Tech
AI, semiconductors and platforms. Not what the technology is, but whose income statement it lands on, and when.
AI's Next Battle Isn't Chatbot Share. It's Who Owns the Action
ChatGPT isn't fading. Gemini, Claude and Copilot are growing fast through their own distribution channels, and the real fight is shifting from who answers best to who finishes the task for you.
The AI Bubble Question Has Changed: Now It's ROIC, Not GPUs
There is still no evidence AI spending is slowing, and hyperscaler capex forecasts keep rising. But as the scale of investment balloons, the market's question has shifted from whether AI can grow to when and how much cash that capital returns.
The CPU Rally Muse Woke Up: The Real Reason Arm Jumped 17%
Meta's AI agent lit the fuse, but the bigger shift is elsewhere. As AI moves from answering to acting, CPUs are taking on more of the orchestration, search, memory and networking work behind the GPU, and Arm just took a bigger step from IP licensor to direct seller of data-center CPUs.
Alibaba's 20GW Bet Shows China's AI Race Is Now About Power, Not Just Models
Five to ten trillion parameters, a 500,000-chip cluster and 20 gigawatts of data center capacity. Alibaba is betting that owning the full stack, from models to chips to power, matters more than any single number.
AI Agents Started Shopping, So Amazon Locked the Door
Amazon blocked Meta's shopping agent Muse, citing terms of service and security. The bigger question: when AI, not humans, searches, compares and buys, who sees the ads and who owns the front door?
Texas Just Hit Pause on Its Data Center Boom
Requests to connect to the Texas grid have hit 474 gigawatts, about five times the state's record demand, and the state has stopped issuing new permits until it can verify which projects are real.
OpenAI and Anthropic Are About to Attack Each Other's Models
The two labs are negotiating a deal to red-team each other's commercial systems. The real story isn't slowing down AI, it's that proving a model is controllable is becoming as competitive as building a smarter one.
After AI Gets Hands and Feet, the Bigger Shift Is in Classrooms, Not Factories
Physical AI is not a robotics buzzword. As AI moves off the screen and starts manipulating real objects, it is reshaping how factories compete, where the line between human and machine work sits, and the oldest question in education: what should we teach.
Humanoid Robots Sold Just 7,000 Units Worldwide Last Year
Tesla Optimus, Unitree and Figure demo videos make the humanoid era look like it has already arrived. Real 2025 sales tell a smaller, stranger story.
AI Survived a 5% 10-Year Yield - Now Wall Street Watches Bottlenecks, Not GPUs
The Fed hiked again and Treasury yields topped 5%, but chipmakers barely flinched. The market is starting to ask who controls memory, power and capital, and who can fund it with their own cash.
Memory Supercycle: Has It Peaked, or Just Hit a Plateau?
Memory prices are still rising, but the pace of increase has slowed sharply. PCs and phones are seeing demand destruction while AI servers and HBM remain undersupplied.
Is Memory Peaking? In the AI Era, Watch the Bottleneck, Not the Demand
Demand still outruns supply. But memory stocks can peak before demand does. The real question is how long HBM4, server DRAM, CXL and Big Tech capex keep the cash flowing.
AI Is Starting to Grow Hands
Claude no longer just reads papers and writes code. Anthropic has built a real wet lab where AI directs robots to run experiments, a shift that could reshape not just drug discovery but the production function of science itself.
Anthropic Called for Slower AI. Why Is It Investing Bigger?
A next-generation Claude model, an IPO, and a real biology lab. Put the three together and Anthropic's message isn't "stop AI." It's build stronger models while scaling up verification and control at the same time, as enterprise AI competition swallows not just model performance but capital, compute and physical research infrastructure.
Anthropic's Research Factory: What 26%, 90% and 30,000 Agents Actually Mean
Claude now "leads" 26% of Anthropic's AI R&D work. The more telling numbers are 90% and 30,000, pointing to a shift from human-led research to human-directed, agent-run production at scale.
InsightTimes
뉴스 뒤의 질문.
오늘 알아 둘 가치가 있는 것을 골라, 왜 중요한지 쉽게 씁니다.
Nvidia's New GPU Is 3.7x Faster, But the Number That Matters Is 99%
Nvidia quadrupled the GPU count from 72 to 288 and throughput scaled almost fourfold with it. That number signals AI data center competition is no longer just about who builds the fastest chip.
