Meta

Morgan Stanley, JPMorgan, Bank of America and Citi are now blocking a category of customer that didn't exist two years ago: the customer's own AI agent. Axios reported this week that unvetted third-party agents are showing up inside live…

🇧🇷Brazilian President Lula da Silva’s campaign has petitioned the top electoral court to disqualify right-wing rival @FlavioBolsonaro ahead of Sunday’s runoff, accusing @Meta of restricting Lula content on @Instagram, @Facebook and…

Sam Altman on the amount of compute needed if 50% of the world were using personal AI assistants to its full capabilities: "How many GPUs do you want working for your personal assistant? When people talk about the global market for GPUs…

$META Does timing matter? Honestly, not that much to me. It matters a lot in the short term and very little in the long term. This week is a perfect example. META posted a strong quarter (time the rip +5%), but the guidance didn’t give us…

$META is down over 6% today… Sucks when bad macro and earnings collide. https://t.co/Zaw75v36RR

Quite surprised that AI stocks are red considering how strong SpaceX and Meta's earnings were. Fundamentally, both proved that there's zero issues on the demand-side which is a positive step in helping sentiment. It therefore seems like…

The spend limit on most internal agents today lives inside a system prompt, which makes it a suggestion, not a control. Walmart shipped what an actual control looks like this week, and it has nothing to do with retail specifically. Agent…

Agent Gateway is Walmart's answer to a problem every site with a catalog now has: give agents a real way to read your data, or watch them scrape it blind and break the moment your layout changes. Walmart built a dedicated MCP server this…

Four labs shipped a standing AI agent inside ten weeks of each other, and that's not a coincidence, it's a race nobody wants to come in fourth. Grok Bot opened early beta on August 11. Meta Muse launched September 8, free to start, with a…

Three different labs now ship an agent that messages you before you ask it to. Anthropic rolled out Claude Duet this week, a research preview on Pro and Max plans that runs on Opus 5.5, reads your calendar and inbox every few hours, and…

BofA: TrendForce states Nvidia's AI-server share moderates from 66.9% in CY26 to 62.6% in CY27 as “Others” rises from 26.0% to 29.6%. Custom ASICs increase from 26.1% to 29.6% of AI server shipments, led by hyperscaler programs. $NVDA…

$META: "we want to make sure that we have built enough compute to be able to serve what we think is going to be a billion-plus people." https://t.co/ss2AVL9KQX

$META on Muse: "I think one of the biggest constraints so far is that because this is a compute-intensive service to run, we have not launched it globally yet." Meta wants more compute, not less!

$SPY rose about 0.5% today to around $773, bouncing from an early low of $768.41 as chip and AI stocks led the market higher. Here are the top stories you need to know. 1. $SMCI jumped about 10% after raising its fiscal first quarter…

Another major shortage is starting to develop across the AI supply chain. We're talking about multilayer ceramic capacitors, or MLCCs, which are tiny components used in everything from AI servers and GPUs to smartphones, networking…

Expert call covering GUC – Google CPU & TPU v10 ASIC outlook, Microsoft Maia 200 ramp, TK1 vs. TK3 margin dynamics, TSMC capacity support, and 2027 hyperscaler volume trajectory. $MFFT $META $GOOGL https://t.co/VvdNhBSJFv

Nomura: global data center power usage by application $NVDA $AMD $VRT $GOOGL $MSFT $META $AMZN

NVIDIA already has TensorRT-LLM. It also has Dynamo. This week it shipped a third local inference server anyway, and the part worth sitting with is why. Flare runs on Rust, not Python, and it's built narrow: just the GPUs with the…

Morgan Stanley sees the cost per new MW peaking at $51.8 million in FY27, before sliding to $47 million in FY29E. $MSFT $NVDA $AMZN $GOOGL $META https://t.co/IC6zlAbmhs https://t.co/8zXgV111pw

Nvidia is now writing checks to keep its own customers alive. According to The Information, $NVDA is in talks to invest hundreds of millions of dollars in Firmus, the Australian AI cloud provider that just pulled its IPO. Firmus was trying…

BREAKING NEWS🚨🚨: TODAY META CONFIRMS THEY ARE RUNNING TPU INFERENCE IN PRODUCTION. What does this mean for the CUDA moat? For those who read our public newsletter, this is not a surprise, as on Nov 28, 2025, we publicly wrote that Meta…

Once again, the GPU depreciation bear thesis is dead. Elon on the $SPCX call: "The value of compute per watt will increase over time, not decrease... that's why you see things like even old computers like H100s renting for more than they…

