Jim Taylor: What’s driving Aussie equities this week

Here are the main factors driving the ASX this week according to Pendal portfolio manager JIM TAYLOR. Reported by portfolio specialist Chris Adams

• Find out about Pendal Focus Australian Share fund

THE market is oscillating between de-escalation rhetoric from the US and escalatory bombing actions, such as Iranian nuclear and steel making facilities being hit on Friday night.

Last week, optimism that a fairly rapid end to the conflict would minimise economic damage from high oil prices and damaged confidence gave way to concerns about the likely negative impact from high energy prices.

President Donald Trump’s extension of the deadline on Thursday for opening the Strait of Hormuz could not hold back further equity selling on Friday.

The S&P500 finished down 2.1% for the week – its fifth straight week of losses and the worst weekly losing streak in about four years.

It has fallen an average of 0.3% per day since the war began, but the intraday swing from high to low has averaged 1.3%, adding to investor consternation.

The Nasdaq was -3.2% for the week, has now fallen for 10 of the last 11 weeks, and is in correction territory.

The S&P/ASX 300 was up 1.0% for the week, helped by a rebound in resource stocks.

Equity market sentiment indicators took a leg down last week, after being fairly muted for the previous three weeks.

The situation in the Middle East remains binary – it could easily get a lot worse but equally could see an improvement in sentiment very quickly.

The issues arising from the conflict are exacerbated by inflation in many countries having either stalled above levels that central banks are comfortable with (e.g. US) or having been on an upward trajectory (e.g. Australia).

There has been a rapid reversal across many developed and emerging market economies from an expectation of rate cuts to rate hikes.

This means a key underpinning of the supportive rate cutting/cyclical upswing upside scenario has moved very quickly to become a significant headwind.

For example, expectations for US rates have moved from 50 basis points (bps) of cuts in CY26 to circa 8bps of hikes.

On the AI front, documents leaked from Anthropic talked about the company’s next model (codenamed “Capybara”) and its enhanced capabilities. The documents said that “compared to our previous best model, Claude Opus 4.6, Capybara gets dramatically higher scores on tests of software coding, academic reasoning, and cybersecurity, among others.”

Macro and policy Australia

February’s consumer price index (CPI) was modestly below consensus expectations, with headline flat month/month and +3.7% year/year (versus +3.8% expected) and trimmed mean +0.2% month/month and +3.3% year/year (versus +3.4% expected).

However, this is largely irrelevant given the shift in energy prices in March.

Consensus expectations for 1Q 2026 quarterly inflation have moved up to ~1%, from 0.90% previously. The 30% rise in fuel costs will drive up headline inflation in the coming months.

The Roy Morgan-ANZ weekly consumer sentiment has plunged back to Covid lows, accompanied by a surge in inflation expectations. The RBA will be sensitive to any risk of the latter becoming unanchored.

In addition, it is expected that any prolonged fuel price increases will manifest in weaker non-discretionary retail spending.

Goldman Sachs has estimated that a 10% increase in fuel would see a 1.1% hit to retail sales, while a 30% increase would translate to a 3.4% hit, all else being equal.

This has weighed on the domestic discretionary retailers, which are typically down 10-20% since the war began.

Macro and policy US

It was a quiet week for US data. Initial and continuous claims data suggest that labour markets are holding steady for now.

However higher rates have seen mortgage applications down ~10% in each of the last two weeks.

While markets have moved quickly to reprice the expected path of US rates, thus far it does not appear to be pricing a material hit to US growth. The longer the conflict persists, the more fragile this assertion will feel.

Macro and policy Europe

European Central Bank president Christine Lagarde demonstrated the bind that central banks find themselves in. 

“We willnot act before we have sufficient information about the size, persistence, and transmission of this shock,” she noted, “…but we will not be paralysed by hesitation:our commitment to achieving a 2% inflation target over the medium term is unconditional.”