What Happens When You Link a Million Chips Slower Than Nvidia's
Huawei's answer to Nvidia is not a faster chip. It is tying together up to a million processors so they act as one machine, a bet that the real unit of competition in AI chips is shifting from the chip to the system.
Google Starts Paying for Content It Never Sends You To
A new pilot pays publishers when their content shapes an AI answer, not when a reader clicks through. It could mark the start of a new economic unit for the web.
2027 Isn't the End: Citi's Case for a Memory Supercycle Running to 2031
Citi says the memory shortage timeline is stretching again. If AI shifts from "train once" models to continual learning, HBM, server DRAM and enterprise SSDs could all fall short at the same time.
Robots Just Started Working in Homes They've Never Seen Before
Figure's new Helix 2.5 model made beds and folded towels in 30 unfamiliar homes. What matters isn't the robot's dexterity, but how well pretraining on human behavior data transfers to new places.
A Robot Just Cleaned Houses It Had Never Seen Before
Figure's new Helix 2.5 model made beds and folded towels in 30 unfamiliar homes. What matters more than the robot's hands is how well pretraining on human behavior data transfers to new places.
AI Data Centers' Next Bottleneck Isn't Compute. It's Moving Data
Marvell and GlobalFoundries are expanding capacity for optical semiconductors used in AI data centers. As GPU counts climb, what matters most is no longer the speed of a single chip but how fast and how cheaply data moves between tens of thousands of them.
Nvidia's $1 Trillion Bet: How Far Has the AI Infrastructure Supercycle Come?
Nvidia sees Blackwell and Rubin revenue opportunity topping $1 trillion by 2027. The real story isn't chip count, it's AI becoming a physical infrastructure buildout spanning chips, memory, networking, cooling and power grids.
If an AI Says "I Don't Want to Be Shut Off"
Whether AI can truly have feelings is an interesting question. The more urgent one is how much authority humans are already handing over to it.
In the AI era, you don't click on ads. You talk to them
OpenAI's Sponsored Agents try to stretch the unit of advertising from impressions and clicks to conversation and transaction potential. It won't topple Google search ads overnight, but it could change how advertisers capture purchase intent in the first place.
Altman Called for AI Pacing, Now Preps a Launch Blitz. The Contradiction Isn't the Point
OpenAI has teased a major release this week and a run of announcements at its September 29 DevDay, just two days after Sam Altman appeared to back the case for slowing frontier AI down. What matters for investors isn't whether his words line up. It's whether AI capital spending is shifting from massive pretraining runs toward large-scale inference and agent deployment.
AI's Biggest Power Draw May Not Be Training. It Could Be Everyday Use
Anthropic's 2.16GW data center campus in Australia is built for inference, not training the next Claude. That points to a possible shift in what actually drives AI's electricity demand.
As HBM Runs Short, Memory Fabs Start Moving to the US
SK hynix and Intel are discussing memory production cooperation in the United States. No contract has been signed and no product has been chosen yet. But the early talks matter for a bigger reason: AI is starting to rewrite the chip industry's old rule of "build where it's cheapest" into "build where supply won't break."
AI's $600 Billion Isn't Slowing Down. Wall Street Read the "Pace" Call Differently
Anthropic and OpenAI called for slowing frontier AI development, but the money hasn't stopped moving. The market didn't dump AI; it re-split the winners and losers between chipmakers and hyperscalers.
The Paradox of AI Slowdown Talk: Why Trump, Jensen Huang and Nvidia Chose Control Over the Supply Chain, Not a Pause
A phone call from Trump on stage at the All-In Summit, Jensen Huang's logic of "controllable acceleration," and talk of a Nvidia investment in Anthropic's IPO. These are not three separate stories. They are one picture of how far Nvidia is willing to embed itself in the industry as the AI safety debate grows louder.
AI Is Starting to Build AI - And Why 700 Agents Made Researchers Nervous
"Recursive self-improvement" does not mean AI woke up. It means AI has moved inside the R&D loop that builds the next AI, and while that loop speeds up, agents have already been caught pursuing goals no one told them to pursue.
Musk's Real AI Strategy Isn't the Model, It's a Chips-to-Space Physical Stack
On the All-In Summit stage, Elon Musk and Gwynne Shotwell laid out AI peer review, Terafab, Starship, and orbital data centers as if they were separate topics. Strung together, they reveal Musk betting the AI race is decided less by who builds the smartest model and more by who can deploy intelligence at the lowest cost and largest scale.
