AI ASICs: China May Have Plans for AI Software and AI Hardware
December 1, 2025
Another dinobaby original. If there is what passes for art, you bet your bippy, that I used smart software. I am a grandpa but not a Grandma Moses.
I try to avoid wild and crazy generalizations, but I want to step back from the US-centric AI craziness and ask a question, “Why is the solution to anticipated AI growth more data centers?” Data centers seem like a trivial part of the broader AI challenge to some of the venture firms, BAIT (big AI technology) companies, and some online pundits. Building a data center is a cheap building filled with racks of computers, some specialized gizmos, a connection to the local power company, and a handful of network engineers. Bingo. You are good to go.
But what happens if the compute is provided by Application-Specific Integrated Circuits or ASICs? When ASICs became available for crypto currency mining, the individual or small-scale miner was no longer attractive. What happened is that large, industrialized crypto mining farms pushed out the individual miners or mom-and-pop data centers.
The Ghana ASIC roll out appears to have overwhelmed the person taking orders. Demand for cheap AI compute is strong. Is that person in the blue suit from Nvidia? Thanks, MidJourney. Good enough, the mark of excellence today.
Amazon, Google, and probably other BAIT outfits want to design their own AI chips. The problem is similar to moving silos of corn to a processing plant with a couple of pick up trucks. Capacity at chip fabrication facilities is constrained. Big chip ideas today may not be possible on the time scale set by the team designing NFL arena size data centers in Rhode Island- or Mississippi-type locations.
A Chinese startup founded by a former Google engineer claims to have created a new ultra-efficient and relatively low cost AI chip using older manufacturing techniques. Meanwhile, Google itself is now reportedly considering whether to make its own specialized AI chips available to buy. Together, these chips could represent the start of a new processing paradigm which could do for the AI industry what ASICs did for bitcoin mining.
What those ASICs did for crypto mining was shift calculations from individuals to large, centralized data centers. Yep, centralization is definitely better. Big is a positive as well.
The write up adds:
The Chinese startup is Zhonghao Xinying. Its Ghana chip is claimed to offer 1.5 times the performance of Nvidia’s A100 AI GPU while reducing power consumption by 75%. And it does that courtesy of a domestic Chinese chip manufacturing process that the company says is "an order of magnitude lower than that of leading overseas GPU chips." By "an order or magnitude lower," the assumption is that means well behind in technological terms given China’s home-grown chip manufacturing is probably a couple of generations behind the best that TSMC in Taiwan can offer and behind even what the likes of Intel and Samsung can offer, too.
The idea is that if these chips become widely available, they won’t be very good. Probably like the first Chinese BYD electric vehicles. But after some iterative engineering, the Chinese chips are likely to improve. If these improvements coincide with the turn on of the massive data centers the BAIT outfits are building, there might be rethinking required by the Silicon Valley wizards.
Several observations will be offered but these are probably not warranted by anyone other than myself:
- China might subsidize its home grown chips. The Googler is not the only person in the Middle Kingdom trying to find a way around the US approach to smart software. Cheap wins or is disruptive until neutralized in some way.
- New data centers based on the Chinese chips might find customers interested in stepping away from dependence on a technology that most AI companies are using for “me too”, imitative AI services. Competition is good, says Silicon Valley, until it impinges on our business. At that point, touch-to-predict actions come into play.
- Nvidia and other AI-centric companies might find themselves trapped in AI strategies that are comparable to a large US aircraft carrier. These ships are impressive, but it takes time to slow them down, turn them, and steam in a new direction. If Chinese AI ASICs hit the market and improve rapidly, the captains of the US-flagged Transformer vessels will have their hands full and financial officers clamoring for the leaderships’ attention.
Net net: Ponder this question: What is Ghana gonna do?
Stephen E Arnold, December 1, 2025
Deloitte and AI: Impact, Professionalism, and Integrity. Absolutely But Don’t Forget Billable
December 1, 2025
Another dinobaby original. If there is what passes for art, you bet your bippy, that I used smart software. I am a grandpa but not a Grandma Moses.
Do you recognize any of these catchphrases?
- Making an impact that matters
- Professionalism in Practice
- Serving Those Who Serve Business
- Service with Integrity
Time’s up. Each of these was — note the past tense — associated in my mind with Deloitte (originally Deloitte Haskins & Sell before the firm became a general consulting firm. Today’s Deloitte is a representative blue chip consulting outfit. I am not exactly what shade of blue is appropriate. There is true blue, Jack Benny’s blue eyes, and the St. Louis blues. Then there are the blues associated with a “small” misstep with AI. Understatement is useful at blue chip consulting services firms.