“If the shock gives rise to a large though not-too-persistent overshoot of our target, some measured adjustment of policy could be warranted. To leave such an overshoot entirely unaddressed could pose a communication risk: the public may find it difficult to understand a reaction function that does not react…. Otherwise, self-reinforcing mechanisms would kick in, and the risk of de-anchoring would become acute.”

Energy update

The net impact on global oil flows from the Strait’s closure appears to have improved last week, as an additional 3 million barrels per day (bpd) were redirected through other pipelines.

However, the release from strategic petroleum reserves has been slower than expected.

Goldman Sachs estimates that of the ~20 million bpd of normal Hormuz flows, 6.9 million bpd have been offset elsewhere.

Notwithstanding the net energy surplus position of the US, there is little indication that production is going to ramp to help fill the global deficit.

A Dallas Fed survey of large US oil producers showed that at this stage ~70% are not expecting to change their production in 2026. Less than 10% expect to increase it significantly.

A shift in ownership of oil wells in the US towards larger firms has meant that US oil production is less sensitive to increases in price than has historically been the case.

Our resources analyst, Jack Gabb, was in Perth last week, focused on the diesel outlook for miners. His observations are:

• Despite emerging shortages among independent distributors, the major miners all appear relatively comfortable at this stage.
• For example, BHP has not seen any change in supplies (typically receiving three to four shipments into Port Hedland per week).
• RIO noted concern, but no alarm. The company has a month’s worth of storage capacity and line of sight on another month.
• South32 also has five to six weeks visibility but are a smaller user.
• Stocks outside of the diversified miners are lower, typically with just one to two weeks on site. However, most reported visibility on deliveries from the likes of Ampol and Viva well into May.
• That said, action plans are being drawn up in the event supply is disrupted. For example, by reducing stripping/mining rates, changing rosters and processing stockpiles.

US West Coast tech/AI trip

Jim Taylor spent a week in San Francisco and Silicon Valley meeting with software and AI-related companies including Nvidia.

His key question on supply dynamics was around the major constraints to further data centre (DC) and AI factory buildouts. Here, his observations were:

• The conversation about data centre supply constraints has shifted from GPU hardware to physical space.
• Access to power remains a key constraint with companies developing their own power sources to bridge often long timelines in accessing the grid.
• New microchip technology is increasing the tokens that can be produced per watt of electricity consumed by between two and 35 times. (A token is a unit of text that AI can read and produce).
• Development yields and re-leasing spreads on existing data centre capacity both remain supportive for continued development. The scale of the new builds means that co-financing of projects is becoming increasingly important.

The other side of the question is demand – how is the demand for tokens developing and is there much growth after model training peaks? Observations here were:

• Demand for tokens is driving the revenue of the major large language models (LLMs) such as GPT-4, Claude, and Gemini up by US$1-2 billion per week. Token consumption is up ~4x since January.
• As token cost comes down (and is expected to continue to fall) economically justifiable use cases for extreme heavy token consumption increase.
• Training of LLMs remains a significant and ongoing use case, however inference usage is very much at the infancy of its journey. (Inference can be thought of as the AI model using what it learned in training).

Nvidia

Nvidia is at the core of the AI system and a leader in terms of its direction. Taylor’s key observations here were:

It is going all-in on inference

The battleground from a chips perspective over training LLMs is not over, but Nvidia has made it very clear the company is focused on getting ready for the explosion in inference activity that is coming.

The measure of success in this area will be the efficiency of token production in DCs and AI factories. The key metric will be tokens per second per megawatt (TPS/MW). The DCs are constrained by power and so to create extra revenue need to produce more tokens per MW of power available. Each new generation of chips is producing more tokens at lower costs. Decreasing cost of tokens will be a feature of the industry.

The current generation of hardware delivers 2x the number of users per MW at low levels of interactivity and up to 10x at higher levels of interactivity, compared with the previous generation. Further optimisation of the hardware sees the multiplier at the extreme high end of demand increasing to circa 35x.

The efficiency of production of tokens is critical to opening up the highest-value/highest-cost use cases for deep research into new medicines, training robots and widespread adoption of driverless vehicles, among many other areas of study.

