Executive Summary

We maintain a Neutral to cautiously Bullish outlook for the Global Technology sector over the next 6-12 months. The AI buildout remains the strongest capital cycle in tech history and the four largest hyperscalers are guiding to approximately 700-730 billion dollars of combined 20 26 capex, which represents growth of over 75 percent from 20 25. This capex is not speculative, but rather already contracted and spending is trickling through to chipmakers, equipment suppliers, and power infrastructure with >3y visibility. However, a handful of AI-exposed names have seen their valuations sharply re-rate, chipmakers are increasingly circular in their financing structures with their largest customers, and index concentration has reached generational highs within a handful of mega-caps. We believe the strength of the fundamental demand story is real and likely underappreciated in terms of duration, however risk-reward in the sector has become more binary and favors picking infrastructure and picks- and-shovels names over broad index exposure.

Macro Backdrop

Driving these valuations is the hyperscaler capex boom we’ve covered extensively throughout the year. Spending continues to exceed even unserious expectations. Amazon increased 20TC guidance to approximately 220 billion dollars, up from a prior plan of 200 billion dollars. Alphabet increased its guidance range to 195-205 billion dollars, more than doubling 2023 spend and coinciding with Google Cloud’s contracted backlog now topping ~460 billion dollars, approximately double last year’s level. Microsoft is pacing towards 190 billion dollars in calendar 2026 while Meta has updated their range to 130-145 billion dollars. All together the big four are pacing towards ~725-730 billion dollars this year and data from Goldman Sachs highlighted in early September suggests that the combined deployments from these four companies reached ~432 billion dollars during the second half of the year alone, firmly putting a 1 trillion dollar annual run rate into reach for 2027.

This money is colliding with a chip and packaging supply chain still recovering from the pandemic binge. Taiwan Semi’s CoWoS advanced packaging node, required for AI accelerators such as Nvidia’s Blackwell and Rubin families, still has a booking backlog out to 52-78 weeks as of August despite near doubling capacity throughout 2026 to reach between 120k and 150k wafers per month. Leading-edge nodes demonstrate a similar bottleneck. As of August TSMC’s 2nm process node has bookings out to 2028 and lead times of 78-156 weeks. The company has also stopped taking new customer startups for their 3nm node due to capacity constraints. In comparison, legacy nodes such as 7nm and above show lead times between 4-17 weeks and indicate significant open capacity. These internals paint a picture of a two-tier semiconductor market. Leading-edge AI silicon remains in production purgatory while mature-node fabs that serve autos, industrials, and commoditized consumer electronics remain flush with capacity.

China tensions continue to be a major bullish factor for share prices. Restrictions on AI chip exports to China have been ratcheted up several times already in 2026 by the Biden administration, first in January 2026 with a rule that put in place requirements for foreign-made advanced semiconductors to be routed through the US for security testing and imposition of a 25% tariff prior to re-export, and again in August 2026 with the Bureau of Industry and Security’s move to close a loophole that allowed Chinese companies to circumvent GPU compute limits by renting chips through Asian data centers in Thailand, Singapore, and elsewhere. Congressional efforts to further restrict China’s access to advanced semiconductrics continue to move through reconciliation as well, with the Chip Security Act and the MATCH Act both aiming to tighten smuggling punishments and bring US ally export restrictions more in-line with Washington’s vision. Taiwan Strait tensions have also meaningfully escalated in 2026, with Taiwan hosting it’s largest-ever military drills in August alone. While our independent risk models put chances of a strait incident at around 20% this year, the tail-risk is real: TSMC accounts for ~95%+ of cutting-edge chip production and a disruption would likely grind lead times to above 6 months and induce +25-35% on prices for critical nodes. It’s one reason why the market has not yet abandoned the risk premium discount it applies to Taiwan-exposed players.

Key Drivers

Hyperscaler capex durability: Capex assumptions. The thesis here is simple: capex guidance has been increased rather than lowered at almost every earnings call in 2026, and cloud backlogs (Google Cloud's ~$460B figure is the best example) are booked years in advance rather than on speculation. This bodes well for chipmakers, equipment suppliers, and DC REITs, but means "beats" will have to be better than ever to prevent sharp corrections on any hints of slowing from a single hyperscaler.

Semiconductor supply bottlenecks remain bullish for equipment makers. CoWoS packaging capacity and leading-edge foundrytrices remain the bottleneck on AI chip supply, not end demand or chip design. This means TSMC, ASML, and packaging/substrate suppliers can still charge what they want for leading-edge capacity, enjoy multi-year visibility on their revenues, and should continue to enjoy steady demand from hyperscalers and Chinese neoclouds even if 2026 capex growth falters.

