Big Tech stopped paying you back

For most of the last decade, Big Tech rewarded shareholders with a simple deal: buy the stock, get the cash back. Buybacks and dividends were the currency of trust. Apple repurchased $600 billion in shares over ten years. Alphabet and Meta returned tens of billions annually. Amazon was the lone holdout, but even it ran a $10 billion buyback program in 2022. That deal is over. HSBC's latest analysis of the "Tech-7" (Alphabet, Amazon, Apple, Meta, Microsoft, Nvidia, and Oracle) reveals a capital allocation shift that would have been unthinkable five years ago. In 2026, these companies are expected to spend 61% of their operating cash flow on capital expenditure, up from 46% in 2025. The share going to buybacks will fall to 16%, down from 22%. Dividends drop to 5% from 6%. For every dollar of operating cash flow, shareholders now get back roughly 21 cents. The rest goes to data centers, GPUs, and cooling systems. Shareholders are funding an AI arms race whether they like it or not.

The numbers behind the shift

The scale is hard to overstate. The four major hyperscalers, Alphabet, Amazon, Meta, and Microsoft, are projected to spend between $635 billion and $720 billion on capital expenditure in 2026, depending on whose estimates you use. Bridgewater Associates pegs it at $650 billion. S&P Global puts it above $700 billion. Either way, it represents a roughly 60% to 70% increase over the $381 billion to $410 billion these companies spent in 2025, which itself was a nearly 50% jump from the $240 billion spent in 2024. The individual numbers are staggering. Amazon leads with a $200 billion capex plan, nearly all of it earmarked for AWS. Alphabet is forecasting $175 billion to $185 billion, doubling down on Gemini and Google Cloud. Microsoft is on pace for roughly $145 billion, channeled into Azure and its OpenAI partnership. Meta projects $115 billion to $135 billion, focused on Llama models and AI-powered ad infrastructure. Meanwhile, the cash flowing back to shareholders is shrinking. In Q4 2025, combined buybacks by Amazon, Alphabet, Microsoft, Meta, and Oracle fell to $12.6 billion, the lowest level since early 2018 and a 74% decline from the roughly $48 billion peak in 2021. Bloomberg reported that Alphabet and Microsoft spent roughly $11 billion on buybacks that quarter, while Amazon and Meta held off entirely. Amazon hasn't repurchased stock since 2022. The free cash flow picture tells the same story. In 2024, the four biggest U.S. internet companies generated a combined $237 billion in free cash flow. By 2025, that dropped to $200 billion. The projections for 2026 are grim. Pivotal Research estimates Alphabet's free cash flow will fall nearly 90%, from $73.3 billion to $8.2 billion. Morgan Stanley projects Amazon will swing to negative free cash flow of $17 billion. Barclays sees a similar collapse for Meta, forecasting negative free cash flow in 2027 and 2028. As Barron's noted, these companies would need to grow operational cash flows by about 30% just to maintain the same free cash flow levels as 2025.

Nobody can afford to stop

Here is the uncomfortable logic driving every one of these spending decisions: if your competitor is pouring $200 billion into AI infrastructure and you don't match them, you lose. This isn't speculation. It's game theory. In a competitive landscape where AI capabilities increasingly determine who wins cloud contracts, who dominates digital advertising, and who controls the next generation of software platforms, underinvestment is an existential risk. Microsoft's Azure cloud grew 33% year-over-year in Q3 FY25, with AI contributing 16 percentage points of that growth. Google Cloud revenue grew 48% year-over-year to $17.7 billion in Q4 2025. Amazon Web Services posted its fastest growth in 13 quarters. The companies spending the most are the ones winning the most. The result is a classic arms race dynamic. No individual company can afford to pull back, even if collective restraint would be better for all of them. Each company's rational decision to spend more forces every other company to spend more. The equilibrium keeps shifting upward. HSBC calls it a "megacycle." The bank expects the Tech-7 to generate $1.3 trillion in operating cash flow before taxes and interest in 2026, and projects group revenue of $2.8 trillion, up from $2.3 trillion in 2025. The money is there. It is just not going where shareholders want it to go.

