The AI bubble: why the AI boom is a financial-market risk
As of: 2026-07-04
The "AI bubble" refers to the possibility that share prices around artificial intelligence have risen faster than real profits and productivity gains. Whether it really is a bubble and whether or when it bursts is open – but several supervisory authorities warn of stretched valuations and market concentration.
Prices around artificial intelligence have risen enormously in just a few years. Whether this is a speculative bubble is a possibility – not a fact. This page puts the known data into context, shows the web of companies involved and explains why a concentration in a handful of stocks can also be a cluster risk for Swiss retirement provision.
What is the thesis?
The thesis is: the share prices of AI companies may have risen faster than real profits and demonstrable productivity gains. A bubble is always a possibility, not a fact – whether and when it bursts cannot be predicted seriously.
Several institutions warn of stretched valuations and market concentration: the International Monetary Fund (IMF), the Bank of England, the European securities regulator ESMA, the Bank for International Settlements (BIS) and the US Federal Reserve. These are risk assessments by supervisory authorities – not an accusation of accounting fraud.
Valuation and concentration
The seven largest US technology stocks (the "Magnificent 7") account for roughly 33–35 % of the US benchmark S&P 500 and about 22–25 % of the global MSCI World index. Their expected price/earnings ratio (forward P/E) is around 31–36, compared with roughly 20–21 for the broad rest of the market.
The long-term Shiller CAPE valuation measure is close to 40 – near the dot-com peak of about 44 in the year 2000. High valuations do not predict a crash, but they increase the drop height if expectations are disappointed.
The capital-expenditure and energy appetite
The large cloud companies (Microsoft, Amazon, Alphabet, Meta) have ramped up their data-centre investments massively: from around 226 Mrd USD (2024) to about 410 Mrd USD (2025), with guidance of roughly 725 Mrd USD for 2026. The investment bank Goldman Sachs estimates cumulative spending for 2025–2030 at about 5,3 Bio USD.
Electricity use is rising too: according to the International Energy Agency (IEA), data centres used around 415 TWh in 2024 (about 1,5 % of global electricity) and could reach around 945 TWh by 2030. The key question for the bubble debate is whether these outlays are matched by lasting revenues.
Circular financing: the web of companies
Striking is the mutual entanglement of the players: chipmakers invest in AI labs, which in turn buy chips and cloud capacity from the same or affiliated firms. Observers speak of "circular financing", because part of the revenue appears to flow between the same participants.
The adjacent relationship chart shows the doubly confirmed core deals. Important: many large figures are letters of intent ("up to", LOI) that have not yet flowed – these are marked as "planned". The risk is not a single deal but the contagion risk: if one link in the chain falters, it can hit many others.
Comparison with the dot-com bubble
What it shares with the year 2000 is high concentration and high valuations. An important difference: today's groups are highly profitable – together they generate around 350 Mrd USD of free cash flow – and finance the build-out largely from their own resources rather than debt.
According to the IMF, the current risk is therefore more a shareholder than a banking-system risk: losses would primarily hit investors, less so the credit system. The BIS, however, points out that debt financing of data centres is increasing.
How institutions assess the risk
No one can predict the timing of a market drop; what is demonstrable is the range of institutional assessments. The International Monetary Fund (IMF) described equity valuations as "stretched" in its Global Financial Stability Report (October 2025) – overvalued by around 10 percentage points – and warned of a possible sudden, sharp correction; market concentration is said to be higher than during the dotcom bubble.
The Bank of England (Financial Stability Report, December 2025) judged many valuations to be materially stretched, especially in AI tech, and saw an increased risk of a sharp correction. The US Federal Reserve (November 2025) placed valuations near the upper end of their historical range and named AI sentiment for the first time as a possible trigger of a correction.
The EU securities regulator ESMA (February 2026) stresses operational concentration risks more than a crash: dependence on a few infrastructure and third-party providers can have systemic effects. The Bank for International Settlements (BIS, January 2026) warns about the shift from cash-flow to debt financing and the growing, opaque private-credit market. Concrete crash probabilities or dates cannot be quantified seriously – individual figures of that kind are individual opinions, not a forecast.
