Models depreciate while data appreciates - the net rate determines the investment type.
You provision $360K in GPU servers for your ML pipeline. Finance capitalizes the purchase and tells you it won't touch EBITDA. Eighteen months later, a new chip generation cuts your servers' market value in half - but your Balance Sheet still shows them at $240K Book Value. They are really worth $120K. That $120K gap between what your Financial Statements claim and what the market would actually pay is economic Depreciation outrunning the accounting schedule - and it quietly distorts every ROI calculation you make downstream.
Depreciation is the systematic reduction of a Capital Asset's Book Value over time. The accounting version is a schedule you choose; the economic version is a verdict the market imposes. For tech Operators, the Net Rate between depreciating models/infrastructure and appreciating data determines whether a combined investment is a Compounder or a Wasting Asset.
Depreciation is the systematic reduction of a Capital Asset's Book Value over time. It operates on two levels that Operators need to keep separate.
Accounting depreciation is a policy choice. When you record a purchase as a Capital Asset on the Balance Sheet (instead of expensing it directly on the P&L), you spread the cost across the period you expect to use the Asset. A $120K server rack depreciated over 3 years puts $40K/year on your Operating Statement as a Depreciation expense - not the full $120K in year one.
Economic depreciation is a market force. It is the rate at which an Asset's market value actually declines - driven by wear, Obsolescence, or Competitive Erosion. Your accounting schedule says the servers lose value evenly over 36 months. The market might disagree sharply when a new chip generation ships at month 18.
Amortization is the same concept applied to intangible Capital Assets - software, licenses, acquired technology. Same mechanics: spread the cost, reduce Book Value each period.
The key distinction: accounting depreciation is a number you choose. Economic depreciation is a number the market imposes. When they diverge, your Balance Sheet stops reflecting reality.
Depreciation occupies a unique position on the Financial Statements. It is an expense on the P&L, but it gets added back when calculating EBITDA. This means the choice between capitalizing a purchase (recording it as a Capital Asset that depreciates over time) versus expensing it (hitting the P&L immediately) directly changes your EBITDA number - even though the total Cash Flow impact is identical.
For PE-Backed businesses where Valuation is often a multiple of EBITDA, this is not an accounting footnote. It is a lever that changes what the business appears to be worth.
But the Operator's deeper risk is not in the accounting treatment. It is in the gap between Book Value and market value. If your Capital Assets are depreciating economically faster than your books reflect, you are carrying phantom value on the Balance Sheet. You think you have $240K in Assets. You actually have $120K. Every decision built on that inflated number - Capital Allocation, ROI underwriting on new investments, even Budget planning - rests on a bad input.
This problem compounds in tech businesses. Physical Capital like production lines depreciates on a relatively predictable curve. Custom software, ML models, and specialized infrastructure face Obsolescence risk that is lumpy and fast. The book schedule almost never captures it.
The Accounting Mechanics
The simplest method: take the purchase price, subtract whatever value you expect the Asset to retain at the end of its useful period, and divide evenly across the years you plan to use it.
Annual Depreciation = (Purchase Price - Expected Remaining Value) / Years of Use
Example: $120K of servers, 3-year expected use, $0 expected remaining value.
P&L vs Balance Sheet vs EBITDA
The cash leaves your account on day one (-$120K Cash Flow). But the P&L absorbs the cost gradually (-$40K/year as Depreciation expense). And EBITDA? It adds Depreciation back - so the $120K capital purchase has zero EBITDA impact across all three years.
This is why Capital Investment decisions deserve more scrutiny than their EBITDA appearance suggests. A business can spend aggressively on Capital Assets, show strong EBITDA, and be bleeding Cash Flow the entire time.
Economic Depreciation: The Tech Operator's Problem
Tech assets face Obsolescence risk that makes actual depreciation schedules lumpy and hard to predict:
Where the Net Rate Framework Applies
Here is where the prerequisites converge. While your models and infrastructure depreciate, the data you collect by running them can be an appreciating Knowledge Asset. Your ML model depreciates. Your proprietary training data and customer interaction logs appreciate - they form a Data Moat that competitors cannot replicate cheaply.
The Net Rate between these two forces determines the investment type:
Before any Capital Investment: Estimate the economic depreciation rate, not just the accounting schedule. How fast will this Asset lose real market value? For tech assets, default to assuming Obsolescence arrives faster than your plan. If economic depreciation exceeds the value the Asset generates over its life, you are building a Wasting Asset before you start.
