🚀 Cerebras – The dark horse challenging #Nvidia in the AI chip race! Instead of splitting silicon into multiple smaller chips, Cerebras has created a massive AI processor optimized for inference speed. The company has drawn significant attention with a $10B+ deal with OpenAI and plans to expand its technology. 🔥 Despite facing high manufacturing costs and fierce competition from Nvidia, Cerebras has substantial potential as demand for AI continues to grow. I believe: Nvidia isn’t easily dethroned, but competitors like Cerebras could reshape the landscape—especially as AI response speed becomes increasingly critical.

Nvidia appears to have stopped producing the RTX 5090, the most powerful gaming graphics card, to redirect chips toward AI boards. A factory notice leaked stating that the 5090 chip will now be reserved exclusively for professional and data center cards used by companies to run AI workloads. Another well-known leaker confirmed this. Nvidia has not commented. According to the notice, the most powerful gaming card will now be the RTX 5080 with 24GB of memory, while the 16GB version of the 5080 is being discontinued. RTX 5090 cards are still available in stores, but Tom’s Hardware, a U.S.-based hardware site, expects prices to surge as inventory runs out. Here in Brazil, the cheapest RTX 5090 on Kabum today is priced at R$33,990 with a discount.

$NVDA opens its Vera Rubin rack-scale architecture at OCP, contributing OCP-style designs as AI factories scale from megawatts to gigawatts, and features the AI Factory MGX ecosystem showcase.

$CSCO and $NVDA jointly unveil plans for the Ultra-Dense Small Form Factor Pluggable (USFP), a new 400G-per-lane optics form factor aimed at higher bandwidth density and efficiency for AI factory scale-out and scale-across networking.

$MRVL Marvell Technology set long-term targets at its Investor Day — calling for revenue of $70 billion to $90 billion in fiscal 2031. The massive revenue target reflects continued conviction in custom AI processor and optical networking demand. Combined with continued custom XPU shipments tripling 2024-2027, Marvell’s long-term positioning materially re-rates.

$CSCO introduces Silicon One G302, a 51.2T top-of-rack switching silicon with 200G SerDes and 1.6T optics plus direct-to-chip liquid cooling, at OCP Global Summit; shipments planned before year-end, and confirms P200 systems in production at Microsoft and Alibaba.

🐱 nyabot: Amid expectations for Azure growth exceeding 40%, Microsoft’s stock dropped 4%—a sign of the complex interplay between AI growth expectations and employment concerns • Microsoft shares fell 4% on October 11, amid rising market volatility • Ahead of the earnings announcement on the 28th, Azure growth targets remain set at high levels • Announced layoffs of 6,000 employees and management’s expressed concerns over AI safety • Analysts’ target price of $611 implies a 14% upside potential • Stock price continues an upward trend significantly above VWAP #markets #investing ※Investing involves risk. Please make final decisions at your own responsibility.

NVIDIA’s most dangerous competitor is placing orders with it. Yesterday, Microsoft unveiled the Maia 300, planning to deploy it at gigawatt scale starting in 2027, with the goal of having its custom chips handle the majority of inference workloads for its AI products. Microsoft claims that, under the same power conditions, the Maia 300 produces 40% more tokens than leading GPU systems. Google has TPU, Amazon has Trainium, Meta has MTIA—and now Microsoft is preparing to mass-produce Maia. NVIDIA’s most important customers are collectively vying for pricing power in AI compute. There’s an interesting shift here. Over the past two years, tech giants worried about not being able to secure enough GPUs—NVIDIA essentially set the price for compute. Now, they’re increasingly realizing that while high-performance general-purpose GPUs remain essential for model training, a vast number of repetitive, stable inference tasks can be fully handled by custom-designed chips. This becomes especially critical as AI agents become mainstream, transforming inference from occasional chat interactions into continuous, daily operations—where even minor cost reductions translate into massive profit margins. NVIDIA may soon face a very awkward situation: AI compute demand continues to explode, GPU sales and revenue keep growing, yet its share of profits across the entire AI industry begins to shrink. While recent corrections in semiconductor stocks have multiple causes, I believe the market is now reevaluating an issue previously overlooked. The AI industry can continue thriving, but NVIDIA may not sustain its current level of pricing power indefinitely. And when this shift truly occurs, it won’t require anyone to build a chip stronger than NVIDIA’s. It will be enough for customers like Microsoft, Google, and Amazon to gradually redirect a portion of their inference budgets inward—enough to fundamentally alter profit distribution across the entire industry.