Every hyperscaler racing to dent Nvidia's lead just converged on the same move within a day of each other. Google open sourced TorchTPU at the PyTorch Conference in San Jose this week, co-built with Meta's PyTorch team, letting existing…

$SPY rose about 0.2% today to around $768, bouncing from an early low of $764.25 as the Dow led and tech lagged. Here are the top stories you need to know. 1. South Korean neocloud Elice plunged 32% in its trading debut, dragging the KOSPI…

CEO of Google, Sundar Pichai, on compute demand and AI bubble possibility: Q: Ben Horowitz relayed Andy Jassy had something like compute is sold out till 2028, are you seeing the same thing? "If today a customer is coming out and asking us…

CEO of Lumentum, Michael Hurlston, on the impact of yields on capex spend: "The CFO of Alphabet came on just a couple of days ago and was talking about their capex going up year over year. For the most part, I think, the hyperscalers…

$MU $SNDK $SKHY $NBIS Nomura’s memory team just put out a really interesting note on the AI scare, the numbers underneath are some of the strongest I’ve seen in this drawdown. Nomura sees AI going from 10GW of data center power in 2025 to…

Expert Takeaways: Data Centers 1. Open-Source Models Closing the Gap > Open-source models are catching up to state-of-the-art frontier models faster than expected, driven by lower costs and improved performance. > Starting in Q2, customers…

Nomura: data center capacity to expand to 74GW in 2028F, up from 26GW in 2025. Of the 74GW, 48.5GW coming from the US big tech. $SPCX $ORCL $CRWV $META $MSFT $AMZN $GOOGL

AI data centers are now running into a FIBER shortage. TrendForce estimates the global optical fiber supply gap will reach 180 million fiber kilometers in 2026. Factory utilization is around 92% and inventories have fallen to roughly one…

Picture circus-tent structures going up over a one-gigawatt construction site in New Albany, Ohio, built that way on purpose because tents cut deployment time by 70 to 80% versus permanent buildings. That is Prometheus, and it is already…

$SPY rose 0.48% to $766.91 today, with AI and chip stocks leading a bounce after last week's selloff. Here are the top stories you need to know. 1. Chip stocks bounced back as the Philadelphia Semiconductor Index rose about 2% to snap an…

Mizuho: Sandisk > NAND pricing is projected to remain solid through 2027E. > Major eSSD ramp driven by hyperscalers: META, GOOGL, and OCI. Acceleration in KV cache and AI SSDs expanding into 2027/28E. > NBMs provide long-term visibility…

We're about to witness the BIGGEST infrastructure spending boom in human history. According to Morgan Stanley, hyperscalers and neoclouds are projected to spend a staggering $1.58 TRILLION in 2027 and $1.81 TRILLION in 2028 when operating…

$NVDA $META Meta's Million-GPU Monster Is Hungry Meta just published how it runs Prometheus, its push toward a million-GPU cluster in Ohio. The number that stuck with me is buried in the reliability section. To run a job on 32,768 GPUs,…

A £2 billion data centre site in Loughton, Essex was supposed to go live in 2027 with roughly 23,000 Nvidia chips inside it. This week it came out that Microsoft walked away from its anchor tenant commitment there, and the reason has…

Goldman Sachs has raised the global servers TAM by 31%/45% in 2027/28E to ~$1.1 trillion and ~$1.6 trillion, respectively. New AI servers shipment forecasts imply AI chip demand of 18m/29m/38m in 2026-28E. $DELL $HPE $NVDA $AMD $GOOGL…

UBS has revised their memory spend % of top 12 hyperscaler capex from 63% to 66% in 2027, or ~$961 billion. For reference, top 12 hyperscaler capex in 2026 is projected to be roughly $1 trillion. Can you imagine...2027 capex on MEMORY…

Earnings season is back ‼️ Many AI names are reporting this week. Tuesday: $SPCX Wednesday: $META $LRCX $TXN Thursday: $NOK $BE $MXL $WDC

BofA on AI comparison to the dotcom: "Unlike the idle fiber of 2000, deployed AI infrastructure runs near 100% utilization, and even older A100/H100/B200 (2020/2022/2024 launch, respectively) GPU rental rates are up 12% -37% YTD, a sign of…

J.P. Morgan: AI CapEx > 2026 Capex Forecast: Growth estimate raised from +80% to +85% YoY. This adds over +$270B in incremental spending (a new record step-up, beating 2025). > 2027 Capex Forecast: Growth estimate raised from +50% to +70%…

