AZ Sestante Chief Investment Officer Andrea Ciaccio reviews markets for September 2026 — from Anthropic’s call to “pace the frontier” and its pursuit of a $2 trillion IPO, to a narrow, tech-led US equity market, a round of global rate hikes and the RBA lifting the cash rate to 4.60%. He argues that AI has become the most consequential risk for investors’ portfolios, and explains why a 5% Treasury yield now looks attractive. Read the full commentary below, or download it as a PDF.
Download the full commentary (PDF)
By Andrea Ciaccio, Chief Investment Officer, AZ Sestante — as at 30 September 2026
Market Review
In Frank Herbert’s “Dune”, humanity emerges from a devastating war against artificial intelligence with a defining commandment: “Thou shalt not make a machine in the likeness of a human mind”. Decades after its publication, the seminal novel’s central themes are no longer confined to science fiction but are increasingly playing out in the real world. Or at least, that is the unsettling conclusion suggested by Jacob Coxon, a pre-training researcher who left Anthropic, warning that “there is a strong chance that we could all die in the immediate future” if AI models continue to grow more powerful at the current “reckless” pace. His words were echoed by Evan Hubinger, Alignment Science Lead at the same tech startup, who believes there is a greater-than-10% probability of human extinction within the next decade caused by a “superintelligence arising from recursive self-improvement”. As former and current employees sounded the alarm, Dario Amodei, co-founder and CEO of Anthropic, entered the debate on September 12th with an essay titled “We must pace the frontier”. In it, he proposed a three-pronged approach consisting of independent monitoring of AI models during development, industry-wide regulation among democratic countries and, ultimately, international cooperation to establish global safeguards, including through engagement with China.
Amodei’s central argument is that the international community must deliberately slow the frontier globally. Without such moderation, competitive pressure, especially between the US and China, could push companies and governments into a “race to the bottom”, prioritising speed and technological edge over safety. He is not advocating an outright halt to AI progress. Rather, he cautions that the accelerating capabilities of advanced AI could outstrip humanity’s ability to understand, govern and control it. Yet his thesis exposes a fundamental geopolitical dilemma. “Pacing” unilaterally could in fact allow China to close the competitive gap or gain a strategic advantage, potentially undermining the very security that restraint is intended to protect. He therefore maintains that the US and its democratic allies must preserve their technological lead while establishing conditions for responsible, coordinated progress. To this end, he supports restricting China’s access to the most advanced AI chips and semiconductor-manufacturing equipment. In essence, he is simultaneously arguing for slowing the AI race and for ensuring that America maintains its lead over China in the “Cold War 2.0”. AI titans Elon Musk and Sam Altman took to X to express support for Amodei’s concerns, with the former stating that “Dario is right” and the latter endorsing the proposal to provide independent evaluators with employee-like access, calling it “a great idea”.
During September, the AI safety question moved into the political sphere, compounding the backlash against the expansion of AI infrastructure. On the 21st, Texas Governor Greg Abbott ordered state environmental regulators to stop issuing permits for data centres pending audits of their effects on electricity-grid reliability and water resources. The decision represented a U-turn given Abbott’s previous effort to position the Lone Star State as a major destination for AI investment. It also demonstrated that anxiety over escalating utility costs was gaining traction among Republicans and was not exclusive to progressive politicians. Two days later, Senator Bernie Sanders and Representative Greg Casar introduced the “Ban Artificial Superintelligence Act”, arguing that voluntary safety commitments were insufficient to address potentially existential risks. On September 29th, President Trump issued an executive order titled “Inaugurating the Era of Super Intelligence”, directing federal agencies to use “Super Intelligence” instead of “Artificial Intelligence” in official communications. The President justified the rebranding by arguing that “artificial” implied something “fake”, whereas modern AI had evolved beyond merely imitating or automating human intelligence. His reasoning was that the technology had become more capable than its original name suggested and could amplify human ingenuity, enable scientific breakthroughs and achieve outcomes previously considered impossible.