When evaluating P&L performance: Check whether strong EBITDA is masking heavy Capital Investment spending. Depreciation gets added back to show EBITDA - but look at actual Cash Flow to see the complete picture. A business showing $2M EBITDA with $1.8M in annual Capital Investment is not as healthy as the EBITDA line suggests.
When assessing your tech portfolio: Map each major Capital Asset on two dimensions: its book depreciation schedule and your honest estimate of economic depreciation. Where the gap is large, you are carrying inflated Book Value. Where economic depreciation is slow (or the Asset actually appreciates, as with proprietary data), you may have undervalued assets that your Balance Sheet understates.
When the Net Rate turns negative: If your depreciating infrastructure and models are losing value faster than your data and Knowledge Assets are gaining it, the investment thesis has broken. This is the signal to stop defending the Depreciating Asset and redirect Capital Allocation toward the component that is actually appreciating.
You purchase $360K of GPU servers for your ML training pipeline. Finance records them as a Capital Asset and depreciates them over 3 years with $0 expected remaining value. Annual Depreciation: $120K/year. Monthly: $10K.
After 18 months, Book Value = $360K - (18 x $10K) = $180K. Your Balance Sheet shows $180K in server Assets.
At month 18, a new GPU generation ships. Your servers now sell on the market for $90K - half of Book Value. Economic depreciation has outpaced the books by $90K.
P&L impact through month 18: $180K in cumulative Depreciation expense. EBITDA impact: $0 (Depreciation is added back). Cash Flow impact: -$360K (all on day one, 18 months ago).
Your Balance Sheet claims $180K in Assets. The market says $90K. Every ROI calculation using the book number overstates your position by $90K.
Insight: Book depreciation is a schedule. Economic depreciation is a verdict. For tech assets facing Obsolescence, the verdict almost always arrives ahead of schedule. The Operator who tracks only Book Value will miss the moment their Asset becomes a Wasting Asset.
You invest $240K building a custom ML model for product recommendations (capitalized over 3 years at $80K/year Depreciation). You also spend $60K/year collecting and curating proprietary customer preference data (expensed on the P&L each year). Evaluate the investment after 2 years.
Model depreciation: Book Value after 2 years = $240K - (2 x $80K) = $80K. But a competitor releases a comparable open-source model. Your model's market value drops to roughly $30K. Economic loss: $210K. The books overstate by $50K.
Data Appreciation: You have invested $120K in proprietary preference data ($60K x 2 years). This data reflects real customer behavior that cannot be replicated without running the same product for the same duration. A potential acquirer estimates replacement cost at $350K+. The data has appreciated roughly $230K above your cost.
Net Rate calculation: Model economic loss = -$210K. Data appreciation above cost = +$230K. Combined Net Rate = +$20K. The investment is slightly positive overall.
EBITDA lens: The model's $80K/year Depreciation was added back to EBITDA each year (zero impact). The data team's $60K/year was expensed and reduced EBITDA by $60K/year. Ironically, the component that hurt EBITDA (data collection) is the one that created real value, while the component invisible to EBITDA (the model) is where value was destroyed.
Insight: The model was a Depreciating Asset. The data was an appreciating Knowledge Asset. The combined Net Rate was positive - making this a Compounder - but only because the data side compensated for the model side. Operators who evaluate tech investments as individual line items miss this. Evaluate the portfolio.
Depreciation is both an accounting schedule (you choose it) and an economic reality (the market imposes it). When these diverge, your Balance Sheet misleads every downstream decision that depends on Asset values.
Capital Asset purchases bypass EBITDA but still consume Cash Flow. A business showing strong EBITDA with heavy Capital Investment can be deteriorating underneath - always check Cash Flow alongside EBITDA.
For tech Operators, the Net Rate between depreciating models and infrastructure versus appreciating data and Knowledge Assets determines whether the combined investment is a Compounder or a Wasting Asset. Track both sides, not just the accounting line.
Treating the accounting depreciation schedule as truth. Book Value and market value diverge constantly for tech assets facing Obsolescence. The books say your servers are worth $180K. The market says $90K. If you only track Book Value, you will miss the moment your Capital Asset becomes a Wasting Asset - and every Budget and ROI calculation downstream will inherit the error.
Celebrating EBITDA improvement from capitalizing purchases without tracking Cash Flow impact. Moving a $500K spend from operating expense to Capital Asset makes EBITDA look $500K better per year (because Depreciation gets added back). Cash Flow does not change at all. In PE-Backed businesses valued on EBITDA multiples, this creates real incentive to capitalize aggressively - which is exactly why sophisticated buyers scrutinize Capital Investment levels relative to EBITDA during M&A due diligence.
