• Freestyle
  • Posts
  • Is AI Forcing a Midlife Crisis on Cloud Hardware?

Is AI Forcing a Midlife Crisis on Cloud Hardware?

Freestyle is where we examine the changing tides of technology from our front-row seats. These are raw, evolving thoughts—half-baked ideas meant to spark conversation. The real refinement happens when you reply, challenge, and build on what we put out there. 🤝 

Cloud Hardware’s Mid-Life U-Turn: Why AWS and Azure Are Reversing Course on Lifespans

Last month, we asked whether AWS or Azure was poised to reign as the AI King, expecting Q4 2024 earnings to settle the debate—yet as the numbers rolled in, an even bigger twist emerged: both cloud giants abruptly reversed course on how they depreciate their hardware.

For years, cloud giants Amazon Web Services (AWS) and Microsoft Azure seemed to believe their servers and switches were like fine wine – getting better (or at least lasting longer) with age. Both companies repeatedly extended the “useful life” of their data center hardware, delaying retirement and depreciation. Microsoft’s CFO famously announced in 2022 that Azure’s servers and network gear could depreciate over 6 years instead of 4, crediting software optimizations and better tech for stretching asset life. This accounting tweak was a boon for the bottom line – Microsoft expected to save $3.7 billion in FY 2023 alone from the change. Not to be outdone, AWS performed similar magic: in early 2022 it bumped server lifespan from 4 to 5 years (and networking gear from 5 to 6), and by 2024 AWS was convinced it could squeeze out a 6th year on its servers, adding an extra $900 million to Q1 2024 profit thanks to lower depreciation expenses. In short, life was good – and getting longer – for cloud hardware, and depreciation schedules kept stretching in tandem with hyperscalers’ confidence.

That’s why it came as a surprise in this month’s earnings when AWS abruptly hit the brakes on this trend. After years of lengthening hardware lifespans, Amazon’s finance chief announced a plot twist: for certain AI-focused infrastructure, 6 years was too long – they’re cutting it back to 5. This marked the first major reversal of the cloud depreciation trend, and it has industry-watchers asking: What’s changed?

AWS’s 2025 Reversal: When 6 Years Became 5 (Again)

In AWS’s Q4 2024 earnings call, CFO Brian Olsavsky revealed the company had completed a new “useful life study” and didn’t love the results. The pace of technological advancement in AI/ML was accelerating so fast that some of AWS’s shiny new gear will get old faster than expected. His response: starting January 2025, a subset of servers and networking equipment (think AI training servers with GPUs and their high-powered switches) will have a 5-year lifespan, not 6. Essentially, AWS is un-doing last year’s extension – at least for the AI-related hardware.

This wasn’t just a minor policy tweak – it carries a price tag. Shortening the depreciation schedule means AWS will recognize costs sooner, shrinking operating profit by an estimated $700 million in 2025. It’s a bit ironic: in 2024 extending server life boosted Amazon’s profit, and in 2025 shortening it will dent the profit. (One might imagine AWS’s accountants ruefully swapping out their “stretch” rubber band for a “snap-back” one.) Moreover, Amazon disclosed it had retired some servers and networking gear early, taking a one-time $920 million depreciation charge – essentially admitting that certain hardware wasn’t going to make it to its previously expected old age.

Microsoft’s Azure is on a similar path, although they haven’t made a public announcement specifically about depreciation lives shortening (yet). Azure was the first of the big clouds to push to a 6-year server life, back in 2022, and enjoyed the immediate financial upsides. But Microsoft is facing the same competitive and technological pressures as AWS. In recent earnings guidance, CFO Amy Hood has been upfront that Microsoft’s capital expenditures are rising sharply to build AI capacity, and she indicated that the “mix of spend will begin to shift back to short-lived assets” – meaning more servers and chips, fewer ultra-long-lived buildings – which will “increase depreciation” going forward​. It’s as close as you get to saying: we know we just gave ourselves a longer timeline, but now we’ll be buying so many new servers (and replacing them faster) that the effective life will shorten again. Microsoft hasn’t explicitly said “we’re going back to a 5-year depreciation for servers,” but their commentary leaves breadcrumbs pointing that direction.

Why Shorter Lifespans?