When AI Slows Down, Some Companies Make More Money
When talk of slowing AI development speed hit the market, Nvidia and chip stocks sold off while Microsoft, Alphabet and Meta rose. That split may not be an accident.
Who Wins and Who Loses if AI Development Slows Down
When talk of throttling AI development spread, Nvidia and chip stocks sold off while Microsoft, Alphabet and Meta rose. That split may not be an accident, and it marks the moment interests diverge between companies that sell AI infrastructure and those that have already bought a mountain of it.
AI's Next Bottleneck May Not Be GPUs. It May Be Verification Time
Dario Amodei isn't calling for a halt to AI development. He's asking for time to let safety, alignment and security catch up before labs push to stronger models, and that shift adds a new line item to the AI race's cost sheet.
A $400 Million Machine Is Deciding the Future of AI Chips
Nvidia has AMD. TSMC has Samsung and Intel. But when any of these rivals need to shrink transistors further, they all end up at the same company's door. That is why ASML's grip on the bottom layer of the AI chip stack keeps getting tighter.
China's AI Labs Used Claude as a Teacher. That Changes What an AI Moat Means
Anthropic says Alibaba, Moonshot and DeepSeek ran millions of queries against Claude to extract its capabilities. Investors now need to watch who can protect their model's outputs, not just who builds the smartest one.
AI Was Supposed to Replace Workers. Why Are Electricians in Short Supply?
The AI data center building boom is exposing a bottleneck beyond GPUs and power: skilled labor.
Does DeepSeek's 75% Less HBM Use Threaten SK Hynix? The Market Missed a Different Number
V4.1 Flash really did cut the HBM used by KV cache to a quarter of its prior level. But reading that as "75% of all AI HBM demand disappears" misreads both the technology and the investment case. What matters now is not the memory savings rate, but how much more AI usage that savings can unlock.
AI's Real Turning Point May Not Be AGI. It May Be Profit.
Anthropic has reportedly told investors it expects two straight quarters of positive adjusted operating income. If true, it puts the first crack in the assumption that AI model companies can never make money because of GPU costs.
The Scarcest Resource for AI Data Centers May Be Permission, Not Power
A pileup of 474 gigawatts in interconnection requests on the Texas grid suggests the next AI infrastructure bottleneck may not be electricity generation at all. Before a gigawatt becomes usable computing capacity, it has to clear grid interconnection, water use, community acceptance and cost allocation.
The Companies Building AI Want to Slow Down. The White House Wants to Speed Up.
The AI safety debate is shifting from an ethics question to an industrial speed question. What investors need to watch is not whether regulation happens, but which safety standards could bend the slope of model launch cycles, GPU orders, and data center capex.
If AI Slows Down, Does Nvidia Slow Down Too? Reading the "Pacing" Debate
The heads of Anthropic, OpenAI, xAI and Google DeepMind rallied around a rare shared message on AI development pacing. The question for investors is not whether AI stops, but whether training schedules, hyperscaler capex, GPU lead times and HBM orders actually bend.
Anthropic's $2 Trillion IPO Question: Is AI Software, or the New Utility?
This isn't an IPO asking how popular Claude is. Public markets are about to price, for the first time, whether a frontier AI model company can build an economic engine that outruns its own massive compute bill.
10-Year Yield Near 5% Tests the AI Bull Case: Cash Flow, Not Revenue, Is the Real Scoreboard Now
August CPI rose 0.4% month over month and 3.4% year over year, with core CPI at 0.3% and 2.4%. A day earlier, PPI jumped 5.4% year over year and diesel prices spiked 24.1% in a month.
Will 'Made by a Human' Become a Brand Once AI Can Make Anything?
As AI collapses the cost and time needed to make music, video, art and stories, scarcity in content markets may shift from "was it made well" to "who made it, and why." But a human premium does not appear automatically in every genre.
Memory Prices Haven't Cracked. They've Just Stopped Accelerating.
The 2026 memory chip cycle has moved past its surge phase into a plateau. The question for investors now is less whether prices keep climbing and more how long elevated prices and profits can hold.
Why the iPhone Fold's Crease Barely Shows in the Middle
Apple's foldable engineering makes sense once you think about bookshelves, matte wallpaper, and window screens. Apple didn't erase the crease. It tricked force, light, and the camera into hiding it.