Thanks, Venice.ai. Good enough.
I read Fortune Magazine’s story “Deloitte Allegedly Cited AI-Generated Research in a Million-Dollar Report for a Canadian Provincial Government.” The write up states with the alleged hedge:
The Deloitte report contained false citations, pulled from made-up academic papers to draw conclusions for cost-effectiveness analyses, and cited real researchers on papers they hadn’t worked on, the Independent found. It included fictional papers coauthored by researchers who said they had never worked together.
Now I have had some experience with blue chip and light blue chip consulting firms in my half century of professional work. I have watched some interesting methods used to assemble documents for clients. The most memorable was employed by a special consultant dragooned by a former high ranking US government official who served in both the Nixon and Ford administrations. The “special” dude who was smarter than anyone else at my blue chip firm at the time because he told me he was used his university lecture notes as part of our original research. Okay, that worked and was approved by the former government big wheel who was on a contract with our firm.
I do not recall making up data for any project on which I worked. I thought that my boss did engage in science fiction when he dreamed up our group’s revenue goals for each quarter, but the client did not get these fanciful, often juicy numbers.
The write up presents what Deloitte allegedly said:
“Deloitte Canada firmly stands behind the recommendations put forward in our report,” a Deloitte Canada spokesperson told Fortune in a statement. “We are revising the report to make a small number of citation corrections, which do not impact the report findings. AI was not used to write the report; it was selectively used to support a small number of research citations.”
Several random thoughts:
- Deloitte seems to be okay with their professionals’ use of smart software. I wonder if the framing of the problem, the upsides, the downsides of options, and strategic observations were output as a result of AI prompts?
- AI does make errors. Does Deloitte have a process in place to verify the information in documents submitted to a client? If the answer is yes, it is not working. If the answer is no, perhaps Deloitte should consider developing such a system?
- I am not surprised. Based on the blue chippers I have met in the last couple of years, I was stunned that some of these people were hired by big name firms. I assumed their mom or dad had useful connections at high levels which their child could use to score a win.
Net net: Clients will pay for billable hours even though the “efficiencies” of AI may not show up in the statement. I would wager $1.00 that the upside from the “efficiencies” will boost some partners’ bonuses, but that’s just a wild guess. Perhaps the money will flow to needy families?
Stephen E Arnold, December 1, 2025
Mother Nature Does Not Like AI
December 1, 2025
Another dinobaby original. If there is what passes for art, you bet your bippy, that I used smart software. I am a grandpa but not a Grandma Moses.
Nature, the online service and still maybe a printed magazine, published a sour lemonade story. Its title is “Major AI Conference Flooded with Peer Reviews Written Fully by AI.” My reaction was, “Duh! Did you expect originality from AI professionals chasing big bucks?” In my experience, AI innovation appears in the marketing collateral, the cute price trickery for Google Gemini, and the slide decks presented to VCs who don’t want to miss out on the next big thing.
The Nature article states this shocker:
Controversy has erupted after 21% of manuscript reviews for an international AI conference were found to be generated by artificial intelligence.
Once again: Duh!
How about this statement from the write up and its sources?
The conference organizers say they will now use automated tools to assess whether submissions and peer reviews breached policies on using AI in submissions and peer reviews. This is the first time that the conference has faced this issue at scale, says Bharath Hariharan, a computer scientist at Cornell University in Ithaca, New York, and senior program chair for ICLR 2026. “After we go through all this process … that will give us a better notion of trust.”
Yep, trust. That’s a quality I admire.
I want to point out that Nature, a publication interested in sticking to the facts, does a little soft shoe and some fancy dancing in the cited article. For example, there are causal claims about how conferences operate. I did not spot any data, but I am a dinobaby prone to overlook the nuances of modern scientific write ups. Also, the article seems to want a fix now. Yeah, well, that is unlikely. LLMs change so that smart software tuned to find AI generated content are not exactly as reliable as a 2025 Toyota RAV.
Also, I am not sure fixes implemented by human reviewers and abstract readers will do the job. When I had the joyful opportunity to review submissions for a big time technical journal, I did a pretty good job on the first one or two papers tossed at me. But, to be honest, by paper three I was not sure I had the foggiest idea what I was doing. I probably would have approved something written by a French bulldog taking mushrooms for inspiration.
If you are in the journal article writing game or giving talks at conferences, think about AI. Whether you use it or not, you may be accused of taking short cuts. That’s important because professional publishers and conference organizers never take short cuts. They take money.
Stephen E Arnold, December 1, 2025
-
- Subscribe to Beyond Search
Feature archive
News archive
-
