How to invest in AI. These companies form an ecosystem of artificial intelligence and are difficult to ignore

The AI business model is burdened with enormous infections and training, and the winners are probably entities controlling the full technological stack – from chips, through energy, to the cloud. Added to this is a record demand for electricity, which according to American EIA is to beat further records in 2025–2026 thanks to data centers.

We serve Companies in which you can invest and actively contribute to the development of the AI ecosystem.

IMPORTANT: included in the company's text and valuations are only informative and do not constitute a recommendation for the purchase or sale of financial products.

From lithography to memory

Walking from the bottom of the ecosystem, we can distinguish production tools (ASML and its Litography EUV), then Foundry (TSMC), system designers (NVIDIA, AMD), Memory (SK Hynix, Micron), Servers and Networks (Super Micro, Dell, Arista). In addition, infrastructure of data centers and cooling (Equinix, Digital Realty, Vertiv), cloud hyperskalers (Microsoft, Amazon, Alphabet), and finally AI model laboratories (OpenAI, Anthropic, Cohere, Mistral) and end applications.

The ASML solution is necessary for the production of the latest chipswhich makes the company a “bottleneck” for the entire industry. Recently, the company's management reported that “it may not be able to generate the assumed growth”, despite the fact that orders were better than expectations. It is about the risk of customs duties and higher machine costs. This year, ASML shares have generated a profit of 4 percent. Investors counted on more, but Donald Trump's policy clearly influenced this sector. Regardless of what Trump says and does, ASML remains one of the most important elements of the puzzle we call artificial intelligence.

The heart of today's boom is invariably Nvidia. The company still dominates the GPU/accelerator system for machine learning. The Data Center Nvidia department is responsible for the largest revenues and generated as much as $ 39.1 billion in one of the last quarters, which illustrates the scale of demand.

SK Hynix, a key supplier for Nvidia, declares in turn Almost total sales of HBM memory at 2025. The first plan to double the sale of these systems this year – after a record operational profit in Q2 2025. Micron, another memory supplier, reported that his HBM at 2024 was sold out and most of the power at 2025 was already contracted. The company announces further investments in HBM.

HBM memory (High Bandwidth Memory) is specialized DRAM systems connected to TSV through throughout, with a very wide interface ensuring a huge bandwidth with lower energy consumption. They are “fuel” for modern AI, because they remove the bottleneck of memory that arose when the GPU computing power grew faster than the bandwidth of traditional RAM. NVIDIA shares this year generated a profit of 30 percent, SK Hynix up to 54 percent, and Micron is in the plus of 32 percent.

Read also: Aleksander Kwasniewski, asked about the swearing -in of Karol Nawrocki. “We don't live in paradise”

Foundry and production equipment, i.e. who physically produces these chips

TSMC increases the share of the HPC segment (including AI) in its revenues: from 43 percent in 2023 to 51 percent in 2024, AW and KW 2025 already 59 percent It shows how much AI drives the demand for the latest lithographic processes. ASML provides EUV/High-on EUV without which this process would be impossible, but Taiwan TSMC is equally important. Her shares increased this year by 20 percent.

Super Micro Computer (shares increased since January by 72 percent) also It became a symbol of AI servers boombut the variability of results (reduced forecasts, reporting problems) shows the risk of this segment. Dell, in turn, reported $ 12.1 billion. new orders for AI servers in one quarter and $ 14.4 billion. in Backlog. Arista Networks exceeded $ 2 billion. revenues in the first quarter of 2025, indicating the transformation of the network under AI. Della shares have increased by 11 percent since January, and Arista – by 3 percent

Going further, you need to consider data centers, energy and cooling. Equinix and Digital Realty increase the pace of investment, planning billions of dollars a year and thousands of new server wardrobes. Vertiv, supply and cooling supplier, indicates a forecast growth of 160 percent. power demand due to AI. American energy is preparing for record electricity consumption. If we look at the shares of these companies, Equinix has recorded a loss of 16 % since January, Digital Realty achieved a small profit of 1.5 percent, but Vertiv generated a result of 15 percent. in the plus.