NPO and CPO adoption by customers presented by LightCounting at OCP. $AMZN $GOOGL $META $MSFT $ORCL

Meta published the first hyperscaler report on running AMD GPUs in large-scale training. Rather than porting from CUDA, it revised model structures around MI350X's 1.6x HBM capacity vs B200: up to 1.5x throughput over the ported baseline.…

UBS: AI power TAM growing to $91.1 billion by 2030E, up from $17.7 billion in 2026E. Of the $91.1 billion, $49 billion dedicated to GPU power TAM, while the other $42 billion for ASIC. $GOOGL $MSFT $META $ORCL $NVDA $AMD $AMZN

According to Korean media, Microsoft, Google, Amazon and Meta are speeding up their in house AI chips. That could open up a SECOND group of HBM buyers beyond $NVDA. Microsoft's Maia 300 is expected to ship more than 1 MILLION units next…

‘Doing What It Takes’ in AI. OpenAI, Meta & TSMC. ARD #185 …OpenAI’s ad sprint, Meta’s open Muse on Apple’s Macs, TSMC on its own terms, and Apple’s Ternus era. $TSM $META $AAPL Full writeup and source links: https://t.co/phJT1JuRTJ…

HSBC: ASIC will account 57% of total HBM demand by 2028e (from 31% in 2025) $GOOGL $MU $MRVL $META $MSFT $AVGO $AMZN

🎙️ “Hey Luna, take this block. Now give it back. Follow me.” In this demo by @AI_with_Eric, his local AI assistant Luna brings together #LocalAI and #robotics to turn voice commands into physical actions. With the #reBotArmB601, #Qwen 3.8…

Bloomberg shared this visual showing the estimated cost of building a data center. Of the estimated $29.2 billion, $17.1 billion (or 58.5%) is dedicated to chips & servers. $NVDA $DELL $GEV $AMD $GOOGL $META $MSFT https://t.co/ay4TmgkRo4

Expert Takeaways: Meta's Optical Demand > Meta signed a $6B deal with Corning for MPO fiber cables to power its next-gen AI data center infrastructure. > Zhongji, Eoptolink, and Cohr lead Meta’s optical supply chain, with new entrants…

J.P. Morgan is forecasting a ~300GW shortfall in global AI compute by 2031E. $NVDA $AMD $GOOGL $MSFT $META $AVGO $AMZN https://t.co/iZrvcU2svc https://t.co/yzzhhhEN7G

Top stories in AI today: - Meta’s Muse Wonder lands at the frontier - Andrej Karpathy sketches AI’s ‘sharp funnel’ - Put Meta’s new Muse Wonder to work - Microsoft’s Maia 300 takes aim at Nvidia https://t.co/SpJmTKfs96

𝟏𝟖 𝐭𝐞𝐜𝐡 𝐠𝐢𝐚𝐧𝐭𝐬. 𝐓𝐡𝐫𝐞𝐞 𝐬𝐭𝐫𝐚𝐭𝐞𝐠𝐢𝐞𝐬. 𝐎𝐧𝐞 𝐫𝐚𝐜𝐞 𝐭𝐨 𝐠𝐢𝐯𝐞 𝐀𝐈 𝐚 𝐛𝐨𝐝𝐲. @OpenAI, @amazon, @Apple, @AntGroup and @XiaomiTechnolo1 are building or pursuing their own humanoid bodies. @GoogleDeepMind,…

$SPY finished nearly flat today, down about 0.06% and closing near $778 after dipping as low as about $777. Here are the top stories you need to know. 1. $MSFT announced Maia 300, calling it its biggest bet yet on custom silicon, and said…

Microsoft Maia 300 is a hiring story as much as a chip launch. KORE1's take: the bottleneck isn't the silicon, it's the ML compiler engineers who make it useful. https://t.co/tFzlBZpkzh https://t.co/EDLcmTKqMP

Who designs the chip after Microsoft Maia 300? The same small pool Google, Amazon, Meta, and OpenAI are recruiting from. KORE1 breaks down which chip roles get squeezed first. https://t.co/tFzlBZpkzh https://t.co/liozv8z01k

Microsoft Maia 300 enters production in 2027, and the SIA projected about 67,000 U.S. chip jobs at risk of going unfilled by 2030. KORE1 expects compiler and validation engineers to feel it first. https://t.co/tFzlBZpkzh…

Citi: CSPs data center capex is expected to increase to ~$1.6 trillion in 2030E, with global capex spend increasing to ~$3.8 trillion. Google to see the strongest capex growth through 2030. $GOOGL $AMZN $META $MSFT