The counter-position (bull case)
Arguments for sustained high valuations include real profits and cash flows as well as contractually committed demand (such as order backlogs at AWS and Oracle). Valuations are also below the extremes of the dot-com era.
It remains open whether AI delivers broadly measurable productivity gains: studies by Goldman Sachs (2026) and the research bureau NBER reach different conclusions – this is contested and not conclusively established.
Relevance for Switzerland and for your own provision
Anyone investing broadly in "the world" is more concentrated in a few US tech stocks than many people think: around a quarter of the MSCI World is in the seven largest US tech names. This affects many pillar-3a and pension-fund products with a global index reference.
The Swiss National Bank (SNB) also holds US technology stocks directly – including around 10 Mrd USD of Apple and around 11 Mrd USD of Nvidia. The SNB's total US equity portfolio is estimated at around 167 Mrd USD (according to swissinfo). According to a Complementa study 2025, the average equity quota of Swiss pension funds was 32,5 % with a foreign share of 49,3 %.
Warning signs – indicators, not a forecast
Observable indicators (not a prediction): high forward P/E and CAPE values; high index concentration; a gap between investments and actual revenues; increasing debt financing of data centres (according to the BIS, fund lending to AI firms rose from around 3 Mrd USD in 2010 to over 40 Mrd USD in 2025).
Further signals: large insider sales (Nvidia chief Jensen Huang sold shares worth over 1 Mrd USD in 2024–2025 under a pre-arranged sales plan under SEC rule 10b5-1) and sentiment shocks such as the DeepSeek shock of 27 January 2025, which triggered a sharp AI sell-off (Nvidia lost around 589 Mrd USD of market value in a single day, according to Bloomberg/CNBC).
What you can do (general provision principles)
These are general provision principles, not investment advice: review your own cluster risk in pillar 3a, pension fund and ETFs; diversify broadly; keep a liquidity buffer; know your own time horizon; and periodically bring positions back to their target weighting (rebalancing).
For personal financial decisions, please consult independent, qualified professionals. PrepperGuide is not a financial institution and gives no recommendations.
The web of companies
Show the list of companies
| Company | Role | Ticker / Exchange |
|---|---|---|
| NVIDIA | Chips / GPU | NVDA · NASDAQ |
| AMD | Chips / GPU | AMD · NASDAQ |
| Broadcom | Chips (custom) | AVGO · NASDAQ |
| Microsoft | Hyperscaler / cloud | MSFT · NASDAQ |
| Alphabet (Google) | Hyperscaler / cloud | GOOGL · NASDAQ |
| Amazon | Hyperscaler / cloud | AMZN · NASDAQ |
| Meta | Hyperscaler / cloud | META · NASDAQ |
| Oracle | Cloud / infrastructure | ORCL · NYSE |
| Super Micro | Servers / infrastructure | SMCI · NASDAQ |
| Dell | Servers / infrastructure | DELL · NYSE |
| Vertiv | Data-centre infrastructure | VRT · NYSE |
| Intel | Foundry (contract manufacturing) | INTC · NASDAQ |
| Arm | Chip IP | ARM · NASDAQ |
| Nebius | GPU cloud | NBIS · NASDAQ |
| TSMC | Foundry (contract manufacturing) | TSM · NYSE |
| ASML | Manufacturing equipment | ASML · NASDAQ |
| SoftBank | Investor | SFTBY · OTC/Tokyo |
| OpenAI | AI lab | private (not listed) |
| Anthropic | AI lab | private (not listed) |
| xAI | AI lab | private (not listed) |
| Mistral | AI lab | private (not listed) |
| CoreWeave | GPU cloud | private (not listed) |
| Stargate (JV) | Infrastructure joint venture | private (not listed) |
Show relationships as a table
| From | Relationship | To | Key figures |
|---|---|---|---|
