Big Tech earnings 2Q26: Demand confirmed; Higher burden of proof

The post-earnings share price performance also suggests that investors are no longer rewarding AI-driven revenue growth in isolation. Instead, the market is increasingly focused on the cost required to generate that growth. Among the four major hyperscalers, the key differentiator was not the strength of demand, but rather each company's ability to finance its AI investments without relying on external funding.

Alphabet’s subsequent recovery suggests that investors are still willing to tolerate near-term FCF pressure, where confidence in monetisation remains high, although short-term price action alone is insufficient evidence of durable conviction.  

Meta, on the other hand, declined as weaker forward guidance and rising costs outweighed evidence of AI-driven improvements in its advertising business, while visibility on near-term returns remained limited.

Figure 1: Price movements 1 day after earnings

AI-monetisation indicators – cloud growth accelerated across the board

Cloud growth acceleration was the biggest positive takeaway this quarter. AWS delivered its fastest growth in 18 quarters, Azure exceeded its own 39–40% guidance while surpassing an annualised revenue run rate of USD100 billion, and Google Cloud nearly doubled its already elevated growth rate. More importantly, the simultaneous acceleration across all three cloud providers effectively rules out the notion of market share gains at the expense of one another. Instead, it suggests that the overall addressable market continues to expand, with demand rising across the industry.

Order backlogs also continue to reinforce the bullish investment case. Collectively, the three hyperscalers now have approximately USD1.7 trillion of contracted future revenue, with backlog growth continuing to outpace recognised revenue across all three companies. This is a clear indication of supply constraints rather than demand weakness. In other words, reported revenue is being determined by the pace at which new capacity can be deployed, rather than by the volume of customer orders. Management commentary was remarkably consistent: Alphabet reiterated that it continues to operate in a supply-constrained environment, Microsoft stated that Azure demand still exceeds available capacity, while Amazon acknowledged that it does not expect to have sufficient capacity to meet customer demand through 2026 and 2027.

Meta does not have a cloud business, making its AI monetisation indicators less direct. Instead, evidence of AI-driven returns can be seen in its advertising business, where large language model (LLM)-powered recommendation and targeting systems increased ad clicks by 8.3% and conversions by 15.7% on Facebook, contributing to 27% growth in advertising revenue and 73% growth in Family of Apps Other revenue.

However, given the lack of direct AI monetisation disclosures and forward guidance that fell short of market expectations, Meta's share price declined following its earnings release. The absence of a clearer timeline for material incremental returns increases the duration and execution risk of Meta’s AI investment case, even if the long-term opportunity remains substantial.

While this approach has the potential to create greater long-term value, it also requires investors to be patient as they wait for meaningful returns on Meta's substantial AI investments.

In short, for cloud companies, this quarter earnings continue to proof that near-term cloud revenue is a capacity-delivery forecast, not a demand forecast. Demand-side uncertainty lies further ahead, at the point where today's record backlogs either convert into recognised revenue or are renegotiated.

Figure 2: Remaining Performance Obligation rising

Figure 3: Cloud Growth accelerating

Cost discipline: is the spending being managed?

To support their aggressive AI capital expenditure plans, hyperscalers are increasingly focused on measures to improve cost efficiency. The most notable initiatives have been workforce rationalisation and greater vertical integration through the development of proprietary AI silicon.

The most immediate cost-saving measure has been headcount optimisation. Meta reduced its workforce by approximately 8,000 employees in May, bringing total headcount down 3% quarter-on-quarter to just over 75,000, incurring USD1.18 billion in severance costs. Microsoft also implemented a voluntary retirement programme alongside severance charges and an Xbox impairment. While these restructuring initiatives are intended to improve long-term efficiency, they weighed on near-term profitability. Meta's total expenses increased 55% year-on-year to USD42.0 billion, reducing its operating margin to 31%. Management attributed the increase to higher compensation, infrastructure investments, legal expenses, and notably, rising third-party AI token costs. Meanwhile, Alphabet cited continued operational discipline across the business as a key driver behind its expanding operating margin.

Custom silicon emerging as a potential margin defence, and it is at the beginning of being scaled. The major hyperscalers are all ramping their in-house AI accelerators, including Alphabet's TPU v7 (Ironwood), Amazon's Trainium 3, Microsoft's Maia 200 built on TSMC's 3nm process, and Meta's MTIA family, with the 300- and 500-series announced in March based on a RISC-V architecture.