$NVDA just made some HUGE announcements about the future of AI infrastructure. At the Open Compute Project Summit, NVIDIA announced plans to open up more of its next generation Vera Rubin infrastructure designs while expanding its…

UBS: "As a reminder, for C2027 we anticipate ~50GW of compute shipments (requiring ~57GW of powered data center including PUE factor) against current visibility for ~38GW of powered shell when excluding early-stage power plant projects. In…

"Bernstein has raised its forecast for the global data center XPU ASIC market in 2028 from $140–160 billion to $250–300 billion, with projections for 2026 and 2027 at $50–70 billion and $130–150 billion, respectively." $GOOGL $AVGO $MRVL…

David Lazovsky, EVP & GM, Data Center Networking Business Group at Marvell Technology on AI inference: "3 years ago 1/3 of total data center infrastructure was focused on inference, today north of 60% and over the course of the next 2-3…

Former CEO of Intel, Pat Gelsinger, sees more AI data center delays coming due to lack of energy capacity: "We just need more capacity, and fundamentally, energy capacity omewhat dampens how you fork we can get on AI. Because, you know,…

A humanoid construction worker on a real job site, run from a chair Construction still runs on people. Sites are messy, jobs are hard on the body, and full autonomy isn't there yet. So a UC Irvine team tried something simpler-- park a…

You can now run vision-language models on OAK4. Ask the camera if a pallet is fully wrapped and get an answer, generated entirely on-device. New llama.cpp and ONNX Runtime support means no conversion step for VLMs and other larger models.…

Broadcom is reportedly arranging more than fifty billion dollars in financing so OpenAI can buy chips Broadcom hasn't finished building yet. Apollo and Blackstone are the lenders in the room, the same pair behind Broadcom's $35 billion…

Mecka AI just raised a $60M Series B led by Sequoia. NVIDIA, Microsoft M12, Qualcomm, and Samsung also participated, along with angels including Tony Xu, Frank Slootman, and former Tesla Optimus lead Milan Kovac. Total funding now exceeds…

$2,599 buys a Windows laptop that Microsoft says renders AI images 4.3 times faster than a 16 inch MacBook Pro on M5 Pro. The marketing is aimed squarely at Apple. The architecture underneath is aimed at something bigger: deciding, task by…

AI agents will trigger the largest CPU supercycle in history. Mizuho estimates agentic CPU sockets will surge from 1.3 million in 2026 to 16 million by 2030. Including head node processors, AI could drive nearly half of all server CPU…

a16z's Alex Immerman, David George, and Sarah Wang on why a data center can lower your electricity bill: "It may seem counterintuitive, but a data center can help lower your electricity rate. A recent US study showed that for every 10%…

The US grid plans to add 86 gigawatts this year. Here's where it's coming from: - Solar: 43.4 GW - Batteries: 24.3 GW - Wind: 11.8 GW - Gas: 6.3 GW - New nuclear: 0 GW The AI labs are planning for hundreds of gigawatts. Solar is the one…

Citi on Meta's Muse: At 100M daily users, our framework implies roughly 200,000–390,000 Blackwell-class GPUs, about 2,800–5,500 NVL72 racks, and approximately $7–19B of one-time Nvidia revenue, including scale-out networking. On a per-user…

Excellent and thoughtful piece here. https://t.co/mjHueUXIIX

Left: Texas's grid only maxes out about 200 hours a year, roughly 2% of the time. Right: New data centers that agree to cut back during peak hours can skip the wait for more capacity. US grids could add 100 GW of them without building a…

"At the end of 2024, Texas' grid operator had 63 GW of large new customers in its queue. By this June, that figure reached 474 GW, more than five times record peak demand." "Then Texas hit pause." "Developers routinely spam requests across…

Expert Takeaways: Marvell > Marvell will receive roughly $8 billion plus equity warrants in its Google deal, with the Google LPU v10 projected to generate about $10 billion in revenue over 2029–2030. > LPUs pair with TPUs at a 2-to-1 ratio…

Expert Call: Memory - Key Takeaways > Contracts are moving from flexible 2–3 year terms to 5yr+ take-or-pay structures with upfront deposits and price floors/ceilings, prioritizing downside protection for suppliers. > Tight conventional…

Today was absolutely insane in the stock market Here's everything you need to know: 1. Weak September jobs report continues fueling bets the Fed skips an October hike 2. The Nasdaq hit a record high as megacap tech led 3. $NVDA, $META,…

