A volatile week in the AI trade kicked off with an Anthropic S-1 that surprised many with positive takeaways.
Funda has a detailed and structured 58 point tweet summarising the S-1.
Positive feedback mainly converged on two points:
Anthropic has a good reserve of cash and has been surprisingly capital efficient in getting its business to where it is today.
Anthropic’s broad Fortune 500 base of customers spend a relatively small percentage of their expenses on AI. This can either be viewed positively (runway) or negatively (airpocket / bottleneck), with bears pointing to Funda’s observation about forward deployed engineers being a potential bottleneck.
Advantest, SK Hynix and Sandisk posted good to great earnings, but ultimately Marvell’s bid for Lumentum and Intel put a stop to any good vibes returning to the market.
Kioxia later today will have to demonstrate sufficient appetite for price hikes to avoid a miss, and probably a continued appetite for buying back shares. We’d expect management to continue presenting a strong outlook, especially since Sandisk said its New Business Model (LTA) pipeline exceeds its supply through the end of the decade.
It has now been six months since I started working with Funda. I have greatly benefitted from my conversations with talented colleagues, whose numbers continue to grow, allowing us to make sense of more of the market.
It has not escaped your notice that my confidence in reading the market has grown since June, something I have to attribute entirely to the data and insights from Funda. I have to say I did not expect this to happen, as I initially viewed the main Funda advantage as the flow of channel checks. I think these capabilities will sharpen in the coming weeks.
As independent analysts with a trusted voice in the market, we also have a larger degree of access to companies who recognise the value in engaging with us. We will be trying to deliver on that promise when we meet the amazing line-up at the Tomorrow X conference in two weeks. You can purchase a pass for 20% off here. Please also reach out by replying to this email if you would like to meet in Austin between 16 to 18 November.
Finally, in the same vein, we spoke with Aolani this week, a Singapore-founded neocloud and NVIDIA Cloud Partner building high-performance AI infrastructure across Asia.
Aolani reached out to me saying they were seeing US neolabs, inference platforms and physical AI customers seeking compute in Southeast Asia for the first time. They were kind enough to share their views on the recent market volatility.
Could you share a little about financing progress and plans for data centers in Southeast Asia?
Despite recent market volatility, we continue to be bullish on the long-term fundamentals for compute demand in Southeast Asia. Global compute demand will only continue to trend upwards, and we are seeing a lot of those tailwinds shifting to Southeast Asia. The region benefits from favourable supply-side trends – Europe and US are highly challenged to add new DC capacity given public scrunity and structural infrastructure bottlenecks, while single buyer frameworks in Southeast Asian countries like Thailand, Malaysia, Indonesia allow much faster scaling of power capacity.
We’ve seen only increasing demand for SEA compute this year, with many US customers (neolabs, inference platforms, physical AI) seeking compute in the region for the first time. DC capacity in the region is pretty much booked out until 2H 2027. Beyond our capacity already operational or announced in Malaysia and the Philippines, we are also working on further capacity in Malaysia, Thailand, India, and Europe.
AI infrastructure is set to be the largest financing opportunity in Southeast Asia for the next 5-10 years at least. We’ve also seen the Asian lending market for GPUs come a remarkable way from when we first began speaking to banks in 2025, and we think there’s still headroom for the GPU financing market in Asia to grow – insurers, sovereign wealth or provident funds, bond markets, even futures markets to come.
Against this backdrop, a key strategy of ours is opening up new financing solutions and sources of capital while ensuring diversification of capital sources and disciplined capital efficiency. The transition to asset-backed financing for example, is one to watch as financiers recognize GPU residual value over time. We currently work with a variety of global and Asian institutional investors and lenders for our equity and debt needs.
Can you tell us a little bit about some of the concerns you are hearing from data center / neocloud investors in APAC in recent weeks? What about pushback or mitigants to those concerns?