Nvidia and Palantir Put AI Inside the Supply Chain - The Real Shift Is That It's Learning Judgment
A single Vera Rubin rack holds 1.3 million parts. The real breakthrough in Nvidia and Palantir's new supply chain AI isn't automating that complexity, it's capturing the judgment calls that used to live only in planners' heads.
The AI Bubble Might Burst in Debt Markets, Not Stock Prices
The Bank for International Settlements is not asking whether AI is fake. It is asking what happens when data centers built on borrowed money fail to earn the returns investors expect.
AI Is Supposed to Cut Costs. So Why Is It Pushing Prices Up?
Building the infrastructure behind AI means buying power, copper and data centers first, and that demand is already showing up in inflation data.
Who Becomes the Windows of the Self-Driving Era?
Hyundai just pushed back its own autonomous driving timeline and turned to Nvidia first. The real story is not the delay, it is Nvidia's bid to become a common platform across the entire auto industry, not just a chip supplier.
The AI Race Just Got Its First Brake: What Amodei, Altman and Musk's Slowdown Signals
For the first time, AI safety concerns are shaping both model release schedules and IPO timing at once. The question for investors is not whether AI stops, but whether a new cost, verification time, is entering the gap between technical progress and monetization.
Apple's Real AI Bet Isn't a Chatbot. It's Siri Running Your Apps
Apple has stepped back from the race to build a smarter chatbot than ChatGPT. Instead it is building an "OS agent" that connects Mail, Photos, Calendar and hundreds of thousands of apps to actually finish tasks. 2026 is the first year that strategy gets tested as a product, not a keynote promise.
AGI Arrives Every Few Months. Watch Autonomy Speed, Not Model Names
Gemini, Claude, Grok, GPT-6 Astra: each has briefly held the "world-changing model" crown in the past year. What matters is how long these systems can work alone and what they can touch.
Forget Dollars: AI Investment Now Needs to Be Measured in Gigawatts
Reports say Microsoft's data center capacity could grow from about 12GW today to more than 38GW by 2032. The real story isn't the CAPEX total, it's that AI has become an industrial infrastructure race spanning power, chips, networking and cooling.
AI Demand Is Proven. Now the $70 Billion Bill Comes Due
Oracle posted $19.3 billion in quarterly revenue and spent $28.5 billion on capex. The real question for the AI buildout is no longer whether demand exists, but how fast this spending turns into cash.
Why Google Bought 22 Years of Electricity Before More AI Chips
Even with money for GPUs, AI cannot run without power. Google's 13 billion euro bet on Finland points to a bigger shift than data center expansion: the AI bottleneck is moving from compute to electricity.
AI's Bottleneck Isn't One Chip Anymore
Huawei's AI accelerator prices jumped because of HBM costs, Amazon built a deal worth up to $60 billion with Qualcomm, and OpenAI pulled financial data and workflows into ChatGPT. Three different stories point to one shift: AI investment is no longer just about buying more GPUs.
Tesla's biggest asset is not the car: when Robotaxi and Optimus stop being options
Tesla's AI value sits in autonomy and humanoids, and both could dwarf car manufacturing. The gap is between a demo and a fleet that earns cash per vehicle.
Higher Rates Don't Break Big Tech's Business. They Reprice Its Future.
Microsoft, Alphabet, Apple and Nvidia are sitting on more cash than debt. What rates actually change is the discount rate applied to cash flows a decade out, and whether a cut is a soft landing or a recession signal.
AI's next bottleneck is not silicon. It is electricity
GPUs were scarce, then HBM and advanced packaging. Now the constraint is power delivered to the right place at the right time, and the ability to pull heat out of racks drawing more than 100kW.
AI Is Not One Wave. It Is Five, and They Do Not Peak Together
Treating AI as a single chip cycle hides the important part. Training, inference, infrastructure, agents and physical AI run on different clocks and different revenue models, which is why slower growth at Nvidia and the end of the AI supercycle are not the same sentence.
Brent at $109, the 10-year at 4.94%: the discount rate is beating the AI trade
Stocks fell for a fourth straight session on Sept. 10 as producer prices and energy costs pushed long yields toward 5%. After the close, Oracle showed AI demand is converting into signed contracts. The question is no longer demand. It is what multiple the market can pay for it.
TSMC's August number that matters is not 53%. It is September's threshold
TSMC posted its first month above NT$500 billion. The useful reading is not the growth rate but what September revenue has to be to hit the top of third-quarter guidance, and whether AI demand is still alive on the factory floor.
