Hey! We are Nemo.
Nemo, short for Never Miss Out, is a mobile investment platform that delivers curated, data-driven investment ideas to your fingertips. It offers commission-free trading across stocks, ETFs, crypto, and CFDs, along with AI-powered tools, real-time market alerts, and themed stock collections called Nemes.
Download the App
Scan the QR code to download the Nemo app and start investing on Nemo today
• 10 min read
• Published on 7 August 2026
published-on
Read article
published-on
Read article
published-on
Read article
published-on
Read article
published-on
Read article
The Energy Drain. General-purpose chips are massive energy hogs. The recent AMD Taalas AI inference chip acquisition flips the script completely. They're betting on hardwired AI silicon that works like a scalpel. It strips out the wasted power and does one specific job brutally well.
The Inference Pivot. Training AI is yesterday's news. The real battleground is deployment. Smart capital is hunting for the ultimate AI inference semiconductor, because tech giants now demand cheaper, faster ways to run models every single second. Execution is everything. Period.
The Challenger Play. The AMD vs Nvidia rivalry is heating up for anyone evaluating AI chip stocks 2026. It's easier than ever to build diversification with small amounts today. You can grab fractional shares on a regulated broker, enjoying commission-free trading exposure to both sides of this massive turf war.
The Expiry Date. AI models evolve incredibly fast. A custom chip built for today might become totally useless tomorrow. Plus, big tech could just manufacture their own hardware, meaning this whole AMD AI chip deal might face a brutally shrinking market. You'll want to lean on real-time insights and AI-driven research to track these rapid shifts, as unexpected changes could easily impact your portfolio returns.
View the full Basket:Enterprise AI Hardware Supercycle | Theme Overview
To my mind, the semiconductor industry is starting to look a lot like a very expensive, highly sophisticated game of musical chairs. For the last couple of years, everyone has been scrambling for the exact same seat. That seat, of course, belongs to Nvidia. Investors have thrown billions at anyone promising to build a bigger, faster, more flexible graphics processing unit. It has been a straightforward, albeit crowded, strategy. But now, AMD has decided to stop playing the same game.
With its recent acquisition of Taalas, AMD has made a calculated pivot. Rather than trying to beat Nvidia at its own general-purpose game, AMD is placing a wager on a completely different philosophy. They are betting on silicon that is hardwired to run a single, specific AI model with brutal efficiency.
It is a fascinating move. It is also fraught with peril.
If you want to survive the next phase of tech investing, you need to stop worshipping at the altar of raw computing power.
Efficiency, not sheer brute force, is where the next battle will be fought. Let me explain why this matters, and why the market for running AI models could become far more fragmented, and far more risky, than the current consensus suggests.
View the full Basket:Enterprise AI Hardware Supercycle | Theme Overview
To understand what Taalas actually does, you have to look at how chips are built. Nvidia’s dominant chips are essentially the world’s most powerful Swiss Army knives. They are general-purpose accelerators. You can throw almost any computational workload at them, and they will figure it out. That flexibility is brilliant when you are experimenting, but it comes at a steep price. General-purpose chips carry an enormous amount of programmable overhead. They waste energy, they generate heat, and they cost an absolute fortune.
Taalas, by contrast, designs chips that do exactly one thing. They build silicon hardwired for a single AI model.
Think of it as a scalpel. A scalpel is useless if you need to open a bottle of wine or screw in a loose floorboard. But if you need to perform surgery, nothing else will do. By stripping away all the flexible, programmable architecture, a model-specific chip can operate with breathtaking efficiency. It lowers the power consumption drastically. It slashes the cost per operation. For the massive tech companies running billions of AI queries every single day, those incremental energy savings compound into massive financial victories.
In 2022, the market was absolutely obsessed with training AI. That was the era of brute force. Today, the landscape is shifting. The models have been built. Now, the challenge is running them cheaply enough to actually turn a profit.
View the full Basket:Enterprise AI Hardware Supercycle | Theme Overview
We need to draw a sharp line between training an AI and running an AI.
Training is the equivalent of sending an AI to university. You feed it mountains of data over a period of months, forcing it to learn patterns. It requires the massive, parallel processing power that Nvidia provides so beautifully. However, training happens relatively infrequently.
Inference is the day job. It is what happens every single time you ask a chatbot a question, or ask an image generator to draw a cat in a spacesuit. Inference is continuous. It happens at an unfathomable scale. And crucially, it rewards low latency and extreme energy efficiency over raw, flexible power.
This is exactly why the market for purpose-built inference chips is opening up. If you want to grasp the sheer scale of this transition, it is worth looking at the Enterprise AI Hardware Supercycle | Theme Overview to understand how the underlying capital flows are shifting. The money is moving from the laboratory to the factory floor. AMD knows this, and the Taalas acquisition is a direct attempt to capture that deployment phase.
