division dedicated to consumer AI and research in 2024, is training its models with the goal of eventually replacing workloads from OpenAI, the people said. Achieving that might take years. The company has said that OpenAI will continue to be our partner on frontier models and that its philosophy is to use the best models available.

Microsoft trails OpenAI in Consumer AI. The co-pilot smartphone app has been downloaded 95 million times, while ChatGPT has been downloaded more than a billion times, according to Censor Tower, a research firm. Microsoft in August said that it had started to test publicly a homegrown AI model that could be used for its co-pilot chatbot. researchers and engineers, especially those recently hired away from Google's deep-mind AI lab, are focused almost entirely on advancing Microsoft's own models. Already, Microsoft is using non-open AI models for some of its other software. It now deploys models from Open AI's rival Anthropic to power AI tools within its 365 products, end quote. Nothing more for you today. Talk to you tomorrow.