Among the four, we believe Alphabet and Amazon currently hold the strongest positions. Alphabet began shipping TPU systems to customer data centres during the second quarter, although management expects only a relatively small portion of revenue from existing TPU system sale agreements to be recognised this year, with a more meaningful contribution expected as deployments accelerate through 2026 and the majority of revenue recognised in 2027. Amazon, meanwhile, is reportedly evaluating the sale of its Trainium chips to third-party data centre operators, which could create an additional high-margin revenue stream. Internally, Alphabet is already running the majority of its inference workloads on TPUs rather than merchant GPUs, highlighting the maturity of its custom silicon ecosystem.

Performance metrics are also becoming increasingly compelling. Amazon stated that its Graviton processors deliver approximately 30%–40% better price-performance than comparable alternatives, while Microsoft's Maia 200 and Alphabet's TPUs are reported to offer roughly 30% better performance per dollar. These improvements strengthen the economic case for proprietary silicon as AI inference workloads continue to scale.

As highlighted in our previous article, we are also observing customers shifting their focus from maximising token usage to optimising the cost of each token generated. For many workloads, customers are increasingly willing to accept slightly lower-quality outputs in exchange for significantly lower inference costs. This behavioural shift should further accelerate the adoption of proprietary AI chips, particularly as inference now accounts for roughly two-thirds of total AI compute, expanding the proportion of workloads that can migrate to captive silicon.

Margin capture, not cost avoidance. Custom silicon can shift part of the economics from external suppliers to the hyperscalers, enabling them to capture the  economic value.  Moreover, cloud margin/profitability expansion is consistent with this thesis, portrayed by AWS's operating margin expanding to 39.4% (+650 basis points year-on-year) and Google Cloud's operating margin reaching 35.6%.

However, proprietary silicon does not eliminate the need for elevated AI capital expenditure, it simply improves the return generated on those investments.

Figure 4: Operating Margin stabilising

Figure 5: Cost savings (management commentary)

Source: Company earnings. IFAST compilations. Data as of 7 August 2026.

CAPEX momentum remains strong

Capital expenditure (CAPEX) continued to accelerate across the hyperscalers this quarter, with Apple remaining the only major technology company that did not meaningfully increase investment. Amazon led the group with CAPEX of USD54.2 billion, up from USD32.1 billion a year ago and marking the largest single-quarter investment among its peers. Alphabet reported CAPEX of USD44.9 billion (+100% YoY), followed by Microsoft at USD41.0 billion and Meta at USD31.1 billion.

Reflecting the stronger-than-expected spending plans, we have once again revised our combined 2026 CAPEX forecast upward to USD760 billion, representing approximately 85% year-on-year growth, while maintaining our 2027 forecast of USD1.1 trillion. Although Microsoft lowered its reported CAPEX guidance, this primarily reflects an accounting reclassification whereby certain leases were moved from finance leases to operating leases. As a result, the associated cash payments are now recognised within operating cash flow rather than capital expenditure, without affecting the underlying level of investment. Adjusting for these reclassified operating leases, we continue to estimate Microsoft's total AI-related investment at approximately USD190 billion in 2026. Higher memory and component costs remain the primary drivers of upward CAPEX revisions, with both Microsoft and Meta explicitly citing rising component prices in their capital expenditure commentary.

The implications of today's elevated CAPEX will become more apparent over the coming years as these investments flow through the income statement as depreciation expense. The magnitude of this impact depends largely on the composition of spending. Microsoft disclosed that approximately two-thirds of its CAPEX is allocated to shorter-lived assets such as GPUs and AI server racks, which carry an estimated useful life of around six years, while the remaining one-third is invested in longer-lived infrastructure assets such as data centre buildings. Although the other hyperscalers did not provide a comparable breakdown, we expect Amazon to face the greatest relative depreciation burden, given its substantially higher absolute CAPEX relative to its adjusted earnings base. Collectively, we expect depreciation expense across the group to increase by an average of 46% in 2026, followed by a further acceleration of 44% and 37% in 2027 and 2028, respectively.

Overall, our view remains unchanged. While AI-related CAPEX guidance continues to trend higher, the magnitude of upward revisions has begun to moderate compared with previous quarters. We believe this is an encouraging development, as it suggests the pace of incremental spending is starting to normalise, which should help alleviate investor concerns over further deterioration in free cash flow while preserving the long-term AI investment thesis.

Figure 6: CAPEX revised higher for 2026

Figure 7: Depreciation Expenses higher for the next 3 years

Free Cash Flow and Debt – the casualties

Free cash flow (FCF) has emerged as the primary casualty of the AI investment cycle. All major technology companies, with the exception of Apple, reported a significant decline in FCF compared with the prior year as elevated AI infrastructure spending weighed on cash generation.

