Last week : Oil price (flat but remains high) FED & BOJ hiked (+25bps each) stronger USD (+1%) Bond yields up (+10bps)
WEEKLY TRENDS
Last week, it was Central Banks decisions’ week, first with the FED and its +25bps hike, the first in 3 years (reversing 3 cuts in 2025, markets now wait for at least 2 more hikes until end of 2027) then BOE who kept rates unchanged as expected on Thursday, and finally with the BOJ on Friday morning who raised rates by 25bps, having its rate at a 31 year high (1.25%).
All those decisions were expected by the markets, but the facts are, with higher USD rates, the USD got stronger and the JPY weaker as the markets expect more hikes to come. Oil price remains very high while reserves are at their lows worldwide (estimated at 1.8trn barrels). Meanwhile, the US Senate rejected the Clarity Act, yet BTC surged by 5% last week. RE was down -2.5% and PE was also in negative territory at -1.5% on the week.
As Yields went up, Bonds went lower, YTD global indices valuations are now at -5% for US, EU and EM Bonds. The Nasdaq seemed to remain safe last week with a return of +0.7% thanks to Semi-conductors and Healthcare stocks. NB we now have entered the ‘Blackout period’ for 40% of the SP500 stocks.
MARKETS
Equities
Q2 earnings weekly performances :
ADP (-1%) Bolloré (-1%) Kier (+27%)
NB : OVH (+13%) upgrade to ‘o/w’ from BNPP ; MAN (+10%) upgrade to ‘buy’ from UBS ; CoreWeave (-8%) dilution risk after $3bn convertibles
On 12 September, Dario Amodei, CEO of Anthropic, called on the artificial intelligence industry to slow down the development of new models. For investors, a question immediately arises: should this be seen as the first sign of a turnaround in the AI infrastructure investment cycle? In my view, the risk lies more in the expectations that currently underpin valuations.
First, we need to distinguish between two uses of computing power: model training on the one hand, and model deployment or inference on the other. Whilst training requires a huge amount of computing power, albeit on an ad hoc basis, inference occurs with every query: summarising a document, drafting an email or answering a question. Although each query requires less computing power, the sheer volume of billions of daily uses nevertheless results in considerable overall consumption. According to Deloitte, training in 2023 accounted for two-thirds of the computing power dedicated to AI, compared with an expected one-third in 2026. This proportion is set to continue falling as usage becomes more widespread. A slowdown in model development would therefore, at first glance, have a limited effect on total demand for computing power.
However, this does not mean that progress in model development has become secondary. The main obstacle to their deployment is not merely their ability to perform a simple task, but their reliability when dealing with complex, multi-step processes. If a model has a 95% chance of succeeding at each of the ten stages of a process, the probability of the process as a whole succeeding falls to around 60%. This is one of the reasons why certain applications remain difficult to deploy on a large scale. More reliable and accurate models can therefore unlock new applications, particularly in the business sector.
This is also what makes the slowdown potentially ambivalent. Giving stakeholders more time to improve the reliability of models could enhance their reliability when applied to complex processes and, as a result, accelerate certain corporate roll-outs; however, if the complexity of the models increases at a slower rate, current investments in AI infrastructure could also prove to be based on overly optimistic assumptions and have a significant impact on the profitability of these investments in the short term. The key issue is economic. A data centre’s revenue depends on three variables that are multiplying: the number of tasks entrusted to AI, the amount of computation required per task, and the cost of that computation. The third factor is falling rapidly, driven by advances in chips and software. According to Epoch AI, the cost of computation could fall by a factor of five to ten per year, and this decline would then force infrastructure providers to significantly increase volumes in order to maintain revenue growth. The other two factors, however, are closely linked to the capabilities of the models. The computational power required per task may increase as models ‘think’ more deeply or carry out longer tasks, and the number of tasks increases as new applications become economically and technically feasible. A truly autonomous assistant, for example, could transform the volume of computing power consumed by a business far more significantly than a simple improvement in the quality of responses. Halting the progress of models therefore risks slowing down precisely the two factors likely to offset the structural decline in the price of computing. The roll-out of already mature applications would continue but might not be sufficient to achieve the levels of growth factored into current valuations. The impact would not, however, be uniform across the value chain. Model developers benefit directly from cheaper computing: their training costs fall, as does the price of the tokens they resell, and their margins may improve. Their vulnerability lies instead in price pressure. As open-source models catch up with state-of-the-art models, the gap in capabilities – and therefore the premium that customers are willing to pay – narrows. A state-of-the-art model from eighteen months ago can quickly become almost free.