| NVIDIA | invests in planned | OpenAI | ≤ 100 Mrd USD · 10 GW · LOI 22.09.2025 |
| NVIDIA | holds a stake in | CoreWeave | Backstop ≤ 6,3 Mrd USD (bis 2032) + Equity |
| NVIDIA | invests in planned | xAI | ≤ 2 Mrd USD (via SPV) |
| NVIDIA | invests in planned | Anthropic | ≤ 10 Mrd USD · 18.11.2025 |
| NVIDIA | holds a stake in | Intel | 5 Mrd USD · ~4% · vollzogen Dez. 2025 |
| Microsoft | holds a stake in | OpenAI | ~27% (~135 Mrd USD) · 28.10.2025 |
| OpenAI | is a customer of | Microsoft | Azure · + 250 Mrd USD Abnahme |
| OpenAI | is a customer of planned | Oracle | ~300 Mrd USD · 5 J · 4,5 GW (Stargate) |
| AMD | partnership with | OpenAI | 6 GW + Warrant ≤ ~10% |
| OpenAI | partnership with | Broadcom | 10 GW Custom-Chips |
| SoftBank | invests in | OpenAI | Lead 40-Mrd-USD-Runde · 31.03.2025 |
| OpenAI | partnership with planned | Stargate (JV) | 500 Mrd USD · 4 J · 21.01.2025 |
| Oracle | partnership with planned | Stargate (JV) | Infrastruktur-Partner |
| SoftBank | partnership with planned | Stargate (JV) | Infrastruktur-Partner |
| Amazon | invests in | Anthropic | 8 Mrd + ≤ 25 Mrd USD |
| Anthropic | is a customer of planned | Amazon | AWS · > 100 Mrd USD · 10 J · 5 GW Trainium |
| Alphabet (Google) | invests in planned | Anthropic | ≤ 40 Mrd USD |
| Anthropic | is a customer of planned | Alphabet (Google) | ≤ 1 Mio TPU (> 1 GW) |
| Microsoft | invests in planned | Anthropic | ≤ 5 Mrd USD |
| Anthropic | is a customer of | Microsoft | Azure · 30 Mrd USD |
| ASML | invests in | Mistral | 1,3 Mrd EUR · ~11% · 09.09.2025 |
| Dell | supplies | xAI | ~5 Mrd USD |
| Super Micro | supplies | xAI | Colossus-Cluster |
| TSMC | supplies | NVIDIA | CoWoS-Kapazität (~60%) |
| SoftBank | holds a stake in | Arm | ~87–90% |
Companies in the AI network
Expandable short profiles of the key companies around the AI boom – grouped by their role. Factual information, not investment advice.
Chips & hardware
NVIDIA
Semiconductors — graphics and AI accelerators (GPU)
- Headquarters
- Santa Clara, California (USA)
- Profile
- NVIDIA designs graphics processors (GPUs) and computing platforms that have become the de facto standard for training large AI models. NVIDIA designs the chips itself but has them manufactured externally (above all at TSMC).
- Role in the network
- Central node — at once supplier and investor: announced stakes in, among others, OpenAI (intended up to USD 100 billion, according to Nvidia not yet definitive at the end of 2025), Anthropic, xAI, CoreWeave and Intel (USD 5 billion / ~4%, completed Dec. 2025).
- Key figure
- USD 130.5 bn Revenue fiscal year 2025 (approx. +114%)
AMD
Semiconductors — CPUs, GPUs and AI accelerators
- Headquarters
- Santa Clara, California (USA)
- Profile
- AMD designs processors (CPUs) and AI graphics accelerators (GPUs) for PCs, servers and data centres. In the AI boom, AMD is the most important direct challenger to NVIDIA in AI chips.
- Role in the network
- Partnership with OpenAI (purchase of around 6 gigawatts of computing capacity) including options (warrants) for OpenAI on up to around 10% of AMD — an example of mutual entanglement.
- Key figure
- USD 34.6 bn Revenue fiscal year 2025 (record, approx. +34%)
Broadcom
Semiconductors (networking/custom chips) and infrastructure software
- Headquarters
- Palo Alto, California (USA)
- Profile
- Broadcom designs networking chips and tailor-made AI accelerators (custom ASICs) that large tech groups use to have their own AI chips built. After the VMware acquisition, Broadcom is also a large provider of enterprise software.
- Role in the network
- Broadcom and OpenAI agreed on the joint development of custom AI chips on the order of around 10 gigawatts (no official monetary value stated).