The investors that we engage with generally recognise the long-term fundamentals and understand that in a high-growth market, sentiment and valuations can be highly sensitive to short-term volatility.
From our vantage point, the recent market pullback has prompted greater scrutiny of risk exposure rather than a fundamental reassessment of the sector’s drivers. There are two primary risks that we see being top of mind for investors in APAC: 1. Financing and liquidity risks, given likely rate rises; 2. Geopolitical risks.
On 1, neoclouds vary quite widely in how much financing risk they expose themselves to, with factors such as:
Capital stack – Balance of parent-level unsecured debt vs. asset-level DDTLs secured against GPUs and/or customer contracts, and quality of the DDTLs themselves (non-recourse, investment-grade offtake, etc)
Cost of financing – many neoclouds rely on private credit vs. cheaper financing such as bank or bonds
DC contracts – whether they contract or build DCs ahead of demand, how well-matched their DC tenors are to their offtake contracts
On 2, the Remote Access Security Act is a bill that could potentially control the remote access of NVIDIA GPUs by customers deemed to pose a risk to national security or foreign policy. It has been passed by the House of Representatives in the US, and stalled in the Senate. It expires in Jan 2027, but experts do not believe that it will be passed in the next 18 months.
In this environment, robust risk management remains critical. For us this means maintaining capital-efficient financing structures that can endure through market cycles, diversifying sources of capital and proactively monitoring regulatory developments to ensure we are resilient across a range of market scenarios.
How do lenders currently view financing for neocloud companies? Are they increasingly favoring projects backed by long-term contracts with major CSPs, while imposing more stringent financing terms on standalone neocloud projects, such as higher interest rates, lower loan-to-value ratios, and stricter covenants?
Aolani primarily works with institutional bank lenders in Asia. In our experience, institutional banks generally do not finance neoclouds solely on on their standalone credit. The market norm has always been contract-backed financing with an investment-grade offtaker, and the credibility of offtakers has always been the key focus area of banks’ due diligence process. Today, lenders are also usually doubly protected on their GPU financing, by the underlying offtake contract and the GPU asset as collateral.
Exceptions are made when a parental guarantee is involved, or when the loan is purely asset-backed (small minority of lenders today, mainly private credit). We believe over time that the market for asset-backed lending will grow, as lenders recognize the durability and fungibility of GPUs, demonstrated by the rising value of older GPU generations like H100.
More broadly, we have not observed shifts in terms of lending behaviour. Lenders have always based their terms on detailed due diligence on project quality. While equity investors are more susceptible to market volatility, lenders tend to remain focused on the fundamental credit characteristics of each project.
Are the attitudes of data center developers and operators, particularly powered-shell providers, towards neocloud customers changing? Are they more inclined to allocate available capacity to major CSPs rather than neoclouds? Have you seen any cases where data center developers or operators have delayed, scaled back, or even canceled projects or capacity commitments originally intended for neoclouds?
Virtually all data centres in Asia from the start have been geared towards long-term CSP offtake rather than neoclouds or AI offtake. A key challenge has been the structural mismatch between typical data centre tenors (10-15 years) and AI customer offtaker contracts (3-5 years), which has long hampered DC capacity available to neoclouds.
We haven’t seen a significant negative shift away from neoclouds. If anything, the trend is opposite – DC capacity allocated to neoclouds is growing, as operators recognise that this is a fast-growing segment with better economics.
Where cancellations do occur, they usually originate from end offtakers rather than DC developers or operators. We’ve seen instances of hyperscalers cancelling planned DC builds due to perceived execution or macroeconomic risk. End users may also walk away from contracts if they perceive that delivery timelines can’t be met. Most projects exposed to this risk are large-scale campus type developments (100s of MWs and above), particularly when land, power, and water permits haven’t yet been secured.
Our approach mitigates this, where we deploy in more flexible and modular phases (20-30MW) and only contract after the build is fully derisked with all permits secured.