This price pressure has a direct impact on infrastructure providers: chip manufacturers, capacity providers, electricity generators and network operators sell computing power and must therefore offset the fall in unit price by increasing volumes. It is they who bear the brunt of the $630–690 billion in investment projected for this year, based on the assumption that volumes will grow faster than prices fall. In the short term, the 2026 budgets already committed under contract are relatively protected. The real challenge lies in 2027 and 2028, when budgets will depend more heavily on growth forecasts for AI deployment. If these forecasts begin to be revised downwards, the impact on the markets could be immediate: a fall in spending forecasts, a downward revision of expected profits for providers and a compression of valuation multiples – even as physical demand for computing power remains high. The impact would therefore not necessarily be negative for the cloud giants, as lower investment could also mean more available cash; however, infrastructure providers would be much more vulnerable. A slowdown in AI would not call into question the usefulness of data centres. However, it could undermine the growth assumptions on which current investments are based. If revenues are slow to materialise, the main risk would then be financial, with a direct impact on valuations.
Wisdom Tree, 6 August 2026
Authors : Aneeka Gupta, Associate Director, Research
Few investment themes intersect as directly with the energy transition, national security and technological advances as strategic metals and rare earths. Copper and lithium are essential for powering electric vehicles and the electricity grid. At the same time, certain rare earths such as neodymium and dysprosium are used in the composition of permanent magnets that power electric vehicle (EV) motors and wind turbines. Furthermore, metals such as tin, silver, germanium and antimony now play a key role in defence systems and cutting-edge electronics. Demand for these materials is expected to grow for several decades. The real question facing the market is not about their importance, but rather about who controls them. For the time being, the answer is China, and this is no coincidence. Thirty years of a persistently pursued industrial policy have enabled China to build the most strategic position in critical raw materials of the modern era, centred on the processing and refining stages – the very stages that are the most difficult to replicate.
The July correction is cyclical; the thesis is structural. The standout development was a sharp reversal in rare earths. MP Materials, in the United States, fell by 26%; in Australia, Lynas dropped by 21%, whilst on the Chinese side, producers such as Shenghe Resources Holdings and Grinm Advanced Material Co saw their share prices plunge by 29 % and 47% respectively. China thus found itself at the bottom of the country allocation rankings, posting a decline of 13.7% and recording the most significant negative contribution. The United States followed closely behind, also penalised by the same players in the rare-earths sector. These losses were offset by gains from Indonesian nickel and tin producers (Vale Indonesia (+27%), Metals X Ltd (+21%), Pt Timah Tbk (+16%)) as well as by the rebound seen among South African mining companies. In terms of size, it was the medium-sized companies that suffered the most, with an 11% decline, whilst the major producers once again fared better.
This year, hedge funds have increased their short positions in several US companies in the critical minerals sector, including MP Materials, US Antimony and American Resources. In their view, even though Washington’s support remains substantial, it will take years to erode China’s influence, and this support alone may not be enough. Short positions in US Antimony, which accounted for around a quarter of its market capitalisation at the start of the year, now exceed 40%. As for American Resources, a company that generated no turnover last year, the proportion of short positions has risen from less than 10% to more than a fifth of its shares. This scepticism is based on two points :
Firstly, several of these companies saw their share prices triple in 2025 thanks to government equity investments, loans and purchase agreements, leading to valuations that critics believe are driven more by political announcements than by fundamentals.
Secondly, and on a more structural level, a report by SAFE’s Centre for Critical Minerals Strategy asserts that the main constraint facing the West is financial rather than geological. There is a shortfall in public funding, whilst mines and processing plants are highly capital-intensive and can take a decade or more to build; meanwhile, China retains the ability to adjust prices by flooding the market with supply as soon as competitors reach a certain scale. In our view, none of this undermines the structural thesis; on the contrary, it reinforces it. The argument in favour of a decline rests primarily on the concentration observed within a small number of speculative US stocks, which frequently lack revenue or are dependent on a single commodity.