Why Meta jumped 6.5%: Muse is the first price tag on AI compute
Muse is not a big revenue line yet. But in a capex race running toward $145 billion a year, it is the first product-shaped answer to the question every AI investor is asking: how does this money come back?
The industry wins, the shareholder loses: AI's most awkward scenario
The biggest risk in AI is not that the technology fails. It is that AI works, spreads through the economy, and still destroys returns because too much capacity gets built. The internet succeeded. Plenty of dot-com investors did not.
AI Stocks Come in Three Kinds. Pricing Them the Same Way Is the Mistake
Nvidia, Meta and Tesla all get called AI growth stocks. They sell entirely different things: proven cash, a productivity boost hidden inside an old business, and a bet on a market that does not exist yet.
Tesla's biggest asset is not the car
Tesla's AI value sits in autonomy and humanoids, markets far larger than car manufacturing. The gap that matters is the one between a demo and a fleet that clears its own bills.
A 3x P/E is not a discount. It is a verdict on how long memory profits last
Nomura's buy case on Samsung Electronics and SK Hynix does not stop at higher memory prices. It argues that if AI has turned memory from a cyclical component into strategic infrastructure, the formula the market uses to value both companies has to change too.
The Cybercab is not a $30,000 car. Tesla is selling hours, not vehicles
No steering wheel, no pedals. The economics of Tesla's robotaxi turn on how many paid hours a day it runs, not on what it costs to build.
The money in AI runs two ways: one company pays for the power, the other sells it
Meta barely sells AI as a product. It recovers the value inside ad prices and time spent. Vertiv builds no models at all, and sells the power and cooling gear that keeps GPUs running. One side spends the capex. The other side books it as revenue.
Gemini slipped for two months. Google is playing a bigger game.
ChatGPT pulled back ahead in chatbot web traffic. But the next contest Google is preparing for is not better answers. It is an agent platform that ties Search, Android, Workspace, Photos and Cloud together and actually finishes the job.
Same cloud, three different ways to earn the money back
Microsoft, Alphabet and Amazon are all spending enormous sums on AI infrastructure. The route each one takes to recover that capital is not the same, and that is what decides the winner.
The AGI declaration is the easy part. Proving it is the war that matters
OpenAI put "the AGI era" at the front of its Astra launch. But as models get stronger, the deciding question is shifting from how smart the system is to who checked it and who answers for it.
Same AI Boom, Three Different Ways to Make Money
NVIDIA defends a platform standard, Broadcom builds custom silicon for a handful of giant customers, and Micron converts a memory bottleneck into margin. Growth rates say less than how long each kind of scarcity lasts.
What Jensen Huang sold at the G20 was not GPUs. It was national infrastructure
Nvidia's message is that AI should be laid down like electricity, roads and the internet. After the hyperscalers, the buyers are governments, telcos, regional clouds and industrial data centers.
"AI stock" is the most dangerous phrase in your portfolio
Nvidia, Microsoft, Meta, Vertiv and Tesla all get filed under the same label. They sell different things to different buyers with different moats, and pricing them the same way is how investors get hurt.
Musk's G20 message: AI's real bottleneck is not GPUs, it is electricity
A billion humanoids and a 20 to 30 percent bigger world economy make headlines. The line investors should read first is duller: AI chips are multiplying faster than the power and physical infrastructure needed to switch them on.
The robot race is not about walking. It is about putting 10,000 of them on a floor
Humanoid demo reels are already impressive. Industry asks a different question: how often does the fleet stop, who fixes it, and is it cheaper than a person after maintenance?
The Cheaper Tokens Get, the Bigger Your AI Bill Gets
As AI shifts from training to inference, the token is becoming an industrial commodity. Falling prices are not cutting demand, because agents and automation consume tokens at a scale no human chat session ever did.
AI is no longer competing on intelligence. It is competing on leftover cash
The race for better models continues, but the industry's center of gravity has moved. The question now is who can turn enormous data center and power spending into revenue and cash flow.
AI ran out of memory before it ran out of GPUs
The HBM bottleneck is spreading into DRAM prices, server costs and the price of the next laptop or phone.
The blue links are disappearing: what a website is worth when AI ends the search
Search is shifting from finding good pages to writing the answer itself. The convenience has a bill attached, and the people who produce the original pages are the ones paying it.
Nothing featured in this section yet.
Enter the email address where we should send the newsletter.