Anthropic’s IPO preparations served as a reminder of the enormous financial stakes involved in the AI race. On September 28th, Reuters reported that the lab had generated approximately $4.6 billion in revenue in 2025, a twelvefold increase from the previous year, yet suffered an operating loss of $8.06 billion, up substantially from $2.98 billion in 2024. The company spent $7.33 billion on computing and infrastructure in 2025 alone, three times its 2024 expenditure and more than half of its $12.65 billion in total operating expenses, and unveiled plans for $518 billion in future capex commitments. Its financial risks extend beyond overheads, as two unnamed customers accounted for approximately 24% of its 2025 revenue, while some major clients were not bound by long-term spending commitments. Nevertheless, Anthropic was targeting a valuation exceeding $2 trillion for an IPO after the November US midterm elections, more than double its estimated valuation of $965 billion in May 2026. Meanwhile, having ruled out an IPO in 2026, OpenAI sought to raise at least $30 billion from private investors at a targeted valuation of approximately $1.4 trillion, following a March funding round that secured $122 billion in commitments and valued the company at $852 billion.
International Equities
US equities were mixed in September. The Nasdaq 100 (+3.23% in USD terms) broke to new all-time highs, while the S&P 500 (-0.45%) remained within 2% of its peak recorded on August 13th, modestly retreating during the final trading week of the month. However, this slight decline masked a significantly weaker underlying market, with approximately 78% of its constituents falling during the month and 60% dropping by at least 5%. As a result, the S&P 500 Equal Weighted Index plunged -4.99%, logging its fifth-worst monthly relative performance ever vis-à-vis its market-cap weighted counterpart. Gains in major technology companies offset widespread weakness elsewhere, leaving the headline index less representative of the average constituent. Moreover, market concentration increased further, with the 10 largest companies making up approximately 39.2% of the S&P 500. Not surprisingly, the “old economy”-heavy Dow Jones Industrial Average (-4.29%) fell for most of September, while the Russell 2000 plunged -5.40%, underperforming the Russell 1000 by the most since December 2024. Following its stabilisation in August, the Philadelphia Stock Exchange Semiconductor Index jumped by close to double digits on renewed optimism about demand for computing hardware. On September 21st, AMD surpassed $1 trillion in market capitalisation for the first time, becoming the fourth American chipmaker to join the club after Nvidia, Broadcom and Micron. The “Magnificent Seven” group of stocks took off, besting the rest of US large caps by more than 600 Bps and once again moving in tandem with semiconductor stocks. This reversed the negative correlation observed over the preceding months, when investors had rotated away from the companies financing AI infrastructure (the spenders) towards those supplying the hardware underpinning it (the earners). Meta led gains, adding nearly $400 billion to its market capitalisation following the launch of Muse, a personal AI agent capable of performing tasks such as sending emails, booking travel, filling out forms and making payments. The system can interact with other applications on users’ behalf, although Meta acknowledged internal concerns about the handling of sensitive personal data. Initial consumer adoption was strong. According to Sensor Tower data, Muse amassed approximately 2.8 million downloads in its first two weeks across the US and Canada.
Technology was the only sector to post gains in September, while communication services was broadly unchanged. The remaining groups declined, with materials, financials, consumer discretionary, real estate, utilities, consumer staples and industrials falling between 4% and 7%. As defensives, high-dividend stocks and cyclicals were equally hit, the S&P 500 High Beta trounced the S&P 500 Low Volatility TR by more than 8%.
US equities outperformed the rest of the world (as exemplified by the MSCI AC World ex USA TR Index) by the most since March, although they continued to lag on a YTD basis. Japan was the best-performing market, buoyed by a strong rebound in the value of the Japanese yen (JPY) and a recovery in AI-related stocks and semiconductors. The same sectors lifted emerging-market heavyweights South Korea and Taiwan, propelling the returns of the Asia ex-Japan region above those of the US, despite steep declines in China and India. Conversely, Europe bucked the trend as weakness in autos, financials and basic resources weighed on the broader market. France’s political and budgetary problems amplified the negative sentiment, with technology and energy the only sectors to deliver positive returns. All in all, the MSCI AC World Daily TR was down 1.11% in USD terms and up 1.97% in AUD terms.
Australian Equities
Australian equities dropped in September, interrupting a five-month positive run, and the MSCI Australia Net TR lagged the MSCI AC World Daily TR by nearly 400 Bps. The S&P/ASX 300 shed -2.43% as heavyweight sectors financials and resources fell. Within the former, banks and A-REITs were in the red, while insurance recorded solid gains. Within the latter, diversified and gold miners tracked the price of industrial and precious metals lower, with iron ore, gold and silver tumbling -4.68%, -6.22% and -9.08% respectively. Conversely, energy rose as West Texas Intermediate (WTI) crude surged +5.43% in September, supported by renewed Chinese crude stockpiling and tight refining capacity amid supply disruptions in the Middle East and in Eastern Europe. Technology was hammered as investors sold off software names such as Xero and WiseTech, which Goldman Sachs has grouped within its “consumer inertia” theme. The basket comprises companies that benefit from entrenched customer habits, low engagement and high switching costs, advantages that may be eroded by the rise of autonomous AI agents. Healthcare posted robust gains as CSL continued to rise from oversold levels following its FY26 results and Ramsay Health Care extended its rally. Consumer staples curbed losses, while consumer discretionary, utilities and telecom trailed the general index. Finally, the Top 20 trounced mid-caps and smaller companies as risk aversion intensified.
