By Hagen Ernst, Deputy Head of Research & Portfolio Management at DJE Kapital AG
Artificial intelligence remains the defining theme in the technology sector. One thing is already clear: AI will have a profound, epoch-making impact and will reshape the sector. As was already the case last year, the pace of this development appears to have been underestimated again in 2026. The major hyperscalers — Amazon, Microsoft, Alphabet and Meta — have revised their 2026 investment plans significantly higher. At the same time, AI monetisation is increasingly gaining momentum.
Hyperscalers raise their investment plans again
At the beginning of the year, analysts were still assuming a consolidated investment budget of around USD 600 billion — already a record level. However, following first-quarter 2026 results, it became clear that these estimates had been too conservative.
Meta and Alphabet accelerate investment
Meta now expects capital expenditure of USD 125–145 billion, compared with its previous guidance range of USD 115–135 billion. This represents an increase of 85% year on year. The key drivers are the newly established Meta Superintelligence Labs as well as the expansion of data centres and GPU capacity. However, investors remain critical of the high level of investment in “superintelligence”, as it is unlikely to generate meaningful incremental returns in the foreseeable future.
At the same time, user engagement on Instagram and Facebook continues to rise. Meta can use AI to deliver more targeted advertising and thereby accelerate monetisation. This was also reflected in first-quarter revenue growth.
Alphabet, Google’s parent company, also announced substantially higher investment. Its forecast increased to USD 180–190 billion, from a previous range of USD 175–185 billion. This corresponds to year-on-year growth of 100%. Google is focusing on proprietary AI chips, known as TPUs, cloud infrastructure and model development under DeepMind. The cloud business was particularly impressive, with revenue growth of 63%, driven by the increasing adoption of Gemini.
Amazon and Microsoft focus on their platforms
Amazon is maintaining its investment forecast of around USD 200 billion, which still makes it the leader among the hyperscalers in absolute investment terms. Its AWS division is investing heavily in new data centre regions as well as proprietary AI chips from the Trainium and Inferentia families. In addition, the company recently invested USD 25 billion in Anthropic and USD 50 billion in OpenAI. In return, Anthropic awarded AWS contracts worth USD 100 billion. OpenAI also announced a strategic partnership. With Claude and Frontier, two powerful enterprise AI programmes are now running on AWS.
Microsoft confirmed its spending plans, with an increase of around 25% year on year. Its partnership with OpenAI remains the strategic core of its AI ambitions. Microsoft is integrating AI functionality across its product portfolio — from Azure and Office 365 to GitHub Copilot. However, its cautious approach led to the loss of its exclusive partnership with OpenAI. In addition, Copilot has so far failed to meet elevated expectations. Microsoft has fallen behind in the AI race. At the same time, its Office-related software business faces risks from AI-driven disruption. That said, the cloud business continues to grow strongly.
The next order of magnitude is already visible
Taken together, the leading hyperscalers are heading towards combined investment of USD 725 billion to more than USD 800 billion. This would mark the third consecutive record level and represent growth of more than 60% versus the already historically high levels seen in 2025.
Demand for AI compute capacity is so strong that another significant increase in investment is expected in 2027. However, another increase on the same scale appears unlikely. Around 75% of this investment is flowing directly into AI infrastructure, including GPUs, networking technology, data centres and cooling systems.
AI monetisation is gaining momentum
For a long time, the central criticism was: investment is rising sharply, but who is actually making money from it? In 2026, this question can increasingly be answered in concrete terms.
OpenAI reported annual recurring revenue of USD 20 billion at the end of 2025, compared with USD 6 billion in 2024 and USD 2 billion in 2023.
In April 2026, its annualised revenue stood at around USD 24 billion. OpenAI monetises through ChatGPT subscriptions, API access, enterprise licences and, increasingly, agent-based solutions.
Anthropic is growing even faster
Anthropic’s growth story is even more impressive. The company began 2025 with annual recurring revenue of around USD 1 billion, ended the year at USD 9 billion and reported annual recurring revenue of USD 30 billion in April 2026. By May, this figure had already risen to USD 44 billion. Anthropic has therefore overtaken OpenAI in terms of annualised revenue. The key driver is Claude Code, an AI development assistant. More than 1,000 enterprise customers now pay over USD 1 million per year.