Both Alphabet and Amazon reported negative free cash flow during the quarter. Notably, Alphabet recorded its first negative FCF since its 2004 IPO, occurring earlier than market expectations, which contributed to the sharp negative share price reaction following its earnings release.

Nevertheless, the underlying balance sheet strength of these companies remains intact, with cash reserves still at healthy levels. This suggests that while near-term FCF generation has been pressured by aggressive AI investments, the ability of these companies to generate substantial operating cash flows remains robust.

The deterioration in FCF has also driven a shift in funding strategies. During 2Q26, we continued to observe increased debt issuance and greater use of off-balance-sheet financing structures to fund data centre expansion. As a result, leverage levels have continued to rise across the hyperscaler group. Alphabet and Amazon have been the most notable examples, with both companies returning to the debt markets during the quarter, while Alphabet raised both debt and equity capital. Meta raised USD25 billion to pre-fund future CAPEX requirements and suspended share buybacks entirely, whereas Microsoft has remained relatively conservative and has yet to meaningfully increase external funding.

Looking ahead, we expect funding activities—including debt issuance and off-balance-sheet financing—to continue through the remainder of 2026 and into 2027 as hyperscalers maintain elevated AI infrastructure investments. While current debt capacity remains sufficient, companies will need to demonstrate sustained AI-driven revenue growth and attractive returns on invested capital to justify increasingly aggressive funding strategies.

Figure 8: Debt levels generally increased

Shareholder value creation has taken a back seat

Corporate actions aimed at enhancing shareholder value have also slowed amid the ongoing AI investment cycle. During 2Q26, share repurchase activities were paused across all major technology companies except Microsoft, which continued to execute buybacks and announced USD10.2 billion of share repurchases and dividends. This marks another key differentiating factor for Microsoft, alongside its positive free cash flow generation and lack of significant bond issuance.

The suspension of buybacks represents an alternative source of funding, allowing companies to preserve cash for AI infrastructure investments. However, this comes at the expense of capital that could otherwise have been allocated towards shareholder return initiatives, including share repurchases and dividend distributions.

As hyperscalers continue to prioritise AI infrastructure spending, the trade-off between funding long-term growth investments and returning capital to shareholders is becoming increasingly evident. While elevated AI investment may create substantial long-term value, investors will require greater visibility on monetisation and returns before shareholder value creation activities can meaningfully resume.

Figure 9: Share buybacks have been decreasing

A reminiscence of the dotcom bubble? Not quite yet.

Recent developments—such as rising capital intensity increasingly funded through external capital, circular financing arrangements, greater accounting flexibility (most notably Microsoft's extension of asset useful lives), off-balance-sheet structures carrying obligations not fully reflected on balance sheets, and the concentration of USD1.7 trillion in contracted backlog among a limited number of counterparties—do share some similarities with characteristics observed during the dotcom era.

However, while concerns around a potential AI bubble are understandable, we believe the current environment has yet to reach the stage of speculative euphoria seen during the late 1990s. Unlike the dotcom period, today's leading AI beneficiaries are supported by substantial revenue generation, strong balance sheets, and clear evidence of enterprise demand.

That said, we continue to closely monitor several key developments to ensure our investment thesis remains disciplined. 

Source: Bloomberg Finance L.P., iFAST compilations. Data as of 5 August 2026.

The internet thesis ultimately proved correct, but many early infrastructure investors failed to capture its economics. The relevant parallel for AI is therefore not whether demand materialises, but whether today’s capital providers earn adequate returns on the capacity being built.

Now, the burden of proof has shifted from demonstrating AI demand to showing that new capacity converts into incremental cash returns above the cost of capital, without excessive dilution or balance-sheet risk.

Read more: Microsoft 2Q26 Earnings Update: Passing with Flying Colours, 

Alphabet 2Q26 earnings: Free cash flow turns negative. Should investors be worried?

Meta 2Q26 earnings: AI monetisation may take longer, yet the sell-off offers 89% upside potential 

Amazon 2Q26 Earnings Update: Tale of Two Halves

Reaffirm our Positive call on Big Tech

2Q26 confirmed the demand case and reframed the risk. Revenue is capped by delivered capacity, not by orders— but relieving that constraint now costs more than these businesses generate, and the market is sorting them by who must fund it externally.

Microsoft currently offers the strongest combination of internal funding capacity and visible monetisation; Alphabet and Amazon show stronger cloud momentum but face rising capital intensity, while Meta’s returns remain the most back-loaded and execution-dependent.

We continue to prefer the Invesco NASDAQ Internet ETF, which has 39% exposure towards big tech companies.