The efficiency of the new generation hardware drives volume of production of tokens per unit of power, which creates additional revenue opportunities to sell that volume at better margins.

Open-source models

Nvidia is driving token usage through the provision of open-sourced models across the spectrum of autonomous driving, robotics, biology and a number of other fields.

Open-source models encourage companies to engage in research in these fields by not requiring them to begin at ground zero. This drives demand for tokens and Nvidia hardware and software.

The company’s Omniverse simulation platform allows manufacturers to build physics-based digital twins of factories and run real-time simulations. This can materially reduce production planning cycles and support more autonomous decision making on the factory floor.

Neocloud Providers (NCPs)

Nvidia is supplying NCPs with access to the whole stack of hardware and software to create demand tension with the hyperscalers.

It also provides Nvidia with access to low-cost tokens for its own internal usage. Nvidia is taking equity stakes in these businesses. It offers the NCPs a very low risk entry into the AI factory world.

The rise of Physical AI

Given the scale of manufacturing, healthcare, transport, logistics within global GDP it is no surprise that AI in the physical world is expected to dwarf the scale of AI in the IT world. Think of physical AI as AI + sensors + actuators + real world feedback loops.

While it may not be attracting all the headlines like the LLMs, there is a great deal of activity happening below the radar. For example, Jeff Bezos is trying to raise an AI manufacturing fund of $100 billion which will buy up manufacturing businesses to integrate them with AI. He has also founded a start-up called Project Prometheus – his first CEO position since 2021 – to develop AI for engineering and manufacturing (cars, aerospace, electronics, and more) with $6.2 billion in reported funding as of November 2025.

Tesla has flagged that the training of Optimus (a humanoid robot) will require ~10x the compute of that required for Cybercab, providing some insight into the scale of compute that is going to be required for global physical AI development.

Tesla is saying that it is targeting 30c/mile in total costs for the Cybercabs it is just launching. This compares to costs of running the Model Y of about 72c/mile and the cost of catching an Uber of about $2.50/mile. Thus, Cybercabs may be one of the first use cases with compelling transparent economics that prove up the bona fides of AI.

Other, broad observations from the research trip:

Be careful extrapolating the pace of AI/Agent deployment at tech companies to the broader economy. The speed of adoption at Microsoft or Block is not representative of the broader enterprise or consumer adoption curve. Mid-market and core economy companies are two to three years behind. Native AI/Cloud-based organisations have inherent advantages in terms of data being centralised and easily accessed, which materially reduces lead times for software development. Most enterprises are still busily cleaning and centralising their data to allow interrogation.

First-party data, properly exploited with AI/machine learning tools, is the nirvana outcome. Block’s internal credit bureau and dynamic credit scoring is the exemplar: AI converts proprietary data into a durable competitive advantage where growth is a purposeful journey, not a random walk.

Irrespective of AI, the scope for efficiency and cultural dividends from companies undergoing genuine transformation needs to be front and centre. Expedia, Intel, and Bill.com all offer meaningful operational leverage opportunities that exist independently of AI adoption. The benefit to shareholders of a fresh perspective from new C-suite executives remains a key source of value accretion.

Sales organisations are easier to pivot than start from scratch. Companies with existing channel relationships, sales infrastructure, and customer trust in adjacent markets have a structural advantage in capturing the middle market opportunity over pure greenfield entrants, even those with better technology.

Dealing with large organisations and institutions (“enterprise”) can be incredibly difficult for start-ups. Sales lead times can be long, while the enterprise pain point that the start-up is trying to address and the enterprise IT function are different parts of the business. Enterprise IT is focused on and obsessed with security, privacy and governance, which stretches resources of the new entrant. Wholesale change of customer relationship management is unlikely. Start-ups typically get a sliver of a workflow vertical (e.g. client service) where identifiable deep expertise gets them in the door. They augment, not replace, existing systems. Enterprise sales don’t walk in the door; they need to be curated and nurtured.