- Key figure
- USD 63.9 bn Revenue fiscal year 2025 (approx. +24%)
Intel
Semiconductors — processors (CPU) and contract manufacturing
- Headquarters
- Santa Clara, California (USA)
- Profile
- Intel is one of the most established chipmakers (above all CPUs) and is at the same time building a contract-manufacturing business (Intel Foundry). In AI accelerators Intel is regarded as a latecomer, but it is strategically important because Western chip manufacturing is scarce.
- Role in the network
- NVIDIA took a stake of USD 5 billion (around 4%) in Intel (announced Sept. 2025, completed Dec. 2025).
- Key figure
- USD 52.9 bn Revenue fiscal year 2025 (slight decline)
Arm Holdings
Chip architecture/IP — licensor of processor designs
- Headquarters
- Cambridge (United Kingdom)
- Profile
- Arm develops energy-efficient processor architectures and earns from licences and per-unit fees (royalties): almost all smartphones and, increasingly, AI and server chips use Arm designs. The company is majority-owned by SoftBank.
- Role in the network
- SoftBank holds around 87–90% of Arm; Arm is also a technology partner in the Stargate joint venture.
- Key figure
- USD 4.0 bn Revenue fiscal year (as of 31.03.2025; record)
TSMC
Semiconductor contract manufacturing (foundry)
- Headquarters
- Hsinchu (Taiwan)
- Profile
- TSMC is the world's largest pure-play semiconductor foundry: the company does not design its own chips but produces them for others (including NVIDIA, AMD, Apple, Broadcom). Because the most advanced AI chips are made almost only at TSMC, it is a critical bottleneck.
- Role in the network
- TSMC manufactures the AI chips for NVIDIA and others; the packaging technology (CoWoS) for NVIDIA was reportedly sold out through the end of 2026.
- Key figure
- NT$ 3,809 bn (~USD 122 bn) Net revenue fiscal year 2025
ASML Holding
Semiconductor equipment — lithography systems (incl. EUV)
- Headquarters
- Veldhoven (Netherlands)
- Profile
- ASML builds the lithography machines with which chipmakers create the finest structures. In the most modern EUV technology, ASML is the only provider worldwide — without ASML machines the most advanced AI chips cannot be manufactured.
- Role in the network
- ASML invested EUR 1.3 billion in the French AI lab Mistral and thereby became, at around 11%, its largest shareholder.
- Key figure
- EUR 32.7 bn Net revenue fiscal year 2025 (net profit EUR 9.6 billion)
Super Micro Computer
Hardware — high-performance servers and complete systems
- Headquarters
- San Jose, California (USA)
- Profile
- Super Micro builds servers, storage systems and complete, partly liquid-cooled data-centre racks for AI. The company builds AI chips (e.g. from NVIDIA) into ready-to-deploy systems — a system integrator between chip and data centre.
- Role in the network
- Super Micro supplies server systems to xAI (for its data centre "Colossus").
- Key figure
- USD 22.0 bn Revenue fiscal year 2025
Dell Technologies
Hardware — PCs, servers, storage and IT infrastructure
- Headquarters
- Round Rock, Texas (USA)
- Profile
- Dell makes PCs, servers and storage systems and supplies complete IT infrastructure. In the AI boom, Dell mainly sells AI servers in which the accelerator chips are built.
- Role in the network
- Dell supplies AI server systems to xAI (scope around USD 5 billion).
- Key figure
- USD 95.6 bn Revenue fiscal year 2025 (as of 31.01.2025; approx. +8%)
Hyperscalers & cloud
Microsoft
Software and cloud (Azure) — hyperscaler and investor
- Headquarters
- Redmond, Washington (USA)
- Profile
- Microsoft is one of the largest software and cloud groups; the cloud division Azure rents out computing capacity on which many AI services run. Microsoft is at the same time a large investor in and cloud supplier to the AI labs — a typical dual role in the entangled AI ecosystem.
- Role in the network
- Microsoft holds around 27% of OpenAI (valued at ~USD 135 billion, as of 28.10.2025); in return OpenAI draws Azure capacity. In addition, an investment in Anthropic (up to USD 5 billion).