Hyperscalers are monetising AI primarily through their cloud platforms. Microsoft Azure AI is recording double-digit growth, while GitHub Copilot has more than 1.8 million paying enterprise users. Google Cloud is seeing a significant AI revenue contribution for the first time, driven by the integration of Gemini and AI services delivered via the cloud. AWS is also experiencing increased demand for cloud services, particularly its AI developer platform Bedrock.
Winners and losers in the technology sector
The hyperscalers’ massive AI investments are creating clear winners and losers. Companies directly involved in building AI compute capacity have the greatest structural advantage. The largest share of investment, around 35–40%, is being allocated to AI chips. Nvidia remains the main beneficiary, but alternative suppliers are also gaining market share. Companies developing AI-specific networking chips and custom ASICs for hyperscalers are also well positioned.
Memory chips become the bottleneck
The memory chip segment has seen the most significant re-rating. In the past, the memory chip business was generally regarded as highly cyclical. Today, memory chips represent the biggest bottleneck in the value chain. Strong demand for fast, volatile AI DRAM — high-bandwidth memory, or HBM — has led to capacity shortages and price increases of 80–90% compared with the fourth quarter of 2025.
Several companies have benefited from this, including SK Hynix, which has close relationships with Nvidia. Share price performance has reflected this accordingly. At the same time, earnings expectations in this segment are rising.
Networks are in demand, software comes under pressure
The enormous data throughput in AI data centres requires substantial network bandwidth. Providers of networking and optical transmission technology are benefiting from the expansion of network infrastructure. In addition, traditional corporate networks need to be upgraded in order to handle the increase in data traffic caused by AI agents. This creates opportunities for established enterprise networking providers, which appear moderately valued from a valuation perspective.
Traditional software providers and IT services companies are currently among the biggest losers from the latest developments. The focus is on AI disruption risk. The increasingly sophisticated enterprise applications from Anthropic and OpenAI are fuelling concerns that traditional providers could lose relevance. However, it remains questionable whether AI agents will be able to independently manage complex business processes. Software providers with complex solutions are unlikely to be easily replaced. At the same time, some companies in this segment are themselves active in the AI cloud business and are pursuing ambitious growth targets.
What matters now
At present, there is much to suggest that the speed of AI development continues to be underestimated. Demand for AI compute capacity and the monetisation of AI applications are the key factors. As long as available capacity is insufficient to meet demand, this trend is likely to continue. The most significant bottlenecks currently lie in memory chips and power capacity.
New technologies could change the picture
In March, Google introduced TurboQuant, a vector compression algorithm. It compresses the key-value cache — effectively the “working memory” of a large language model — during inference. This is intended to reduce memory requirements by a factor of six. Such a breakthrough would be negative for memory chip producers.
At the same time, high-performance AI chips are becoming increasingly energy-efficient. Nvidia’s latest Rubin chip generation, for example, delivers ten times the performance per watt of the current Blackwell generation. However, overall compute requirements are also increasing rapidly, meaning that no real breakthrough is yet apparent in this regard.
Between euphoria and solid fundamentals
The AI investment wave of 2026 was underestimated, and further increases in spending are already visible for 2027. At the same time, there has been significant progress in monetisation. AI infrastructure providers in particular are likely to benefit and therefore remain attractive, despite possible correction phases.
Importantly, not every company investing in AI is automatically a winner. What matters is who occupies a position along the AI value chain that is difficult to replace. It remains to be seen whether the gap between the current winners and losers of the AI trend will widen further or begin to narrow again. New technologies such as TurboQuant or more energy-efficient AI chips could also shift the balance of power once more.
Legal information
Marketing advertisement: All information published here is for your information only and does not constitute investment advice or any other recommendation. The statements contained in this document reflect the current assessment of DJE Kapital AG. These may change at any time without prior notice. All statements made have been made with care in accordance with the state of knowledge at the time of preparation. However, no guarantee and no liability can be assumed for the correctness and completeness.