Visions of Freedom

the AI Chip Race : nVidia vs. the World in Silicon Warfare 2024

AI Chip Race

All the news, and all the stock market, seems to celebrate nVidia these days, with their apparent stranglehold on the global AI chip race market. Their domination, and the unprecedented demand for their fastest chips (currently the H100, as of 2024), has been the major factor that has propelled them into position as the most valuable company in the world, surpassing Apple, Microsoft, WalMart, Exxon, Toyota, and everyone else.

As you might imagine, that has given both competitors — and customers — significant motivation to design their own chips, joining the hell-for-leather AI chip race. Why should Microsoft pay nVidia the ransom price of $30 billion a year for AI chips when it is perfectly capable of designing their own silicon? The major caveat is: designing and manufacturing silicon takes a lot of time, and a lot of money. Think: a minimum investment of $10 billion, plus 2 years to design the chip, plus 3 years to actually get the fab online.

So… while nVidia may be the king today, competition is coming… we might just not see fruit from these planted seeds until, say, 2028 at the earliest. Until then, here’s the silicon forecast:

nVidia : THE KING

  • architecture: GPU
    (which they invented)
  • A100 / H100 / Blackwell B20
  • Grace / Hopper — $50k per unit, plus demand pricing
  • technically, nVidia does not own a single chip “fab” (silicon fabrication factory). They merely design, market, and package the chips… TSMC is the actual manufacturer.
  • Question: in the name (under the guise of) quality-testing, is it possible that in its lead position in the AI chip race, that nVidia is actually building its very own next generation of artificial intelligence LLMs/ AGI? since it essentially controls the chip-flow, and theoretically this is a “winner take all” game… why wouldn’t it do this? (for a comparable story in the crypto mining world, witness: Butterfly Labs)

…and then there’s everybody else…

AI Chip Race: the contenders

GOOGLE

  • architecture: TPU
    (which they invented)
  • TensorFlow v6

AMAZON

  • AWS Trainium
  • AWS Inferentia (technically an architecture and an infrastructure, not a chip… but still, an Amazon proprietary AI hardware play)

MICROSOFT

  • Azure Maia 100
  • Cobalt 100

AMD

  • MI300x

OPENAI

TESLA / X.ai

  • hello
  • Dojo
  • dark horse in the AI chip race

APPLE

  • masters of bespoke silicon, therefor well positioned
  • M4 (2024)

META

  • nVidia stockpile
  • intentionally purchasing >10% of nVidia annual output
  • dark horse: custom chip?

TSMC

  • supply chain
  • 90% of the world’s state-of-the-art chips are manufactured in Taiwan
  • assuming AI is a military game changer….
  • …this is a national security threat to America

witness where many of these companies rank within the list of World’s most valuable companies (note: this chart is constantly in flux, tho the top 10, while they shift and shuffle, tend to remain fairly consistent):

…and lest we forget, last but not at all least:

…CHINA

  • (yes, the nation state. this is an American blog, so we tend to view the world through that lens. And in this case, we look at China as a monolithic entity… even tho, actually, it should merit its own breakdown within the AI chip race, as its population is roughly 1/6 of the world’s total, and accounts for 1/5 of the global economy)
  • Chinese hardware companies pursuing the AI chip race holy grail include:
  • Huawei: Ascend 910C (H100 equivalency)
  • Alibaba: Hanguang 800
  • Biren Technology : BR100
  • Cambricon: MLU370
  • SMIC – China’s mainland TSMC equivalent, a foundry for chip design companies

AI Chip Race: Corollaries & Consequences:

As the AI chip race heats up, its activity dovetails nicely into a few other important concepts:

  • AI Sovereignty
  • AI Supremacy
  • Domestic AI Chip Manufacturing Capability
    as a key element of National Security
    • (thus, the CHIPS Act)

Of final note are two up-and-coming technologies, either one or both of which have the potential, in the mid- and long-term, to radically alter the silicon balance of power in AI:

One more thing:

AI datacenters are projected to consume
>8% of total planetary energy generation
by 2030. (Bloomberg)

.

prompt: “shiny chromed computer chip on a glowing green circuitboard with shiny gold circuits. chip has stylized letters ‘AI’ etched on its top surface. glowing traces of light rip across the circuits. narrow depth of field, photographic”

engine: MidJourney 6.1

AcroYogi

, , , ,
Exit mobile version