View the full Basket:Enterprise AI Hardware Supercycle | Theme Overview
Do not make the mistake of thinking Nvidia is simply sitting still while AMD tries to outflank them. Nvidia remains the undisputed behemoth of this sector.
Their hardware is exceptional, but their real weapon is software. The CUDA software ecosystem has created a lock-in effect that is almost entirely unprecedented in modern technology. Developers learn it, they rely on it, and they are incredibly reluctant to abandon it. Custom chips might match Nvidia on a specific mathematical task, but replicating that software ecosystem is a multi-year slog. Nvidia is also pushing heavily into custom designs themselves, working directly with large clients to maintain their grip on the market.
Then we have Intel.
Intel’s position is, frankly, complicated. Their Gaudi AI accelerators look highly competitive on a spreadsheet, yet they have struggled to make a meaningful dent in Nvidia’s market share. But Intel might have a different route to victory here. If the future of inference relies on thousands of different custom, hardwired chips, someone still has to actually manufacture them. Intel’s foundry business, which builds chips for other designers, could theoretically thrive even if their own branded architectures fail to catch on. If the market fragments, the factories could win.
View the full Basket:Enterprise AI Hardware Supercycle | Theme Overview
This brings us to the most uncomfortable part of the AMD thesis. Investing in deep tech is never a safe bet, and this particular strategy carries a rather terrifying structural risk.
AI models are evolving at a breakneck pace.
When you build a general-purpose chip, you can update the software when a new AI model is invented. When you build a hardwired chip, you cannot. A chip designed specifically for today’s leading language model could become commercially obsolete within eighteen months if the underlying mathematics change. You are effectively carving a specific algorithm into stone.
If the leading model architectures shift dramatically, these highly specialised chips become nothing more than very expensive paperweights. The pace of change in this industry has shown no signs of slowing down, and anyone looking at AMD’s strategy must accept that model-specific silicon has a brutally short shelf life. It is a constant treadmill of redesigning, testing, and manufacturing.
Furthermore, integrating a startup like Taalas into a massive corporate machine like AMD is notoriously difficult. The tech graveyard is full of brilliant laboratory ideas that failed to scale in the real world. Timelines slip, costs spiral, and the initial optimism often fades into quiet write-downs.
View the full Basket:Enterprise AI Hardware Supercycle | Theme Overview
Perhaps the most potent threat to both AMD and Nvidia does not come from each other. It comes from their own customers.
The biggest buyers of AI chips in the world are the hyperscalers. Google, Amazon, and Microsoft have the capital, the talent, and the motivation to build their own hardware. Google has been developing its Tensor Processing Units for years. Amazon is pushing its Trainium chips. These companies do not want to pay hefty margins to AMD or Nvidia if they can avoid it.
If the hyperscalers successfully transition to using their own proprietary silicon for inference workloads, the addressable market for third-party chipmakers shrinks dramatically. This is a persistent, overarching risk that could redefine the entire semiconductor landscape over the next five years.
View the full Basket:Enterprise AI Hardware Supercycle | Theme Overview
For retail investors, the AI hardware narrative is becoming far more nuanced. You can no longer just buy a basket of chip stocks and expect guaranteed, astronomical returns. The easy money, if it ever existed, has likely already been made.
AMD’s acquisition of Taalas is a bold, fascinating attempt to carve out a profitable niche in the inference market. It might provide them with a crucial structural edge, or it might become a cautionary tale about the dangers of hardwiring volatile technology.
If you are evaluating these companies, I suggest keeping a very close eye on AMD’s upcoming revenue disclosures. Look for concrete evidence that hyperscalers are actually ordering Taalas-derived technology. Until those orders materialise, this remains an expensive research project with a highly uncertain payoff.
The inference race is just beginning. It will be volatile, it will be messy, and it will punish those who assume the future will look exactly like the past. Keep your expectations grounded, acknowledge the very real possibility of capital loss, and pay attention to the companies building the tools for tomorrow, rather than resting on the triumphs of yesterday.
View the full Basket:Enterprise AI Hardware Supercycle | Theme Overview
View the full Basket:Enterprise AI Hardware Supercycle | Theme Overview
View the full Basket:Enterprise AI Hardware Supercycle | Theme Overview
View the full Basket:Enterprise AI Hardware Supercycle | Theme Overview
View the full Basket:Enterprise AI Hardware Supercycle | Theme Overview
View the full Basket:Enterprise AI Hardware Supercycle | Theme Overview
This article is marketing material and should not be construed as investment advice. No information set out in this article be considered, as advice, recommendation, offer, or a solicitation, to buy or sell any financial product, nor is it financial, investment, or trading advice. Any references to specific financial product or investment strategy are for illustrative / educational purposes only and subject to change without notice. It is the investor’s responsibility to evaluate any prospective investment, assess their own financial situation, and seek independent professional advice. Past performance is not indicative of future results. Please refer to our Risk Disclosure.
published-on
Read article
published-on
Read article
published-on
Read article
published-on
Read article
published-on
Read article