International Fixed Income
Global central banks hiked by a cumulative 125 Bps during the month, with each of the five moves amounting to 25 Bps. The Reserve Bank of New Zealand (RBNZ) kicked things off on September 2nd, bringing its Official Cash Rate (OCR) to 2.75%. The European Central Bank (ECB) followed on the 10th, increasing its policy rate to 2.25%. On September 16th, the FED raised its target rate to a range of 3.75%-4.00%, this time with the FOMC voting unanimously in favour of the decision. During the press conference, Chair Kevin Warsh stated that the committee “removed a dose of accommodation so that financial and credit conditions would be more consistent with our ultimate objectives”. The Dot Plot projected one additional hike in 2026, followed by a pause in 2027. The Summary of Economic Projections (SEP) showed upward revisions to the GDP growth and inflation forecasts, from 2.2% to 2.3% and from 3.6% to 3.7% respectively, while the unemployment estimate was revised down from 4.3% to 4.1%. The Bank of Japan (BOJ) followed two days later, lifting its benchmark rate to 1.25%, while the Norges Bank, Norway’s central bank, raised its policy rate to 4.50% on the 24th.
The other four major central banks that met during the month left interest rates unchanged. These were the Bank of Canada (BOC), the Bank of England (BOE), the Riksbank (Sweden’s central bank) and the Swiss National Bank (SNB), which maintained their policy rates at 2.25%, 3.75%, 1.75% and 0% respectively.
Global yields rose sharply in September as investors priced in more tightening. In the US, the differential between the 3-month and the 10-year yields reached its highest level since June 2022, implying that the FED may be falling behind the curve. The ICE US Treasury 20+ Year Bond Index, a popular benchmark to track the performance of long-duration strategies, dropped the most since December 2024, down -5.16% in USD terms. As a result, the so-called “term premium”, which is the compensation required to buy and hold long-dated Treasuries as opposed to simply rolling over short-term paper at prevailing rates, pushed to new highs for this cycle at 89 Bps. In Japan, the 10-year yield broke above 3% for the first time in more than 30 years, while in Europe the 10-year OAT/Bund spread, that is, the difference between the yield paid by the French government and the yield offered by the German government, reached 127 Bps, its highest level since April 2012. Credit experienced heightened volatility and spread widening, particularly in lower-rated segments. The Bloomberg Barclays Global Aggregate Index hedged back to AUD fell -1.66% for the month.
Australian Fixed Income
On September 29th, the RBA raised the cash rate by 25 Bps to 4.60%, its highest level since November 2011. During her press conference, Governor Bullock adopted a hawkish tone and cautioned that further tightening may be required if inflation persists. She stressed that the domestic inflation problem predates the Iran conflict and remains primarily driven by excess demand and weak productivity growth, while higher energy prices stemming from the conflict are “in addition” to that. Bullock argued that the cost of inaction could ultimately be greater than the near-term pain of further tightening, as the RBA may be forced to raise rates more aggressively and engineer a much sharper economic slowdown to restore price stability at a later stage. For this reason, the Board advocates for growth in aggregate demand to remain “subdued for a period”. The Bloomberg AusBond Composite 0+ Yr returned -0.86% as the yield curve shifted higher, with the 2-, 5- and 10-year yields adding 27, 26 and 26 Bps to 4.96%, 4.97% and 5.35% respectively. At the end of the month, cash rate futures were pricing in one further hike, with a 100% probability of the move occurring within the next six months. The Australian dollar softened across the board, ending the month below 70 cents against the greenback.
Real Assets
Global property plummeted -5.75% in USD terms and -2.81% in AUD terms for the month, with all regions generating negative returns. Asia curbed losses while Europe was the worst performing market on the back of generally higher financial leverage. In Australia, the 10-year real yield climbed to 2.95%, its highest level since June 2008.
Global infrastructure slid -5.57% in USD terms and -2.62% in AUD terms in September. Transportation stocks were the most resilient, supported by airports, while communication infrastructure and electric utilities declined the most.