Loomis Sayles, 13 August 2026
Authors : Elisabeth Colleran, co-Head EM Debts, Portfolio Manager
Emerging markets are now at the heart of two major investment trends: artificial intelligence and energy. Yet a common misconception persists: that the recent energy shock has undermined their resilience. This is precisely where the ‘lie’ lies. Historically, a sharp rise in energy prices has often weighed on emerging economies, particularly those heavily reliant on imports. But this time, their response has been swift. Governments have employed a range of measures, from subsidies and consumption-restriction measures to the diversification of oil supplies.
The exceptional surge in global investment in artificial intelligence is one of the factors underpinning this resilience. Moreover, certain seemingly negative signals should be interpreted with caution. In emerging Asia, the widening of trade deficits is notably linked to rising energy imports. However, it also reflects an increase in imports of capital goods and industrial intermediate goods, which are likely associated with the development of the infrastructure required for AI.
AI relies on extremely complex supply chains. A large proportion of these supply chains pass through emerging markets, particularly in Asia. They encompass IT components, data centre equipment, silicon wafers, electronic components and energy storage systems. China, South Korea, Taiwan, Vietnam, Malaysia and Singapore are key players in this sector. The strong demand for semiconductors is already reflected in the rise in technology exports from South Korea and Taiwan. The Philippines also plays a significant role in assembly and testing, whilst India is gradually developing its design and production capabilities within the AI ecosystem. This wave of investment is not limited to technology.
The rapid expansion of data centres and AI-related infrastructure is also driving a structural increase in demand for electricity. According to estimates from the International Energy Agency, global electricity demand is expected to grow by 3.6% per year between 2026 and 2030, compared with an average of 2.8% over the previous decade. This will require significant investment in the modernisation of electricity grids, storage and electricity transmission infrastructure. Several Asian countries are already among the world leaders in electrification and renewable technologies. China, in particular, dominates the production of solar and electrical equipment. China and South Korea play a key role in battery storage, whilst India is a major producer of wind turbines. By highlighting the need to secure supplies, it is prompting countries to invest more in their own generation capacity and reduce their reliance on imports. China is therefore undertaking a five-year programme to modernise its electricity grid, worth over $700bn.
Raw materials: another strategic pillar
Beyond semiconductors and electricity, the rise of AI also depends on a wide range of raw materials. EM play a vital role. Chile and Peru account for nearly a third of the world’s copper reserves. Indonesia is a major supplier of tin, nickel and copper, whilst China remains a key player in rare earths. In Africa, the Democratic Republic of the Congo dominates the cobalt market by a wide margin, with an estimated share of global production of around 70%. These resources represent a significant structural advantage for emerging economies. They enable these economies not only to benefit from rising global demand but also, in the long term, to move up the value chain by developing greater local processing and production capacity.
The true picture of EM is therefore less one of economies weakened by the energy crisis and more one of players playing an increasingly central role in the major investment cycles linked to AI, electrification and infrastructure.
Contacts
8 Kievyan Street, Yerevan, Armenia
+374 10 712 259 +374 43 004 182
unibankinvest@unibank.am
Disclaimer
The information presented in the document contains a general overview of the products and services offered by Unibank OJSC (registered trademark – Unibank Invest, hereinafter referred to as the Bank).
The information is intended solely for the attention of the persons to whom it is addressed. Further dissemination of this information is allowed only with the prior consent of the Bank.
The information is only indicative, is not exhaustive and is provided solely for discussion purposes. The information should not be regarded as a public offer, request or invitation to purchase or sell any securities, financial instruments or services. The Bank reserves the right to make a final decision on the provision of these products and/or services to a specific customer, including refusing to provide products and/or services if such activities would be contrary to applicable law.
No guarantees in direct or indirect form, including those stipulated by law, are provided in connection with the specified information and materials. The information presented above cannot be considered as a recommendation for investing funds, as well as guarantees or promises of future profitability of investments.