AI diffusion rules v2.0, with Trumpian characteristics.
www.bloomberg.com/news/article...
AI diffusion rules v2.0, with Trumpian characteristics.
www.bloomberg.com/news/article...
The crowding out of other areas by AI will now start to impact customers. Memory, which accounts for 15-40% of device costs has seen prices skyrocket, as the big three prioritise HBM, used in AI chips, over all else.
www.economist.com/business/202...
The economics of tokens is fascinating and not immediately obvious. In this week's @economist.com I look at the steps between token growth and profits.
www.economist.com/business/202...
Full piece is here: www.economist.com/science-and-...
23.10.2025 07:40 β π 0 π 0 π¬ 0 π 0The last is in software: DeepSeek is pushing for a lower precision, more efficient FP8 format for its domestic chips. It creates more complexity for the software, but if done right, is an efficient workaround lower-power chips.(4/n)
23.10.2025 07:40 β π 1 π 0 π¬ 1 π 0Second is systems: a cluster approach by Huawei, by lashing together almost 5X more chips than the comparable solution by Nvidia at the cost of power, complexity (3/n).
23.10.2025 07:40 β π 0 π 0 π¬ 1 π 0Three major approaches. The first is in manufacturing by pushing ASML's older DUV machines to beyond its limits through multi-patterning to get to 7nm and even 5nm process nodes, at cost of yield (2/n)
23.10.2025 07:40 β π 0 π 0 π¬ 1 π 0How good is China's domestic AI chip capability? Depending on who you ask, the country is either hopelessly behind or already catching up. In the @economist.com I try to decode this. tl;dr: Americaβs dominance in AI chips is secure, for now. Export controls have worked. (1/n)
23.10.2025 07:40 β π 2 π 0 π¬ 1 π 0
Yet this is a third way in which the country can lead in AI. India's path may not be like America's or China's. But it could be no less consequential. (3/n)
www.economist.com/leaders/2025...
As OpenAI, PerplexityAI, Google and more give away their products on the cheap, many are worried about what it means for India's tech sovereignty.
India will not make the fastest chips. It will not invent the latest AI models. (2/n)
The AI race between China and America has hidden an interesting trend: the boom in AI in India.
Indians have embraced AI in a scale not seen in other countries. Culturally too, the country seems more optimistic about its impact than in other countries. (1/n)
www.economist.com/asia/2025/09...
And, some personal reflections as someone who was in the trenches and is now writing about it. The greatest thing about this industry is how globalised, yet locally specialised it is. Resilience is good and needed. But the pendulum has swung too far. Time for some common sense.
22.08.2025 06:40 β π 0 π 0 π¬ 0 π 0
Finally Tom Lee Devlin explains why America needs its allies in its chipmaking ambitions (5/n)
www.economist.com/leaders/2025...
Where does that leave Intel? Not in a very good place. I argue that Intel's best changce of success is to focus on the foundry. A foundry-only business would certainly be a gamble. But the longer it dithers, the lower the chance of success.(4/n)
www.economist.com/business/202...
My colleague Noah Snider writes about Japan's resurgence as a chipmaking hub (3/n)
www.economist.com/asia/2025/08...
First up, I look at TSMC. The company at the heart of tech wars and AI. I argue that among all the challenges the company has faced - earthquakes, typhoons, threat of war - exporting its culture abroad may be the hardest (2/n)
www.economist.com/briefing/202...
There's just a lot happening with chips these days! Some time back it would have seemed wild that the US government could become the single largest shareholder of Intel. But here we are. This week's @economist.com
tries to make sense of it. (1/n)
#semiconductors #tsmc #intel
Here is a podcast on the company on @economist.com www.economist.com/podcasts/202...
Thanks to @birdyword.bsky.social @ethanywu.bsky.social Chris Miller and Jon Y
TSMC is still coming to terms with all the attention and its role as a centre point of tech wars (3/n)
25.07.2025 08:27 β π 0 π 0 π¬ 1 π 0The T in TSMC is a big part of its culture. While a lot of focus of governments in onshoring chip manufacturing has been on subsidies and incentives, there has been less appreciation of how much process, talent and culture matter in leading-edge chip manufacturing. (2/n)
25.07.2025 08:27 β π 0 π 0 π¬ 1 π 0Last week I was in Taiwan as part of a profile on TSMC (coming soon!), a company now worth over $1trn. As an ex-chip designer, it was like a pilgrimage, visiting a country where "CoWoS" is not something only geeks know about. A few thoughts (1/n).
25.07.2025 08:27 β π 1 π 0 π¬ 1 π 0
Between 2017β2024, just 5% of Indian VC funding went to deep tech.
By contrast, Chinese startups poured billions into AI, EVs, and semiconductors in one year alone.
Universities fare little better.
India ranks 3rd in publication volume, but no Indian university ranks in the global top 100 for science.
Academic research rarely turns into marketable innovation.
Indiaβs firms are reluctant to invest.
Even its most profitable companies spend just 0.3% of sales on R&D.
In the US, that figure is 8.8%. In China, 2.1%.
R&D spending tells the story:
π΄ India: 0.7% of GDP
π΄ China: 2.4%
π΄ US: 3.6%
Among the worldβs 2,000 biggest R&D spenders, only 15 are Indian.
In 2024, India attracted just $1.2bn in private AI investment.
The US got $109bn.
Austria and Sweden each got more than India.
Not one Indian AI model ranks among the global top 200.
India is a digital powerhouseβbut not yet a technological one.
It leads the world in app downloads, digital payments, and ChatGPT usage.
But when it comes to building foundational tech, it lags far behind.
A thread based on two pieces in this week's
@economist.com
Interesting stat from JP Morgan. Cost of making an iPhone in India is only 2% more than in China.Most of the value is in individual components that are made by other firms (chips, display) the labour intensive assembly is at par. The harder part is recreating Apple's China supply chain in India.
14.05.2025 11:03 β π 2 π 0 π¬ 0 π 0