- Key figure
- USD 281.7 bn Revenue fiscal year 2025 (as of 30.06.2025)
Alphabet (Google)
Internet/advertising, cloud and AI research
- Headquarters
- Mountain View, California (USA)
- Profile
- Alphabet is the parent company of Google and earns mainly from online advertising; through Google Cloud the group rents out computing capacity, and through Google DeepMind it develops its own AI models (Gemini). Google also builds its own AI chips (TPU).
- Role in the network
- Alphabet announced investments of up to USD 40 billion in Anthropic; in return Anthropic intends to use up to around 1 million Google TPU chips.
- Key figure
- USD 402.8 bn Revenue fiscal year 2025 (approx. +15%)
Amazon
Online retail and cloud (AWS) — hyperscaler and investor
- Headquarters
- Seattle, Washington (USA)
- Profile
- Amazon is the largest online retailer in the West and, with AWS, operates the world's largest cloud provider. AWS is a central renter of computing capacity; with "Trainium", Amazon also develops its own AI chips.
- Role in the network
- Amazon invested USD 8 billion in Anthropic and announced up to USD 25 billion more; in return Anthropic committed to using AWS (over USD 100 billion / 10 years).
- Key figure
- USD 716.9 bn Net revenue fiscal year 2025 (approx. +12%)
Meta Platforms
Social networks/advertising — hyperscaler with its own AI models
- Headquarters
- Menlo Park, California (USA)
- Profile
- Meta operates Facebook, Instagram and WhatsApp and develops its own large AI models (Llama). For the AI build-out, Meta invests massively in data centres and chips, but uses its computing capacity predominantly itself rather than renting it out.
- Role in the network
- Meta is one of the four large hyperscalers whose capital expenditure (capex) is driving the AI infrastructure build-out.
- Key figure
- USD 201.0 bn Revenue fiscal year 2025 (approx. +22%)
Oracle
Database/enterprise software and cloud infrastructure (OCI)
- Headquarters
- Austin, Texas (USA)
- Profile
- Oracle is a traditional provider of database and enterprise software and is strongly expanding its cloud business (OCI) towards AI computing capacity. Oracle positions itself as a large renter of data-centre capacity for AI labs.
- Role in the network
- Oracle is a cloud supplier to OpenAI (announced purchase, around 4.5 GW) and an equity partner in the Stargate joint venture.
- Key figure
- USD 57.4 bn Revenue fiscal year 2025 (as of 31.05.2025)
Nebius Group
AI cloud/GPU infrastructure ("neocloud")
- Headquarters
- Amsterdam (Netherlands)
- Profile
- Nebius operates specialised AI data centres and rents out GPU computing capacity for training and running AI models. The company emerged from the Western-oriented part of the former Russian internet group Yandex.
- Role in the network
- Nebius is an emerging GPU cloud provider ("neocloud") in the environment of the large hyperscalers (no doubly confirmed bilateral large-scale contract documented).
- Key figure
- USD 529.8 m Revenue fiscal year 2025 (strong growth)
CoreWeave
GPU cloud — renting of AI computing capacity ("neocloud")
- Headquarters
- Livingston, New Jersey (USA)
- Profile
- CoreWeave is a cloud provider specialised in AI that rents out large amounts of GPU computing capacity. Originally started in crypto mining, the company transformed into a pure AI data-centre renter and went public in March 2025.
- Role in the network
- NVIDIA granted CoreWeave a purchase guarantee (backstop up to USD 6.3 billion) and holds a stake — a close entanglement between chipmaker and GPU renter.
- Key figure
- USD 40.00 / share IPO on 28.03.2025, per share (Nasdaq)
AI labs
OpenAI
AI lab — development of AI models and products
- Headquarters
- San Francisco, California (USA)
- Profile
- OpenAI develops large AI models (GPT) and products such as ChatGPT. With ChatGPT (end of 2022) the company helped trigger the current AI boom; it depends on huge amounts of external computing capacity and capital.
- Role in the network
- Central node: Microsoft holds around 27%; OpenAI draws cloud from Microsoft and Oracle, struck chip partnerships with AMD and Broadcom and is the initiator of the Stargate joint venture.