What’s driving this depreciation U-turn after years of lengthening hardware life? A few compelling (and intertwined) reasons stand out:

  • 🚀 AI Workloads Are Brutal: The new wave of AI servers – packed with power-hungry GPUs – are working 24/7, and it turns out they age in dog years. Under constant heavy load (often 60–70% utilization), datacenter GPUs might survive only 1–3 years before performance and reliability plummet. These chips guzzle 700 watts and run hot, putting enormous stress on hardware and shortening their real-world lifespan. In a recent interview, Daniel Gross noted: “you have a data center, which is, I think, a 30‑year depreciation, and then the GPUs are, I think, accounted for in a five‑year depreciation, but actually are unusable after about 36 months.” In other words, an AWS server filled with Nvidia GPUs may simply wear out or fail faster, making a 6-year depreciation plan look fanciful. AWS’s own experience of retiring hardware early (with that $920 million hit) underscores how AI gear is burning out sooner than general-purpose equipment.

  • ⏱️ Nvidia’s Blistering Innovation Cadence: Even if the hardware survives physically, it can become obsolete overnight thanks to the breakneck pace of AI chip innovation. Nvidia has been refreshing its flagship GPUs annually – a “very aggressive design cycle of roughly a year between major releases,” as one tech strategist noted. Each new generation (A100 to H100 to whatever comes next) brings big leaps in performance. That means after 2–3 years, last-gen chips look painfully slow for cutting-edge AI tasks. By year 5 or 6, a GPU is practically ancient history. No cloud provider wants to be caught with an AI infrastructure that’s generations behind, so they’ll upgrade sooner – and shorten the accounting life accordingly.

  • 🤝 Keeping Up with the Cloud Joneses (Competitive Pressure): The cloud market is ultra-competitive, and big customers demand the latest and greatest hardware. If Azure offers a new GPU instance type and AWS is still on older tech (or vice versa), cloud buyers will take notice. Running enterprise workloads on 5+ year-old servers starts to sound like using a flip phone in the 5G era. In fact, when Microsoft pushed server life to 6 years, industry observers warned that clients might not love their cloud instances running on such aged silicon – there could be 3+ new processor generations in that span. Fast-forward to the AI era, and that concern is even sharper. Cloud providers simply can’t afford to let infrastructure linger too long, not when every year could bring a big jump in AI capability. Shorter depreciation reflects a reality that gear will be turned over faster to stay in the game.

  • 💰 Financial Engineering and Hidden Incentives: Of course, we can’t ignore the accounting and strategic angles. Depreciation isn’t a cash expense – it’s an accounting allocation – and tech CFOs have not been shy about tweaking useful life to manage earnings. In leaner times, extending asset life is an easy way to instantly pad profits by cutting depreciation costs. (Collectively, Big Tech added an estimated $10 billion to earnings by lengthening server lifespans in recent years!) Now, however, that trick has run its course. “We don’t see any mega-cap extending useful life beyond the 6-year schedule, as GPU cycle times are increasing rapidly,” noted Barclays analysts, meaning the era of ever-longer depreciation is over. Overestimating lifespan can backfire – under-depreciating assets inflates short-term profit, only to smack you with a write-down later when the tech needs replacing. AWS’s move to shorten AI hardware life could be seen as preemptive realism: taking a smaller hit now, rather than risking a much larger blow when outdated gear forces an unscheduled write-down. There might also be a tax timing benefit to accelerating depreciation (front-loading deductions before tax rules change or asset values drop), but the primary driver is aligning accounting with actual usage. Microsoft, for its part, hasn’t yet publicly revised its 6-year schedule for Azure hardware – but given they’re pouring $80 billion into AI data centers in FY2025, one has to wonder if Azure’s accountants will eventually face the same music for GPU-packed servers.

The Bigger Picture: Tech Strategy with Financial Side-Effects

For cloud providers, depreciation schedules are more than just accounting arcana – they’re a window into strategy and confidence in technology. The past trend of lengthening hardware life spoke to maturing infrastructure: slower innovation in basic server tech, improved durability, and a dash of cost-optimization. The 2025 reversal, spearheaded by AWS, signals a new chapter where AI-centric hardware is evolving so fast that even trillion-dollar companies feel compelled to shorten their depreciation timetable. Azure and others will be watching closely; no one wants to fall behind either technologically or financially. We can expect more granular approaches – perhaps separate depreciation pools for AI gear versus conventional hardware – as CFOs seek to accurately reflect these differing life cycles without tanking their earnings completely.

