Daily Briefing | 2026-10-20: Policy Support and the AI Compute Race Run in Parallel, Markets Still Driven by Energy and Political Risks

Automatically translated. Read the Chinese original.

Today's main theme

Today's main theme is not a single growth or risk trade, but the coexistence of policy support and financing constraints. China kept its Loan Prime Rate unchanged after the Q3 economic data release, but the NDRC said it would push for the implementation of existing policies and reserve incremental measures; this reflects a policy focus on both stabilizing growth and improving transmission efficiency. The AI industry continues to shift from model capability competition to a battle over infrastructure and developer ecosystems: Nvidia launched open models for enterprises and local devices, while Amazon aims to lower the barrier for migrating PyTorch applications to its homegrown Trainium chips. Meanwhile, Russia-Ukraine diplomacy and Middle East energy risks continue to weigh on risk appetite. The market next needs to assess whether AI investment can translate into productivity and profits, whether policy can boost domestic demand, and whether geopolitics will keep pushing up energy and financing costs.

Key News

1. China Keeps LPR Unchanged, Monetary Policy Places Greater Emphasis on Transmission Mechanism Reform

The People’s Bank of China announced on 20 October that the one-year Loan Prime Rate (LPR) remained at 3.00 per cent, whilst the LPR for terms of five years or more remained at 3.50 per cent, marking the 17th consecutive month without any adjustment. Reuters reported that the market had widely expected the rates to remain unchanged; Following better-than-expected economic growth in the third quarter, policymakers are not in a hurry to introduce a new round of interest rate cuts for the time being. Central Bank Governor Pan Gongsheng had previously stated that the central bank is examining ways to further clarify the role of policy rates, such as the 7-day reverse repo rate, and is pushing ahead with reforms to the loan pricing benchmark to improve the transmission of monetary policy.

This does not amount to a shift towards tighter monetary policy. Keeping interest rates unchanged can go hand in hand with liquidity management tools, structural support measures and reforms to the loan pricing mechanism; the real question is whether the fall in funding costs will be passed on to businesses and households and translate into demand for credit. Going forward, attention should be paid to loan demand, property financing and consumer confidence, rather than focusing solely on whether the LPR is cut.

Source:Reuters report (reprinted by Investing.com)

2. The National Development and Reform Commission has stated that it will ensure the implementation of existing policies and introduce further measures as appropriate.

On 20 October, China’s National Development and Reform Commission (NDRC) stated that it would promote the synergistic implementation of existing and new policies, whilst conducting preliminary research and preparing policy reserves in line with economic conditions. The press conference mentioned accelerating the preparation and commencement of major engineering projects, improving the efficiency of approval processes, expediting the disbursement of funds through new policy-based financial instruments, and introducing private capital into major projects such as nuclear power, hydropower and water conservancy. The NDRC also mentioned the implementation of recently introduced measures, such as re-lending quotas, interest subsidies for residential mortgages, and financial support for the development of the service sector.

This sends a signal to ‘expedite implementation and scale up measures where necessary’, rather than announcing a new stimulus package of a specific scale. The effectiveness of the policy will depend on when the funds are made available, whether projects can commence, and whether private capital is willing to follow suit. It is worth monitoring whether the specific scale, target groups and implementation timetable of any new policies will be announced in due course.

Source:China News Network (CCTV News app)

3. Amazon has launched TorchNeuron in an attempt to lower the barrier to migrating AI applications to its own chips

Reuters reported on 19 October that Amazon has launched the TorchNeuron software tool, which enables developers to run programmes written in PyTorch on the company’s proprietary Trainium AI chips with minimal code changes and via a single configuration adjustment. Amazon states that the tool will be made available free of charge and as open source. PyTorch is a widely used framework in AI development, whilst the NVIDIA CUDA ecosystem has long been one of the key reasons developers choose its GPUs.

The key to this competition lies not merely in the peak computing power of the chips, but in developer migration costs, software compatibility, performance optimisation and the pricing of cloud services. If TorchNeuron can make it easier to migrate existing workloads, Amazon will have the opportunity to increase the utilisation of Trainium and reduce the reliance on NVIDIA GPUs for some AI training and inference tasks. However, the actual impact will depend on the maturity of the tools, the range of supported models, performance, and whether developers are willing to migrate their production environments.

Source:Reuters report (reprinted by Moneycontrol)

4. NVIDIA launches an open model for enterprise agents and on-premises devices

TechCrunch reports that on 19 October, NVIDIA released new open-source weight-based models, including Nemotron 3.5 Super VL for enterprise agents and the lightweight Bolt series of models designed for on-premises devices. NVIDIA’s strategy is to enable enterprises to deploy and customise models on their own infrastructure, whilst promoting the use of AI agents on personal computers and workstations, thereby reducing reliance on cloud services for certain tasks.

Open models are expected to expand NVIDIA’s software and hardware ecosystem; however, performance comparisons published by manufacturers should still be assessed in conjunction with independent testing, specific tasks and deployment costs. For enterprise users, the true decision-making criteria are the total cost of completing a task, reliability, data governance and the maintenance burden—rather than merely the number of parameters or individual benchmark scores. It will be worth keeping an eye on model weights and licensing details, third-party evaluations, and whether enterprises will scale up pilot projects to production environments.

Source:TechCrunch

5. US-Ukraine relations are under strain, and shifts in Washington’s policy towards Russia are adding to the uncertainty surrounding the negotiations

The Associated Press reported on 19 October that US President Trump had publicly criticised Ukrainian President Zelenskyy and stated that Russia’s conduct regarding Ukraine was “not bad”; the report also noted that the US had recently adjusted some of its sanctions targeting the Russian energy sector. These remarks have deepened external concerns regarding US-Ukraine relations and the direction of US policy towards Russia. Meanwhile, Ukraine continues to face attacks by Russian forces on its energy infrastructure, with pressures on the battlefield intertwined with diplomatic negotiations.

This does not mean that a breakthrough has been achieved in the negotiations. For Ukraine, security guarantees, arms supplies and intelligence cooperation remain key issues; for Europe, whether US policy continues to shift will influence defence spending and the assessment of energy risks. Attention should now focus on the talks between the US and Ukraine and with Europe, the specific scope of sanctions exemptions, and whether any ceasefire proposals include enforceable monitoring and security arrangements.

Source:Associated Press

Market Correlation and Today's Observation

• Chinese Assets: The fact that the LPR has remained unchanged does not in itself constitute a new round of monetary easing; the market needs to focus more on policy implementation, credit demand and domestic demand data. If project funding is disbursed more quickly, the infrastructure sector and certain industrial goods may benefit; if private investment and consumption remain weak, policy expectations may not immediately translate into upward revisions to earnings forecasts.
• AI and Semiconductors: Amazon’s software tools and NVIDIA’s OpenModel are both vying for developers and deployment scenarios. In the short term, the market will continue to focus on the financial results, capital expenditure and order books of AI-related companies; in the medium term, however, the outcome should be assessed by comparing actual inference costs, software ecosystem stickiness and the speed of customer migration, rather than judging the winner solely on the basis of statements made at product launches.
• Energy and Risk Appetite: The situation in Ukraine and supply risks in the Middle East are likely to continue to drive a geopolitical premium into oil prices. Should tensions escalate, energy costs and inflation expectations may limit the scope for central banks to cut interest rates; should credible progress be made on the diplomatic front, the risk premium may then recede.
• Today’s checklist: The details of the implementation of China’s new policies; independent performance data for Amazon and Nvidia products; the official scope of US energy measures against Russia; and the reaction of oil prices to geopolitical developments.

Cover image:imgix / Unsplash