Cheaper AI makes the bill bigger

You might expect that improvements in AI efficiency would eventually bring spending under control. The opposite is happening. When DeepSeek released its efficient open-source model in early 2025, some analysts predicted AI costs would fall. Microsoft CEO Satya Nadella had a different take: "Jevons paradox strikes again! As AI gets more efficient and accessible, we will see its use skyrocket, turning it into a commodity we just can't get enough of." Jevons paradox, named after the 19th-century economist William Stanley Jevons, describes what happens when a resource becomes more efficient to use: total consumption goes up, not down. Jevons observed this with coal in Victorian England. More efficient steam engines didn't reduce coal consumption. They made coal-powered industry viable in more contexts, which drove demand higher. The AI version is playing out in real time. Per-token inference costs have dropped roughly 280-fold since late 2022. But total inference spending grew 320% over the same period. Inference workloads now account for 55% of AI infrastructure spending, up from 33% in 2023. Cheaper inference doesn't mean smaller cloud bills. It means companies deploy AI to more tasks, more users, more edge cases. The unit cost drops, but the total bill climbs. This is the "megacycle" HSBC describes. It is self-reinforcing. Efficiency gains don't slow spending. They accelerate it by unlocking new use cases that demand more compute. Analyst estimates have consistently underestimated AI capital expenditure, and the gap between forecast and actual spending keeps widening.

Is this the telecom bubble again?

The most common historical comparison is the late 1990s telecom overbuilding. Between 1996 and 2001, telecom companies laid over 100,000 miles of fiber optic cable, increased network capacity by 186,000 times, and then watched as only 2.7% of that fiber was actually being used by 2002. Half a million people lost their jobs. Seven trillion dollars in market value evaporated. Twenty-three major telecom companies went bankrupt simultaneously. The parallel is tempting but imperfect. The telecom bubble was fueled by a false premise, that internet traffic was doubling every 100 days when it was actually doubling once per year. The companies building the infrastructure were mostly debt-funded upstarts, not the most profitable businesses on the planet. Today's AI spenders are different. They are generating real revenue from AI services. They have enormous cash flows to fund the build-out. And the demand signals, while uncertain in their exact trajectory, are grounded in measurable enterprise adoption rather than purely speculative projections. But the parallel holds in one critical way: the pattern of infrastructure build-outs is always overshoot then consolidation. Railroads in the 1860s and 1870s. Fiber optic in the late 1990s. Cloud computing in the early 2010s. In every case, the infrastructure eventually became essential, but the companies that built it often didn't survive to benefit. Global Crossing laid cable across the ocean floor, went bankrupt, and the cable became the backbone of the modern internet decades later. The infrastructure outlasted the company. The question for Big Tech is whether they are building railroads that will carry commerce for a century, or whether they are laying fiber that will sit dark for years before demand catches up. The honest answer is probably both, and nobody knows the ratio.

Morgan Stanley thinks they're right

Morgan Stanley has taken one of the more bullish positions on the AI infrastructure cycle. The firm estimates that nearly $3 trillion in AI-related infrastructure investment will flow through the global economy by 2028, with more than 80% of that spending still ahead. In March 2026, the firm warned that an AI breakthrough is coming, and most of the world isn't ready for it. If Morgan Stanley is right, the spending is not just justified, it is insufficient. The companies building capacity now will be the ones positioned to capture the value when adoption hits its inflection point. Fewer than 20% of U.S. establishments currently use AI for any business function. The adoption curve hasn't even approached its steepest part. But if Morgan Stanley is wrong, if the breakthroughs plateau or adoption stalls, then what we are witnessing is the largest capital misallocation in corporate history. Not because the technology doesn't work, but because the infrastructure was built for a demand curve that never materialized at the expected pace. Bridgewater co-chief investment officer Greg Jensen has described the current moment as a "more dangerous phase" of the AI boom, marked by exponentially rising investments in physical infrastructure and growing reliance on outside capital. Amazon has already filed with the SEC indicating it may need to raise equity and debt to fund its build-out. Alphabet held a $25 billion bond sale in late 2025, quadrupling its long-term debt to $46.5 billion. These are not the balance sheet moves of companies with comfortable margins of safety. These are the moves of companies betting their financial structure on the future they're building.

