The AI revolution, in plain English

Join 10,000 founders and VCs →
Joined January 2024
CXMT's share of global DRAM revenue jumped to about 10% in the second quarter, up from 4% a year earlier. A new report tells Samsung and SK Hynix how to defend their lead without saying how long it will hold. vector.news/articles/chinese…
1
1
1
155
Over the past year, SK Hynix's share of global DRAM revenue fell 14 points while Samsung and CXMT both gained ground, redrawing the market's balance of power.
39
ByteDance's Seedance 2.0 Pro led a University of Hong Kong benchmark of 12 AI video models, scoring 3.84 on quality against Google's Veo at 3.47. The same study ranked it sixth out of eleven for safeguards. vector.news/articles/chinese…
76
Meta has launched $35 billion in stock offerings, the first piece of a $95 billion plan to fund its AI build-out. The other $60 billion rests on a class of shares that doesn't exist yet. vector.news/articles/meta-ou…
70
Microsoft has offered buyouts to 60,000 workers, a quarter of its staff, after capital spending hit $115.9 billion last year. Data-center, sales and marketing teams were left off the list, and Microsoft won't say what the program will cost. vector.news/articles/microso…
98
Researchers at a security start-up spent under 12 hours running AI agents against Anthropic's own systems. They got into sales, HR and IT tools through third-party software, not the model itself. vector.news/articles/researc…
104
Meta spent $34.9 billion on infrastructure last quarter, more than its operations brought in. Free cash flow swung $12.7 billion in a year, from positive to negative. vector.news/articles/meta-be…
98
Meta spent $34.9bn on capital projects last quarter, $2.1bn more than its operations brought in. The company's plan to close that gap hinges on a new class of stock that carries no vote. vector.news/articles/meta-s-…
1
97
Meta's capital spending outpaced its operating cash flow in the third quarter, a reversal from a year earlier when cash flow comfortably covered it.
37
Entergy Arkansas says a new Google deal will save customers $500 million over 20 years. Regulators' own staff put a rival accounting method's savings at $519 million, and the two sides won't agree on whose number counts. vector.news/articles/google-…
103
The Marine Corps has licensed Palantir's Maven battle software for unlimited use across every command, effective October 1. Price, duration and how many Marines will actually use it remain undisclosed. vector.news/articles/marines…
96
The Pentagon has turned Trump's anti-woke AI order into three contracting templates, dated October 20. Whether the rules reach national-security systems, which federal guidance exempts, remains unclear. vector.news/articles/pentago…
121
Google is funding Voltus to build a gigawatt virtual power plant from home batteries and thermostats, live in 2027. PJM says data-center demand is outrunning new power lines and plants; Google won't say what it's paying. vector.news/articles/google-…
1
1
163
Elon Musk says SpaceX closed $40 billion in debt to buy AI chips and predicts a wave of failures among more than 200 neocloud firms. Days later, two of those firms borrowed billions more. vector.news/articles/musk-wa…
122
Apollo's chief economist says AI has delivered no revenue or margin gains for 493 of the S&P 500's biggest companies. His own data cannot show how many of those firms are making money from it. vector.news/articles/apollo-…
93
Alphabet's free cash flow turned negative, a $5.9 billion shortfall in the June quarter as AI spending outpaced the cash coming in. Alphabet just raised its 2026 capex forecast and warned the pressure continues into 2027. vector.news/articles/alphabe…
1
117
The chart tracks operating cash and capital spending quarter by quarter, showing capex overtaking cash generation and tipping free cash flow negative.
54
DeepSeek ran a matrix test on Huawei's Ascend 950DT chip at 431 trillion calculations a second, 99.8% of the chip's theoretical limit. It says nothing about whether the software sells more Huawei chips. vector.news/articles/deepsee…
1
162
On DeepSeek's own FlashMLA attention benchmarks, Ascend reaches most of its own ceiling but reads prompts and generates answers at roughly a third of Nvidia's B200 throughput.
41