Can you solve this @Meta robotics challenge? 🧠🤖 What if a robot could understand your intended hand gesture just by reading the muscle signals in your wrist? This demo uses Meta’s Neural Band and sEMG signals to classify hand gestures…

just bought one of the most expensive egocentric data collection device from @Meta time to record some long horizon tasks https://t.co/9KXib5nmP8

AI learned to talk from the internet. How does it learn to handle the real world? That’s the challenge for physical AI. In one test cited by Humanoid Analytics, training on more physical data raised a robot’s success rate in unfamiliar…

This might be one of the best projects ever released in human history. Being able to carry a powerful multimodal intelligence inside a small box already sounds unreal. Muse can understand images, language and instructions, plan tasks and…

Worth reading this conversation about AI demand and memory from StatementDog's analyst, Ken. > While NAND prices are still rising in the second half of 2026, increases are moderating to low-double digits per quarter compared to the…

The pattern I keep seeing this month: once a model layer commoditizes, the fight moves one layer down to whoever serves it fastest. RunAnywhere, a YC W26 company, launched Wally this week, an inference stack for open frontier models that…

My read on Xiaomi shipping MiMo-V2.6-Pro this week: the real story isn't the benchmark score. It's who is now capable of putting up that score. Xiaomi, a phone and appliance company, published MiMo-V2.6-Pro and Flash on Hugging Face…

Everyone is still arguing about parameter counts while the real efficiency gains come from a different lever: what a weight is allowed to be. Moondream just shipped Parakeet Redux, a speech to text model built by compressing NVIDIA's…

Rothschild & Co: Overview of the AI infrastructure stack $NBIS $IREN $ORCL $META $GOOGL $AMZN $GOOG

a few notable open model releases *since* this interview of @dylan522p by the way :) Thinking Machines - Inkling - Jul 15 Moonshot - Kimi K3 - Jul 16, weights Jul 26/27 inclusionAI - Ling 3.0 Flash - Jul 23 DeepSeek - V4-Flash-0731 - Jul…

‘Dogs and Cats Living Together’ in AI. Anthropic, China & Amazon. ARD #167 …Arch-rivals near a deal to stress-test each other’s models. Washington and Beijing stand up an AI hotline. And Amazon slams the door on Meta’s shopping agent. Full…

J.P. Morgan: CSP Capex expected to reach ~$1.7 trillion in 2028 🤣 That is a lot of money... $GOOGL $AMZN $META $MSFT $ORCL $SPCX https://t.co/QpnbRLcFz6

The shift I keep waiting for is local inference leaving the hobbyist corner, and this week I think I watched it happen. What used to mean a terminal window and a quantized model file on your laptop just got Apple's blessing: MLX-Serve, an…

Breaking: OpenAI just posted robotics engineering jobs paying up to $500,000, according to Business Insider 4 stocks worth watching if robotics becomes the meta: • TE Connectivity $TEL *Makes the connectors and wiring that link a robot's…

Apple shipped something small this cycle that I think matters more than another point release of a frontier model: macOS 27 now includes an fm command in the terminal, putting Apple's on-device model one shell command away from every…

🇺🇸 Meta is investing $115M in a free training program to prepare American workers to build AI data centers, with jobs guaranteed for graduates. The American Workforce Academy will offer a 4-week course with industry-recognized…

BofA: We do not expect “de-spec” or lower memory content in new GPUs/ ASICs; instead, we assume a further increase, to 1,000GB per Rubin Ultra or 384GB for the new Rubin, vs 288GB currently. $NVDA $AMD $GOOGL $AMZN $MSFT $META

Nice visuals shared by Blackstone in their recently uploaded Youtube video about AI. > Google token use per month has increased from 10T in May 2024 to 3,200T in May 2026. > Anthropic + OpenAI's ARR at $105 billion as of July 2026, up from…

HSBC: USA model launches per year has been accelerating and still far ahead of China. $GOOGL $MSFT $META $AMZN https://t.co/9oqbKNwxCR

NVIDIA just showed something at IFA that I've been waiting for. Not a faster GPU. A distributed inference router. PAIR (Personal AI Router) connects all the devices on your local network and treats them as a single inference cluster. Idle…

Detect, segment, and track in one API call. Meta's SAM 3.1 is now on the Meta Model API. You send a phrase. It returns detections, pixel-precise segmentation masks, and identity-preserving video tracks across frames. Previous stack: object…

Open model adoption just hit a tipping point. Here's what that means for builders: 1/5 - The Volume Evidence 78.4% of tokens on Vercel's AI gateway run on open models. That's not a survey. That's production traffic. The switch is already…