Agent identification and security is critical – no-one is letting agents loose in their corporate networks. Security is the unresolved blocker. Zero-trust frameworks for agents are not yet fully developed. Each agent effectively needs a digital identity and a full set of credentials — and unlike a human sitting outside the system, an autonomous agent is inside the system making real decisions. This is actually harder to secure than traditional user access management. IT owns the security concern but is often not the buyer with the pain point. Up until now bots in e-commerce represent malevolence, from here they represent a transactor. Authorisation goes from proving you are human to proving you are a credentialled bot. Identification and credentialling of agents will be a key focus of enterprises and will be a choke point for their adoption.

Software engineering called to account. Claude code exposes the inefficiencies and poor output that was previously thought to be acceptable; the starting point for software efficiency was an unrecognised issue. Software engineers will be hired (and agreeing to employment) based not just on salaries but also token budgets and the rate of burn of the tokens will be the lead indicator of productivity.

The Great Migration. Companies such as HubSpot, Intuit, Bill.com etc are all increasing in size, with higher monetisation of more complex products. The opportunity here seems real, facilitated by the increased cadence of product enhancement/development.

Sovereign AI. Here is an increasing focus on local availability. Complete outsourcing to foreign parties of all DC/hardware ownership and capabilities in country is considered a sovereign risk that needs to be addressed.

Onsite power production for DCs is being positioned as a bridge to grid connection, but is likely to be a permanent feature. 

Central processing units (CPUs) are back in secular growth, as the graphics processing units (GPUs) critical to AI need a lot of CPUs around it. CPUs also have an energy efficiency dividend to come.

The funding environment has fundamentally changed. In 2021–2023, a former OpenAI or Google DeepMind employee could get a venture or private capital term sheet with no business proposition. Now they need to turn up with a product. Many businesses in the pipeline are in limbo with low likelihood of getting funded. Start-up founders face a binary decision. Those with some funding but not enough to compete with well-funded leaders face a choice: push on with reduced resources and clear milestones or shut down. The days of bridge rounds and extended runways on a hope are largely over.

Are OpenAI/Anthropic friends or foes for traditional business models? The encroachment by the LLMs is something that the market has been very focused on. There are differing views on the nature of the LLMs, with people generally considering Open AI the most commercially driven business (deploying capital, spawning businesses) to Anthropic/GDM who are more focused on solving the scientific complications on the way to artificial general intelligence (AGI). They are pursuing commercial ends to create funding to solve the problems. Expedia, Intuit, Xero and many others have agreed deals with the LLMs. The market is unsure as to the balance of power between the parties, with the LLMs bringing a new distribution channel (1-2% of volume currently) and the companies striving to safeguard their data and IP.

Markets

Gold

Gold has had a positive correlation with equities, rather than acting as a safe-haven offset like US government bonds since the start of the war.

It has dropped over 14% since the start of the month, in significant contrast to prior periods of equity stress.

It has been flagged that around 83 tons of ETF holdings remain loss-making even at $4,500/oz and are hence susceptible to further liquidation, with 85 tons already redeemed since the conflict began.

The put/call skew on the largest gold-backed ETF hit a six-year high last week.

Turkey’s central bank added to the pressure, selling and swapping around $8 billion of gold in the two weeks following the outbreak of war.

It did hold up better than the equity market at the end of last week, though whether we have seen the peak gold price for this cycle remains a key area of debate.

---

About Jim Taylor and Pendal Focus Australian Share Fund

Drawing on more than 25 years of experience investing in top-performing Australian companies and a background in accounting, Jim manages our Long/Short Fund and co-manages our Imputation Fund. He is a Chartered Accountant with membership of the Australian Institute of Chartered Accountants.

Pendal Focus Australian Share Fund is managed by Crispin Murray. The fund has beaten its benchmark in 14 years of its 18-year history (after fees), across a range of market conditions. 

Find out more about Pendal Focus Australian Share Fund here.

Pendal is an independent, global investment management business focused on delivering superior investment returns for our clients through active management.