You purchase $60K of networking equipment. You plan to use it for 5 years with $0 expected remaining value. What is the annual Depreciation expense? What is the Book Value after 2 years?
Hint: Divide total cost by years of expected use for annual Depreciation. Subtract cumulative Depreciation from the purchase price for Book Value at any point.
Annual Depreciation = $60K / 5 = $12K/year. After 2 years, cumulative Depreciation = 2 x $12K = $24K. Book Value = $60K - $24K = $36K.
Your company can acquire $200K of compute capacity two ways: (A) Buy servers outright - capitalize as a Capital Asset, depreciate over 4 years. (B) Use a cloud provider at $50K/year, expensed on the P&L. Both cost $200K over 4 years. Compare the Year 1 EBITDA impact of each option.
Hint: Depreciation is added back when calculating EBITDA. Regular operating expenses are not. Think about what hits the P&L and what gets reversed.
Option A: Year 1 P&L shows $50K Depreciation ($200K / 4 years). EBITDA adds Depreciation back, so Year 1 EBITDA impact = $0. (Cash Flow impact: -$200K upfront.)
Option B: Year 1 P&L shows $50K operating expense. This is NOT added back to EBITDA, so Year 1 EBITDA is reduced by $50K. (Cash Flow impact: -$50K in Year 1.)
Same total cost. Option A shows $50K higher EBITDA each year. This is one reason PE-Backed businesses may prefer Capital Investment over operating expense - it flatters the EBITDA metric that drives Valuation multiples. But Option B has better Year 1 Cash Flow ($50K out vs $200K out). An Operator who only watches EBITDA misses the Cash Flow difference.
You spend $300K on a custom fraud detection model (capitalized over 3 years = $100K/year Depreciation) and $50K/year on labeled fraud data. After 2 years, an industry vendor releases a comparable product for $30K/year, dropping your model's market value to approximately $20K. Your proprietary labeled dataset would cost a competitor $350K to replicate from scratch. Calculate the combined Net Rate on the investment. Is this a Compounder or a Wasting Asset? What should the Operator do next?
Hint: Calculate the total economic gain or loss on each component separately (model and data), then combine them. Remember that the Net Rate tells you the direction, but the trajectory of each component tells you where to allocate next.
Model: Invested $300K, current market value ~$20K. Economic loss = $280K. (Book Value = $300K - $200K = $100K, overstating reality by $80K.)
Data: Invested $100K ($50K x 2 years), replacement cost $350K. Economic gain above cost = $250K.
Combined Net Rate: -$280K (model) + $250K (data) = -$30K. Currently negative - the combined investment is a Wasting Asset by the numbers.
But trajectory matters: Model losses are capped at $300K total (it cannot lose more than you paid). Data appreciation continues with each additional year of collection. The Operator's move: stop defending the Depreciating Asset. Switch to the $30K/year vendor solution (saving $70K/year vs the original $100K Depreciation rate). Keep investing in proprietary data at $50K/year. Future Capital Allocation compounds on the appreciating side instead of evaporating on the depreciating side. Within a year of the switch, the data gains should push the combined Net Rate positive.
Depreciation is the mirror of Appreciation - where Appreciation describes an Asset gaining value over time, Depreciation describes value lost. Together, they feed into Net Rate, which determines whether a combined investment compounds or wastes. Understanding how an Asset gets classified on the Balance Sheet (from the Asset prerequisite) is essential here because that classification determines whether Depreciation mechanics apply at all - only Capital Assets get depreciated; expenses hit the P&L immediately and are done. Downstream, Depreciation connects directly to EBITDA and EBITDA Optimization because Depreciation is added back when calculating EBITDA, making the boundary between what gets capitalized and what gets expensed a lever on the metric PE-Backed businesses are valued on. It also flows into Book Value (the number Depreciation systematically reduces), the Depreciating Asset and Wasting Asset concepts (what happens when Depreciation dominates), and the Compounder concept (what happens when appreciating Knowledge Assets outpace depreciating Physical Capital). Amortization extends the same mechanics to intangible Capital Assets. For Operators doing Capital Allocation and Capital Budgeting, accurate depreciation estimates - economic, not just book - are inputs to ROI, NPV, and Hurdle Rate calculations that determine which investments deserve the next marginal dollar.
Disclaimer: This content is for educational and informational purposes only and does not constitute financial, investment, tax, or legal advice. It is not a recommendation to buy, sell, or hold any security or financial product. You should consult a qualified financial advisor, tax professional, or attorney before making financial decisions. Past performance is not indicative of future results. The author is not a registered investment advisor, broker-dealer, or financial planner.