What shareholders actually bought

For the last decade, the implicit pitch of Big Tech stocks was that you were buying a cash machine. These companies generated so much profit that they could invest heavily in growth and still return enormous sums through buybacks and dividends. The investment thesis was built on capital return. That thesis is being rewritten. What shareholders now own is not a cash machine. It is a claim on AI infrastructure that may or may not generate proportional returns. The cash is still being generated, more of it than ever. But it is being redirected before it reaches investors. This isn't necessarily wrong. If AI becomes as foundational as its proponents believe, early infrastructure investors will be rewarded handsomely. The companies that built cloud infrastructure a decade ago, Amazon, Microsoft, Google, now generate hundreds of billions in revenue from it. The same logic could apply to AI compute. But investors should be clear-eyed about what is happening. They are not receiving less cash because the companies are struggling. They are receiving less cash because the companies have decided the money is better spent on GPUs than on buybacks. It is a bet, made with shareholder capital, on a future that is promising but uncertain. The discomfort is that shareholders didn't vote on this. Nobody asked retail investors whether they'd prefer $50 billion in buybacks or $50 billion in data centers. The decision was made for them, by management teams who believe, perhaps correctly, that the alternative is obsolescence.

The gap between spending and proof

Goldman Sachs chief economist Jan Hatzius pointed out that AI investment contributed "basically zero" to U.S. economic growth in 2025. Much of the hardware powering AI, chips from TSMC, memory from Samsung, is manufactured overseas, so the spending adds to Taiwanese and Korean GDP, not American. Even domestically, the productivity gains haven't shown up in aggregate data yet. But absence of proof is not proof of absence. Goldman's own projection still holds that AI could increase U.S. productivity growth by 1.5 percentage points annually over a ten-year period. Management teams who quantified AI-driven productivity impacts on specific tasks reported a median gain of around 30%. The gains are real. They're just not yet big enough, or widespread enough, to move the macro needle. The question is one of timing. Every major infrastructure build-out in history has had a gap between the investment phase and the return phase. The companies spending today are betting that gap will close before patience runs out, before shareholders revolt, before the debt becomes unmanageable, before a competitor finds a more efficient path. HSBC is betting the same way. Despite the concerns, the bank remains bullish on Big Tech, projecting strong revenue growth and noting that Nvidia alone will contribute 33% of the Tech-7's absolute revenue growth in 2026. The spending is producing revenue. It is just producing more capex faster.

What this means if you hold the stock

If you own shares in Alphabet, Amazon, Meta, or Microsoft, the math is simple. You are getting less cash back than you used to, and the gap is widening. The money that would have supported your stock price through buybacks is going into the ground, literally, in the form of data centers across Virginia, Texas, and Iowa. This doesn't mean the stocks are bad investments. It means the investment thesis has changed. You are no longer buying a cash return story. You are buying an infrastructure story, with all the upside and downside that implies. The upside is that AI infrastructure could be the most valuable asset class of the next decade. The downside is the same as every infrastructure bet in history: the returns might come later than expected, or to different players than the ones writing the checks. For now, shareholders are along for the ride. The companies have made their choice. The capex is committed. The data centers are being built. And the cash that used to come back to investors is being poured into a future that everyone hopes will justify the cost. Whether it does is the $650 billion question.

References