Alternatives
Returns for Alternatives (+0.60%) were mixed across strategic mandates in September. Trend following and discretionary macro capitalised on falling bonds and rising energy prices, while interest-rate-sensitive strategies such as relative value and event-driven declined.
Market Outlook
Anthropic is asking investors to finance a company building increasingly powerful AI while warning those same investors that its technology could pose “catastrophic or existential risks to humanity”. To reconcile this apparent contradiction, its CEO is effectively calling for government intervention, a plea that some critics view as an attempt to shield American AI labs from the competition posed by Chinese “open weight” models. While we are normally fond of nuance, in this case we think that these two positions are fundamentally incompatible. Either Anthropic is worth $2 trillion, or whatever amount public markets are willing to pay for the privilege of owning a stake in it, or AI is a weapon of mass destruction. In fact, if the latter is true, we would argue that, in his otherwise elegant paper, Amodei has overlooked two of the most plausible responses the US administration might consider: nationalising the firm or forcing it to sell only watered-down versions of its models to the public while reserving access to Artificial General Intelligence (AGI), if and when it is achieved, exclusively for government use. Such measures would have far-reaching implications for the startup’s commercial prospects and valuation. Hence, until proven otherwise, our base case remains that Anthropic’s unprecedented disclosure in its IPO prospectus is more of a (bad) marketing gimmick than a genuine threat and should be treated accordingly.
This is not to say that AI is riskless. On the contrary, AI represents the most consequential danger for investors, with the potential to “kill” their portfolios if the investment theme has peaked. In fact, AI has effectively become the US economy. There are two reasons for this. First, the scale of investment in data centres is enormous. The construction of these facilities requires vast amounts of materials, equipment and infrastructure, while generating significant economic activity and employment. Some estimates suggest that AI-related investment may be contributing as much as one percentage point to US GDP growth each quarter. Second, there is the so-called “wealth effect”. Stock markets are at all-time highs, sustained by semiconductor stocks and other AI beneficiaries. As their portfolios have appreciated, the “haves” have become richer and have continued to spend with abandon. An astute analyst has also made the case that some of these equity gains may have spilled over into the property market, helping to keep valuations elevated despite deteriorating housing affordability and high borrowing costs. As rising equity prices have increased the wealth of affluent investors, cash-rich buyers have been able to purchase properties outright, displacing aspiring homeowners who can no longer afford mortgage financing. If AI investment cools off, the economy could be affected through two channels. Companies could reduce capital expenditure, weakening a significant source of growth. In addition, if AI-related equities correct, the resulting loss of household wealth could weigh on consumption.
In other words, a slowdown in AI could quickly morph into a broader economic slowdown. This is what the market may have started to price in since the second week of October, when we saw, for the first time in a long time, a clear deflationary dynamic emerge. The correlation between equities and bonds turned abruptly negative, with technology stocks selling off hard and yields declining in a straight line. This happened despite horrible inflation data, a FED that appears to have lost any sense of urgency about achieving its 2% target (and is widely expected to skip an October hike), and the oil price hovering above $100 per barrel. To top it off, precious metals have also been crashing, which may point to tighter financial conditions and a sudden scramble for liquidity. These moves suggest that the market may have begun shifting its focus away from inflation and towards the risk of a slowdown in economic growth. More specifically, investors may be starting to question whether the AI investment boom can continue at its current pace given the enormous financing requirements involved. A case in point is CoreWeave, a so-called “neocloud” provider that acquires chips, secures data centre capacity and rents GPU computing power to AI developers and hyperscalers. Rather than building all their computing infrastructure themselves, companies such as Microsoft, Meta and other major AI developers can contract with providers such as CoreWeave to obtain the capacity they need.
It is a great business model in that its customers are the largest and most strategically important firms in the world. There is only one problem: a neocloud provider must take on debt and commit to long-term contractual obligations to invest in expensive chips and infrastructure years before it can generate the corresponding rental income. Although a substantial order backlog provides visibility into potential future revenue, it does not eliminate the risks associated with execution, financing costs or the possibility that customer demand and economics change before the capacity comes online. These risks have been apparent for some time, but what changed in October was the market’s willingness to finance them at the prices investors had previously accepted. CoreWeave’s bonds came under significant pressure, with some issues falling below 80 cents on the dollar and yields rising to 15%. This does not mean that the company is currently paying such a high yield on its debt, but that it will have to when it taps the market again. And as long as conditions remain what they are today, we do not think that it will do so. If we are correct in our assessment, that means one of the major assumptions underpinning the AI trade may be starting to break down. Until recently, the prevailing narrative was that companies would be willing to pay any rate to secure the necessary funding in the “AI arms race” as nobody could afford to be left behind. As it turns out, financing costs still matter, even in an industry promising technological transformation.
