![\"Google](\"https:\/\/www.newsbeep.com\/nz\/wp-content\/uploads\/2025\/12\/Algorithm_-_Google_Ironwood_TPU_1764707787306_1764707816762.png\" "\\"Google")
Google which has till now largely rented out custom TPUs is offering to sell them for the first time Last time on Neural Dispatch: The AI maths is broken, and a bubble that many pretend to not see, is deflating<\/a>ALGORITHM<\/p>\nThis week, we talk about stressful times for Nvidia, also because Google\u2019s found a big leap for its TPU business that directly hurts Nvidia\u2019s GPU sales, and its a case of never say never as OpenAI admits that advertisements within ChatGPT may more be a case of when and not if.<\/p>\n
Google TPU vs Nvidia GPU: Who\u2019s winning!<\/p>\n
\u201cCompetitors can price their chips at $0 and Nvidia products will still be a better option,\u201d CEO Jensen Huang said, not too recently. That didn\u2019t age well at all, did it? Meta is now buying (not immediately, staggered) what will be billions of dollars worth of Google TPUs, or Tensor Processing Units. Two important things to note here. First, Google which has till now largely rented out custom TPUs is offering to sell them for the first time. Secondly, despite Nvidia\u2019s insistence to the contrary, TPUs as application-specific integrated circuit (ASIC) will be a better and more cost effective fit in specific AI frameworks. This news hit the Nvidia stock price, quite hard. So much so, Nvidia posted a rather grudging congratulatory note (after all, everyone\u2019s got to still show they\u2019re friends, because AI\u2019s sake\u2026) whilst claiming that \u201cNvidia is a generation ahead of the industry \u2014 its the only platform that runs every AI model and does it everywhere computing is done.\u201d Great.<\/p>\n
What\u2019s the difference between a GPU that Nvidia wants to sell for AI tasks, versus a TPU that Google champions? A GPU is a specialised processor originally designed for manipulating computer graphics. Their parallel structure makes them ideal for algorithms that process large blocks of data commonly found in AI workloads. A TPU is an ASIC designed for neural networks with specialised features, such as the matrix multiply unit (MXU) and proprietary interconnect topology that make them ideal for accelerating AI training and inference. All of Google AI services run off TPUs. Right now, Google has three iterations available for customers \u2014 the Cloud TPU v5e that\u2019s ideal for medium-to-large-scale training and inference workloads, the Cloud TPU v5p which is a powerful TPU for building large, complex foundational models, as well as the Trillium, a sixth generation TPU which pushed the standards for energy efficiency and peak compute performance per chip for training and inference. Next in line is the Ironwood TPU, which Google says is their \u201cmost powerful and efficient TPU yet, for large scale training and inference.\u201d<\/p>\n
One thing is clear. Suddenly, the \u2018Nvidia or nothing\u2019 worldview seems less absolute. Is Nvidia still the gold standard? Perhaps, but we\u2019ll let the market forces decide that. But the armour isn\u2019t shining anymore.<\/p>\n
Nvidia says they aren\u2019t like Enron<\/p>\n
I mean I don\u2019t know everything, but I\u2019d say that if you have to say \u201cwe are not like Enron,\u201d you\u2019re already on the back foot. Isn\u2019t it? Nvidia\u2019s attempts to silence ballooning speculation, over revenue recognition, accounting methodology, and the realistic lifespan of GPUs, are sounding increasingly like a company that is on the back foot. In a note to Wall Street analysts in response to the definitive puncturing of the AI bubble over the past few months (panic is setting in, because circular funding gas been found out) and also American investor and hedge fund manager Michael Burry made a rather blunt post X after Nvidia\u2019s earnings release<\/a>. He had pointed out the practice of Nvidia\u2019s customers now using six-year depreciation schedules instead of the standard practice of two or three years. Burry had argued about the maths of that practice. Nevertheless, Nvidia in the note says \u201cNvidia does not resemble historical accounting frauds because Nvidia\u2019s underlying business is economically sound, our reporting is complete and transparent, and we care about our reputation for integrity. Unlike Enron, Nvidia does not use Special Purpose Entities to hide debt and inflate revenue.\u201d Great, then. The fact that investor forums, analysts, and a few very loud short-sellers even managed to find shades enough for that comparison, is telling. Every over-performing tech giant eventually learns \u2014 gravity isn\u2019t a rumour; it\u2019s a schedule.<\/p>\n From \u201cnever doing ads\u201d to, this\u2026<\/p>\n It was inevitable, wasn\u2019t it? The world\u2019s most powerful conversational product seems to be tip-toeing toward the world\u2019s oldest monetisation model \u2014 advertising. Some things never change, no matter how technology changes. OpenAI\u2019s Sam Altman has hinted that ads are \u201csomething we\u2019ll try\u201d. Don\u2019t get sidetracked by this, because in OpenAI-speak, this translates to \u201ca pilot is already running somewhere, we just haven\u2019t told you.\u201d I\u2019d estimate this means three things. First, AI search is about to look even more like traditional search. Secondly, \u201cobjective answers\u201d may slowly become \u201csponsored objectivity. And third, any dreams of an ad-free, neutral, reasoning-first AI assistant may be headed for a retirement home. The silver lining? If done with transparency, ads might subsidise cheaper tiers, in particular the India-focused education initiatives, without degrading the totality of the product. But the internet has heard that promise before. And some of us will only believe what we see.<\/p>\n
![\"\"\/](\"https:\/\/www.newsbeep.com\/nz\/wp-content\/uploads\/2025\/12\/quoteText.png\")
<\/p>\n