- Key figure
- USD 852 bn Valuation (round with ~USD 122 billion of committed capital, announced 31.03.2026) per media reports
Anthropic
AI lab with a safety focus — development of the Claude models
- Headquarters
- San Francisco, California (USA)
- Profile
- Anthropic is an AI lab, founded by former OpenAI staff, that develops the Claude models with a particular emphasis on AI safety. It is the most important direct competitor of OpenAI and finances itself through several large tech groups.
- Role in the network
- Unusually broadly entangled: commitments from Amazon (USD 8 billion + up to USD 25 billion), Alphabet (up to USD 40 billion), NVIDIA (up to USD 10 billion) and Microsoft (up to USD 5 billion); in return cloud from AWS, Google and Azure.
- Key figure
- USD 965 bn Valuation (Series H, USD 65 billion, May 2026; official announcement)
xAI
AI lab — development of the model "Grok"
- Headquarters
- San Francisco Bay Area, California (USA)
- Profile
- xAI is the AI company founded by Elon Musk that develops the chatbot "Grok" and is connected to the platform X. For training, xAI operates very large data centres (including "Colossus") and is a younger but capital-rich challenger.
- Role in the network
- xAI received an investment from NVIDIA (up to USD 2 billion) and draws hardware from Dell (around USD 5 billion) as well as server systems from Super Micro.
- Key figure
- USD 20 bn Funding round; rumoured valuation around USD 230 billion (early 2026) per media reports
Mistral AI
AI lab — open and commercial language models
- Headquarters
- Paris (France)
- Profile
- Mistral AI is the best-known European AI lab and develops large language models, some of which are available as open source. It is regarded as the European answer to the US AI labs.
- Role in the network
- ASML invested EUR 1.3 billion in Mistral and became, at around 11%, its largest shareholder (Series C, Sept. 2025).
- Key figure
- EUR 1.7 bn Series C round at a valuation of EUR 11.7 billion (Sept. 2025, led by ASML)
Infrastructure, energy & capital
Vertiv Holdings
Data-centre infrastructure — power supply and cooling
- Headquarters
- Westerville, Ohio (USA)
- Profile
- Vertiv supplies the "physical layer" of data centres: power supply, backup power and cooling. Because AI chips consume enormous amounts of electricity and generate heat, powerful cooling is a bottleneck of the AI build-out — and Vertiv a central beneficiary.
- Role in the network
- Industry-wide supplier of critical data-centre infrastructure (power/cooling) for the AI build-out (no single doubly confirmed bilateral contract).
- Key figure
- ~USD 10 bn Annual revenue (order of magnitude)
Crusoe Energy Systems
Construction and operation of AI data centres ("AI Factory")
- Headquarters
- Denver, Colorado (USA)
- Profile
- Crusoe builds and operates energy-optimised AI data centres and describes itself as an "AI Factory" company. It combines energy supply with data-centre infrastructure for AI workloads.
- Role in the network
- Specialised developer/operator of AI data centres (no doubly confirmed contractual link to another node documented).
- Key figure
- USD 1.375 bn Series E funding at a valuation of over USD 10 billion (Oct. 2025)
Stargate (Joint Venture)
AI infrastructure joint venture — construction of data centres
- Headquarters
- USA (distributed sites; flagship Abilene, Texas)
- Profile
- Stargate is a joint large-scale project of several groups to build AI data centres in the USA, above all for OpenAI. It is not an ordinary company but a consortium with distributed construction sites.
- Role in the network
- Joint infrastructure vehicle of OpenAI, Oracle, SoftBank and MGX, with technology partners such as Arm, Microsoft, NVIDIA and Oracle.
- Key figure
- USD 500 bn Announced investment volume (up to; over around four years; intended)
SoftBank Group
Investment/holding company
- Headquarters
- Tokyo (Japan)
- Profile
- SoftBank Group is a Japanese holding company that, through its Vision Funds, invests worldwide in technology and AI companies. Among its most important holdings is the chip designer Arm. SoftBank is one of the largest and most active financiers of the AI boom.