If companies such as CoreWeave cannot raise capital on acceptable terms, some projects may become uneconomic or need to be delayed. This does not mean that AI capital expenditure is about to contract sharply, let alone that the investment boom is over. But it suggests that the pace of expansion may become increasingly dependent on credit conditions and demonstrable returns. On the latter point, there may also be a growing sense of fatigue with bombastic “annual recurring revenue” (ARR) claims when it is not clear what underlying revenue base is being annualised, or assertions that AI labs are already profitable “on an adjusted basis”. Markets may be reaching the point where they want to see the actual books of these companies and have them subject to the same public reporting and disclosure requirements as listed companies, even if they remain private. Finally, the US midterm elections may add a further layer of uncertainty to the AI investment outlook. Opposition to data centre construction has become an increasingly visible political issue, and Democrats have sought to capitalise on this discontent, challenging an administration that has strongly advocated AI investment and the need for the US to maintain its technological lead. With prediction markets currently pricing the probability of a “Blue sweep” at nearly 70%, investors sitting on substantial equity gains may be tempted to sell first and ask questions later.
Against this backdrop, a 5% Treasury yield looks attractive, particularly if, as we believe, the worst of the upward pressure on interest rates is behind us. We are not suggesting that yields are about to fall sharply, or that inflation is no longer a concern. However, in our opinion, the risk of a further substantial repricing higher in rates, accompanied by renewed investor anxiety, has diminished. And if concerns about AI investment spill over into a broader reassessment of growth, government bonds could also benefit from a flight to quality, providing a potential portfolio diversifier on top of a source of income. In equities, diversification remains the name of the game. The volatility in AI-related stocks has increased and a meaningful rotation is taking place beneath the surface of the headline indices. We therefore see no compelling rationale to make a major change in positioning at this stage. Nor would we interpret the recent turmoil as a reason to sound the alarm on equities or to conclude that the bull market has peaked. The evidence is not yet sufficient to make that call. Finally, it is normally assumed that the US dollar will strengthen during periods of market stress because of its reserve currency status and the depth of US financial markets. But positioning matters, and international investors have accumulated an outsized exposure to US equities. If a deterioration in the US growth outlook were to prompt some of them to reduce that exposure, the greenback might weaken rather than strengthen. In that scenario, the Australian dollar could prove more resilient than many investors expect.
Download this commentary as a PDF
AZ Sestante is a specialist investment consultant focused on designing and managing a range of multi-manager model portfolios via SMAs, MDAs, and fund of funds. Our parent company Azimut is Italy’s largest independent asset manager listed on the Italian stock exchange. E: [email protected] | www.azsestante.com
Important information
The Morningstar Historical Corporate Sustainability Score is a weighted average of the trailing 12 months of Morningstar Portfolio Corporate Sustainability Scores. Historical portfolio scores are not equal-weighted; rather, more-recent portfolios are weighted more heavily than older portfolios. Combining the trailing 12 months of portfolio scores adds consistency while still reflecting portfolio managers' current decisions by weighting the most recent portfolio scores more heavily. ESG pillar scores are displayed as a number between 0 and 100 with most scores range between 0 and 25. It is the asset-weighted average of the company environmental, social, governance risk scores for the covered corporate holdings in a portfolio. The scores measure the degree to which a company's economic value may be at risk driven by environmental, social, and governance factors. The risk represents the unmanaged risk exposure after taking into account a company's management of such risks.
This document has been prepared by AZ Sestante Limited, ABN 94 106 888 662, AFSL 284 442 (AZ Sestante). This document is not an offer of securities or financial products, nor is it financial product advice. As this document has been prepared without taking account of any investors' particular objectives, financial situation or needs, you should consider its appropriateness having regard to your objectives, financial situation and needs before taking any action. Past performance is not a reliable indicator of future results. Although specific information has been prepared from sources believed to be reliable, we offer no guarantees as to its accuracy or completeness. The information stated, opinions expressed, and estimates given constitute best judgement at the time of publication and are subject to change without notice. Consequently, although this document is provided in good faith, it is not intended to create any legal liability on the part of any other entity and does not vary the terms of a relevant disclosure statement. All dollars are Australian unless otherwise specified.