Financially, the impact of this shift is significant but manageable for the giants. Amazon’s $700 million hit to 2025 operating income is notable, but in the context of AWS’s $39.8 billion operating profit in 2024, it’s a blip the company can absorb. Microsoft hasn’t yet reported a similar hit, but if and when it revises Azure’s depreciation, there could be a substantial swing given the scale of its data center assets (recall that extending to 6 years added ~$3.7 billion to Microsoft’s 2023 profit, so shortening would do the opposite). Investors and analysts are already factoring this in: some have trimmed earnings forecasts for cloud providers, recognizing that depreciation of all that fancy new AI kit will weigh on margins in coming years. In the grand scheme, these companies are willing to take a short-term accounting hit for long-term strategic gain – they’d rather invest aggressively in AI infrastructure (and reflect its true cost) than hold onto outdated assumptions for the sake of immediate profit.

In an entertaining twist, the cloud titans are learning a classic lesson: what goes up must come down – and what gets extended can also be shortened. The reversal on hardware lifespan is a testament to how quickly the tech landscape is shifting under our feet (or under their racks). Even the most unglamorous accounting line can hint at seismic changes in strategy. For AWS and Azure, it’s a balancing act: keep customers happy with the latest hardware, keep investors content with healthy profits, and ensure that when their servers hit a mid-life crisis, it’s the productive kind.

Ultimately, the shortening of depreciation lives for AI hardware suggests that cloud providers are embracing a faster innovation cycle, effectively saying: “We’ll replace gear sooner, because we need to – and we’ll account for it accordingly.” It’s a pragmatic move, and likely a wise one. In the race to AI supremacy, the useful life of a server may not be what it used to be, but the ability to adapt – even in accounting practices – is more valuable than ever. The cloud is not only about elastic computing; elastic accounting is part of the toolkit too, stretching in good times and snapping back when the tech demands it. As AWS and Azure navigate this new era, the rest of the industry is surely taking notes (and perhaps recalculating their own depreciation spreadsheets). After all, in the cloud business as in life, change is the only constant – sometimes even for the estimated life of a server.

🤙

The opinions expressed in this newsletter are my own, subject to change without notice, and do not necessarily reflect those of Timeless Partners, LLC (“Timeless Partners”). This newsletter is an informal collection of thoughts, articles, and reflections that have recently caught my attention. For discussion purposes, I may present perspectives that contradict my own. I fully expect to change my mind on some topics over time and hope that readers approach these ideas with the same mental flexibility.

Nothing in this newsletter should be interpreted as investment advice, research, or valuation judgment. This newsletter is not intended to, and does not, relate specifically to any investment strategy or product that Timeless Partners offers. Any strategy discussed herein may be unsuitable for investors depending on their specific objectives and situation. Investing involves risk and there can be no assurance that an investment strategy will be successful.

Links to external websites are for convenience only. Neither I, nor Timeless Partners, is responsible for the content or use of such sites. Information provided herein, including any projections or forward-looking statements, targets, forecasts, or expectations, is only current as of the publication date and may become outdated due to subsequent events. The accuracy, completeness, or timeliness of the information cannot be guaranteed, and neither I, nor Timeless Partners, assume any duty to update this newsletter. Actual events or outcomes may differ significantly from those contemplated herein.

It should not be assumed that either I or Timeless Partners has made or will make investment recommendations in the future that are consistent with the views expressed herein. We may make investment recommendations, hold positions, or engage in transactions that are inconsistent with the information and views expressed herein. Moreover, it should not be assumed that any security, instrument, or company identified in the newsletter is a current, past, or potential portfolio holding of mine or of Timeless Partners, and no recommendation is made as to the purchase, sale, or other action with respect to such security, instrument, or company.

Neither I, nor Timeless Partners, make any representation or warranty, express or implied, as to the accuracy, completeness or fairness of the information contained in this newsletter and no responsibility or liability is accepted for any such information. By accessing this newsletter, the reader acknowledges its understanding and acceptance of the foregoing statement.