See also: Giant deposits in the Suwałki region. They can be worth up to PLN 8 trillion

Hyperstanians care about the stock exchange

Google (Alphabet) raised the Capex plan (budget schedule of investment outlays) on 2025 to $ 85 billion, emphasizing the AI infrastructure. Amazon and Microsoft go the same way, while protecting energy and developing their own systems. These are companies that They combine infrastructure, models and distribution, so they have a natural advantage of scale and integration. All these companies are technological giants, generating huge costs.

Considering large investment outlays, Big Techy unfortunately reduce costs, which in practice means a number of exemptions. All these companies are in good condition, but they want to take care not only of innovation, but also shareholders – as a result they focus on large cost reductions and give up many employees. Alphabet shares this year generated a profit of only 1.45 percent, Amazon provided a profit of 5 %, and the Microsoft exchange rate jumped by 21 percent.

It is worth noting that companies such as Anthropic (Claude) or OpenAI (ChatgPT) are not yet on the stock exchange. You cannot invest directly in these entities, but, for example, Microsoft has invested billions of dollars in Opeli and has significant shares in the organization. Anthropic received $ 4 billion. from Amazon (tranche completed in 2024) and up to $ 2 billion. From Google-the start-up was valued at around $ 60 billion. At the beginning of 2025, Cohere, another start-ups for developing AI, collected $ 450 million. from Nvidia and Salesforce and over $ 900 million together. Microsoft also located EUR 15 million in Mistral AI And he concluded a long -term distribution partnership at Azure. A public investor can therefore obtain an intermediate exhibition through Microsoft (OpenAI), Amazon and Alphabet (Anthropic), Nvidia or Salesforce (Cohere), or SAP (Anthropic/Cohere/Aleph Alpha).

Finally, you also have to remember about ETFs. Those focused on AI differ in methodology and may not hit the “winners”, as shown by the Wall Street Journal analysis. Some of the funds weighed Nvidia too poorly and lost to the market despite the accurate thesis. When choosing ETF, you need to check the exposure to the entire value chain and the weighing method of individual companies, not just the proverbial “AI” in the name. Popular ETFs focused largely on the AI to the Global X Robotics & Artificial Intelligence ETF sector (7 % profit this year), Roundhill Generative AI & Technology ETF (profit 24 %) and Wisdomtree Artificial Intelligence ETF (only 0.65 %).

Read also: The Supreme Court judge scares new ministers. We explain what the president can do

How to build your own investment strategy in AI

First, determine whether you want to have a wide exposure (ETF or basket of large infrastructure companies and providers) or selective (several companies from different stages of the chain: chips, memory, foundry, servers, network, DC/energy, cloud, models).

Then set the temporary horizon and volatility tolerance-the hardware segment is cyclical, and the valuations can be corrected rapidly with the slightest slowdown of CAPEX hyperskalers. Also, take care of geographical and currency diversification, as well as regulatory risks (e.g. chip export control, energy policy). Systematic buying of shares can reduce the risk of bad timing. This stage is your portfolio policy – it is only important that it is consistent and consistent.

Who has the greatest potential to remain a strong player? Looking coldly at the barriers of the entry and advantage of the scale, The most difficult to “replace” companies controlling key bottlenecks: NVIDIA in accelerators, TSMC and ASML in production, SK Hynix/Micron in HBM, and on Full Stack – Microsoft, Amazon and Alphabetthat combine models, cloud, own chips or partnerships and access to energy.

Equinix, Digital Realty and Vertiv, and networks – Arista stand out on the physical infrastructure side. This does not exclude the spectacular increases of smaller players (e.g. Super Micro), but the risk goes hand in hand with the prize. It is worth remembering that the advantage will be on the side of those who will master equipment, energy and models, not just a software interface.

Author: Grzegorz Kubera, Business Insider Polska journalist

Note: The valuations included in the text are only informative and do not constitute a recommendation for the purchase or sale of financial products.