Circular financing increasing AI supply chain balance sheet risk. Nvidia has disclosed nearly $108.5B in gross exposure to financing guarantees, most of it related to a single data center in Ohio that will house compute dedicated to OpenAI, and is reportedly looking to corral financing partners like Apollo, Blackstone, BlackRock, Goldman Sachs, and KKR to backstop >$500B of AI infrastructure build-out. The risk here is that vendor financing inflates demand: cash flows from chip supplier to customer as investment, then flows back to supplier as revenue. Reported top-line growth thus becomes difficult to distinguish from real, economy-wide end-demand. This risk is unique to the AI supply chain and could weigh on valuation multiples if/when AI fails to monetize as expected, even though it hasn't affected reported earnings yet.

Power and electricity constraints: Shipping compute hardware to hyperscalers is no longer the limiting factor on their AI expansions; instead, it's a matter of having enough power and electricity to run it all. Over 60% of 2026 hyperscaler AI capex is directed at power build-out, cooling, and DC construction rather than compute hardware. This is positive for utility operators, grid infrastructure providers, and nuclear/gas powered electricity providers servicing data centers, but could slow the pace at which we see committed AI capex turn into productive capacity.

Risk of further US-China export control escalation. Each new round of export controls, whether its the proposed remote-access rule or the Chip Security Act, shrinks the addressable market for high-margin AI chips served into China and adds incremental compliance costs for hyperscalers and neoclouds operating in third-party countries like Singapore or Malaysia. This is a marginal headwind for chip revenue growth, but has been more than countered by non-China hyperscaler capex growth to date.

Regional Lens

Hyperscaler capex concentration is still very much a US phenomenon: Microsoft, Alphabet, Amazon, and Meta make up most of the 725 billion dollars spent globally on AI infrastructure in 2026, while US capital markets also remain farthest along for AI-native IPOs and vendor financing structures like OpenAI’s rumored confidential IPO filing. US investors and institutions are most exposed to both the potential upside of the AI boom cycle and the potential downside of a circular-financing unwind.

The UK has focused on AI research and applied fintech rather than sheer infrastructure spend. After racking up 12.6 billion dollars of venture funding in the first half of 2026 alone (close to 75 percent of all UK VC funding for that period), UK-based AI firms rank fourth on Dealroom’s Global Tech Ecosystem Index between Silicon Valley, New York, and Boston. London-based AI investment was up almost 100 percent year-over-year to approximately 7 billion dollars, buoyed by DeepMind, Isomorphic Labs, and a concentration of fintech unicorns. More than 50 UK fintech startups are valued at over 1 billion dollars each. The UK’s AI strength is intellectual property and financial-services APIs, not computation.

Canada has focused on enterprise AI and sovereign compute policy around Toronto and Waterloo. Canada’s federal government announced its Sovereign AI Compute Strategy last year, which will provide up to 705 million dollars towards building out domestic compute infrastructure as well as up to 700 million dollars to incentivize investment in private data centers. Cohere, founded in Toronto, raised 300 million dollars from the strategy in May and counts RBC, Bell Canada, and SAP among its enterprise customers. The University of Waterloo’s newly announced partnership with Cohere to build out an AI transformation certificate furthers Canada’s approach of targeting AI talent supply and commercialization rather than bleeding edge model training at hyperscale.

The GCC is now the world’s fastest growing source of AI infrastructure capital. In July 2026, Abu Dhabi’s sovereign investment firm MGX closed its first fund at 49 billion dollars, above its initial target of 45 billion dollars. MGX is now aiming to grow its total assets under management to over 100 billion dollars, already co-leading investments in OpenAI and Anthropic while also co-leading Aligned’s $40 billion acquisition of competitor data center provider Aptum. Saudi Arabia’s sovereign fund PIF has similarly deployed $36.2 billion into AI investments over the past year via its HUMAIN fund, and separately announced $23 billion in chip and cloud investments with Nvidia, AMD, and Amazon combined. Sovereign wealth funds have invested more than $350 billion into AI infrastructure since 2025 alone and funds based in the Gulf have been at the forefront of that growth. For investors like Akrabi Group that cover the GCC, focusing on sovereign capital as it relates to AI will look very different: it is much more about demand generation and co-investment in AI infrastructure than equity ownership in chip design IP or software companies that capture the majority of marginal profits today.