- Role in the network
- SoftBank holds around 87–90% of Arm, led the USD 40 billion round of OpenAI (31.03.2025) and is an equity partner in the Stargate joint venture.
- Key figure
- JPY 7.8 tn Group revenue fiscal year (as of 31.03.2026)
Price charts
As of: 2026-08-07
As of: 2026-08-07
Not investment advice – information only. Liability declined.
As of: 2026-08-07
Not investment advice – information only. Liability declined.
As of: 2026-08-07
Not investment advice – information only. Liability declined.
As of: 2026-08-07
Not investment advice – information only. Liability declined.
As of: 2026-08-07
Not investment advice – information only. Liability declined.
As of: 2026-08-07
Not investment advice – information only. Liability declined.
As of: 2026-08-07
Not investment advice – information only. Liability declined.
As of: 2026-08-07
Not investment advice – information only. Liability declined.
As of: 2026-08-07
Not investment advice – information only. Liability declined.
As of: 2026-08-07
Not investment advice – information only. Liability declined.
Frequently asked questions
What is the AI bubble?
The "AI bubble" refers to the possibility that share prices around artificial intelligence have risen faster than real profits and productivity gains. Whether it really is a bubble and whether or when it bursts is open – but several supervisory authorities warn of stretched valuations and market concentration.
Am I affected through my pillar 3a or pension fund?
Possibly indirectly: global equity indices such as the MSCI World consist of roughly a quarter of the seven largest US tech stocks. Many 3a and pension-fund products track such indices. It is worth reviewing your own cluster risk. This is not investment advice – consult an independent professional if needed.
What is circular financing?
Circular financing means that firms invest in one another while at the same time being customers of one another – for example when a chipmaker invests in an AI lab that then buys chips from the same maker. This increases mutual dependence and thus contagion risk.
Further links
- Finance: Valuables in crisis What really counts in a crisis
- Finance: Bank crisis & deposit insurance When market concentration hits the banks
- Finance: Inflation & purchasing-power loss How currency value and savings come under real pressure
- Guide: Setting up emergency supplies Putting preparedness into practice
Related hazard
To the hazard "Financial markets / AI bubble" in the Hazards chapter
Sources
- IWF (IMF): Global Financial Stability Report – Shifting Ground beneath the Calm (2025-10-14) Original source Local copy
- Bank of England: Financial Stability Report – December 2025 (2025-12-02) Original source Local copy
- ESMA: TRV – AI adoption and trends in securities markets (ESMA50-481369926-30599) (2026-02-20) Original source Local copy
- BIZ (BIS): Bulletin No 120 – Financing the AI boom: from cash flows to debt (2026-01-07) Original source Local copy
- US Federal Reserve: Financial Stability Report (2025-11-07) Original source Local copy
- IEA: Energy and AI (2025-04-10) Original source Local copy
- NVIDIA: NVIDIA and OpenAI announce strategic partnership (up to USD 100 bn / 10 GW) (2025-09-22) Original source
-
Fortune:
Nvidia's $100 billion investment in OpenAI raises 'circular financing' questions
(2025-09-28)
Critics call the wave of cross-investments between chipmakers and AI labs 'circular financing'.
Original source Archive -
CNBC:
Nvidia drops nearly 17% as DeepSeek triggers a sell-off in AI stocks
(2025-01-27)
DeepSeek's low-cost model triggered the steepest one-day market-cap loss in US history for Nvidia.
Original source Archive -
swissinfo.ch:
Swiss central bank holds a roughly $167bn tech-led US stock portfolio
(2025-08-01)
The SNB's US equity holdings include around $11bn in Nvidia and $10bn in Apple.
Original source Archive -
Complementa:
Schweizer Pensionskassen-Studie 2025 (Aktienquote 32,5%, Auslandanteil 49,3%)
(2025-05-01)
Die durchschnittliche Aktienquote lag 2025 bei 32,5 %, davon 49,3 % im Ausland.
Original source Archive -
Bloomberg:
AI Circular Deals (graphic of cross-investments among AI firms)
(2026-01-15)
A visual map of the overlapping investments tying chipmakers, hyperscalers and AI labs together.
Original source Archive