Valuation and Positioning

Technology sector valuations have actually compressed over the past several months rather than expanded further. Forward price-to-earnings multiples for the S&P 500 Information Technology sector have fallen from roughly 40 times to around 20 times, a level last seen before the current AI boom began, even as earnings and capex guidance have continued to rise. This divergence, falling multiples against rising fundamentals, suggests the market has become more discriminating about which AI-exposed names deserve a premium rather than rewarding the theme indiscriminately. Broader market valuation remains elevated in absolute terms, with the S&P 500 trading near 22 to 25 times trailing earnings and a cyclically adjusted price-to-earnings ratio above 30, but the technology sector premium to the index has narrowed.

Concentration risk is the more pressing structural issue. The top seven to ten companies in the S&P 500, nearly all technology or AI-adjacent, now represent 30 to 40 percent of total index weight, a level of concentration not seen in decades. This means broad market index returns are increasingly a leveraged bet on a small number of mega-cap balance sheets and their capex discipline. Free cash flow generation at the largest hyperscalers remains strong in absolute terms, but free cash flow margins are compressing as capex intensity rises, since capital that would otherwise fund buybacks and dividends is being redirected into data centers, chips, and power infrastructure. Buyback activity among the largest hyperscalers has moderated relative to prior years for this reason. On balance, we view the sector as fair to modestly attractive at the index level given the multiple compression already realized, but expensive and increasingly binary within the AI infrastructure sub-segment, where financing structures leave less room for execution error.

Companies or Sub-Sectors to Watch

Semiconductor equipment and advanced packaging: TSMC and its packaging supply chain have the least cyclical profile of any companies we cover, maintaining the strongest pricing power in the industry. CoWoS capacity is full through 2026, while 2 nanometer capacity is booked into 2028, granting them multi-year revenue visibility regardless of any individual hyperscaler's capex plans.

AI infrastructure and chipmakers: Nvidia is by far the largest winner from hyperscaler capex, but its ballooning financing exposure (up to $108.5 billion in guarantees disclosed over the last year) forces us to size positions with the circularity of that financing risk explicitly in mind. We're no longer comfortable thinking about revenue growth in isolation.

Hyperscalers and cloud platforms: Microsoft, Alphabet, and Amazon tie profitable, cash-flow generative core businesses (Windows and Office, Search, and e-commerce and AWS respectively) to their AI infrastructure initiatives, giving them more balance sheet flexibility to weather a capex digestion phase than standalone AI infrastructure providers.

Enterprise software with embedded AI: Enterprise software companies that are able to stitch algentic AI into their existing workflows benefit from AI monetization without some of the capital intensity / financing risk of the infrastructure layer. Canadian company Cohere operates in this space through partnerships with SAP and RBC.

Power infrastructure and grid equipment serving data centers: As over 60% of hyperscaler capex earmarked for AI goes towards power & cooling rather than compute hardware, utilities and grid equipment suppliers with major exposure to data center buildouts provide a lower-multiple, less circular way to play the cycle.

Key Risks

A reduction/stop in hyperscaler capex guidance is probably the one biggest catalyst we see for an indiscriminate de-rating because most of the implied infrastructure valuations for AIaaS today assume continued capex growth at >75 percent annually rather than flattening.

Explicitly traced credit tightening is the least significant tail because we know Nvidia has exposure to guarantees and has formed a financing vehicle with Apollo/Blackstone/BlackRock/Goldman Sachs/KKR, so if OpenAI (or another large loss-making AI lab) cannot monetize at scale, losses could cascade through the hardware companies, cloud providers, and private credit funds underwriting new DCB.

Taiwan conflict not necessarily included. Even if China and the US do not go to war, our base case has a 25 to 35 percent price increase and greater than 6 months delay on nodes where TSMC has >90 percent market share of leading-edge fab capacity, upending the entire hyperscale AI hardware stack at once.

Additional restrictions on US exports, such as the recent proposal to eliminate the remote-access loophole that Chinese AI companies have been using to circuminate rent cloud compute from suppliers in countries such as Singapore, could limit revenue upside for semiconductors and neoclouds with exposure to SEA sooner than currently expected, especially if the final rule is broader.

Bottom Line

AI infrastructure cycle in technology is being driven by non-speculative, contracted, multi-year capital dollars. Coupled with semiconductor supply bottlenecks, picks- and-shovels names have a lasting pricing power into 2027. Maintain exposure to the theme via infrastructure, equipment, and power adjacent names. However, view mega-cap concentration risk and circular capital structures as actual tail risks, not background noise, and manage your position size accordingly.