Software Spending Rose 15.5%—Market Cap Fell $1 Trillion

Last Issue’s $100 Was Just $12.90 of Every $100 in IT Spending

Last issue, we looked at BNP Paribas’s estimate that every $100 of AI capex splits into $50 for chips, $20 for power, $15 for networking, $7.5 for cooling, and $7.5 for facilities. Put that same $100 inside a bigger wallet, and its size looks different. In a forecast Gartner released on July 27, 2026, global IT spending this year comes to $6.369 trillion, of which data center systems account for $822 billion. Converted to the same $100 scale, that’s $12.90. Out of that same $100, software takes $23.10 and IT services take $24.70.

AI Data Center Costs and Equipment Replacement Cycles · Issue 236 · INLEVEL9Drawing on BNP Paribas’s investment-allocation chart, this issue examined chip, network, power, cooling, and facility costs—and how they differ across businesses.inlevel9.com

The two $100 baskets overlap, but they aren’t the same. BNP’s map covers the physical assets hyperscalers build, so it includes buildings and power infrastructure. Gartner’s data center systems category is centered on equipment like servers and storage. What overlaps is mostly chips, servers, and networking gear.

This issue looks at the layer above that. Hardware is bought once and used for years, but software is paid for annually as a subscription. Companies are spending more on it—yet the value of the companies collecting that money has been marked down.

---

Software Spending Rose $197 Billion This Year Alone

Here’s what Gartner’s table shows as-is. Software spending rises 15.5%, from $1.271 trillion in 2025 to $1.468 trillion in 2026—a gain of $197 billion. In the same table, data center systems climb 62.5%, from $506 billion to $822 billion, an increase of $316 billion.

The growth rate is about four times faster on the data center side. But in absolute dollars, software is still 1.8 times bigger. That means companies are spending more on software this year than they are on buying servers and storage. It’s a different picture from the narrative that the market has shrunk.

There’s one thing to be careful about when citing these numbers. Gartner revised its forecast for the same year three times within six months. In February, the total was $6.15 trillion with software growth of 14.7%; in April, it was $6.31 trillion, with software at $1.44 trillion and 15.1% growth; in July, it became $6.369 trillion, with software at $1.468 trillion and 15.5% growth. So if you’re going to write “the 2026 software market is $1.44 trillion,” you need to attach a tag noting it’s the April forecast. In issue No. 181, I wrote that an $800 billion market-size figure swings depending on the definition used. Here, it’s the same institution, the same line item, swinging quarter to quarter.

February 3rd, the Same Market Got Repriced in Two Days

In the year spending went up, the stock market went the other way. On February 3, 2026, Anthropic unveiled a set of plugins for Claude Cowork — tools that bundled together work people used to open a screen and do by hand: contract review, NDA classification, compliance procedures. From that day, Thomson Reuters, LegalZoom, and Salesforce took heavy hits, and the selloff spread to Asian IT firms like Tata Consultancy Services and Infosys. The S&P 500 Software & Services Index dropped more than 4% in a single day on February 5th, its eighth straight day of losses, and by that point it was down roughly 20% year-to-date. A trader at Jefferies gave the scene a name: SaaSpocalypse¹.

The size of the loss depends on which source you read. Figures range from $285 billion over 48 hours, to roughly $1 trillion cumulative in Q1, to about $1.6 trillion over the first two months of the year (based on software-and-services ETFs, as cited by the Wall Street Journal), to as much as $2 trillion over the trailing 12 months. The discrepancy comes from which time window you measure and which basket of stocks you count. Pick any one of these four numbers and write “the software market lost $2 trillion,” and you’ve conflated a change in market capitalization with a decline in revenue. Revenue, meanwhile, kept growing the whole time.

Growth rates differ by 1.7x, but valuations differ by 2.9x

Even within the same software universe, the results diverged. Venture investor Tomasz Tunguz sorted 87 publicly traded software and platform companies into five buckets on July 13, 2026, and here’s what the numbers showed.

Bucket	# of Stocks	1-Year Return	Revenue Growth	Revenue Multiple (EV/Sales)2
Infrastructure & Dev Tools	13	+68.5%	21.4%	10.0x
Security	Multiple	+17.6%	~21%	Not disclosed
Business Applications	48	-36.2%	12.5%	3.4x

What stands out here is the difference in the size of these two gaps. Revenue growth rates were 21.4% versus 12.5% — a 1.7x gap — but the multiples the market assigned were 10.0x versus 3.4x — a 2.9x gap. The valuation gap widened far more than the growth-rate gap did. Tunguz concluded that what determined winners and losers wasn’t growth rate, but which category a company sat in. The 48 stocks that include Salesforce, Workday, and ServiceNow make up the majority of publicly traded software companies, and this bucket has the most contracts priced by seat count.

The rebound was narrow, too. According to Meritech’s tally, the market-cap-weighted software index rose 38% in 34 trading days from its April 10 low, but roughly 80% of the recovered market cap went to just 10 companies. The multiples of those top 10, measured against next-twelve-months revenue³, rose from 9.9x to 14.8x, while the median company managed only a 19% gain from the low, landing at 3.3x. That’s why the sentence “the index recovered” and the sentence “my stock recovered” mean different things.

The index itself isn’t singular, either. The SaaS Capital Index’s median multiple fell from 16.9x in August 2021 to 5.6x in December 2025, hit a post-2011 low of 3.2x in June 2026, then climbed back to 3.8x by July 31. Meanwhile, the median multiple tracked by Aventis Advisors stood at 4.6x in August 2026. These two numbers aren’t a contradiction — they’re separate indices with different constituents and different calculation methods. String them together into a single narrative, and you invent a rebound that never happened.

Applying Last Issue’s Three Questions to Software

At the end of issue 236, I laid out three questions to ask when evaluating AI beneficiary stocks: which box does it sit in, how long will it stay in use, and is the bottleneck on the supply side or the demand side. Carry those same questions over to software, and here’s what you get.

Question	Infrastructure (last issue)	Software (this issue)
Which box?	AI chips at $25 or optical transceivers at $5.5	The infrastructure/security box, or the business-app box priced by headcount
How long is it used?	Servers for 5 years, or transformers for 30	A system of record that halts operations if removed, or a screen-level feature a model could replace
What’s the bottleneck?	Transformer lead times and manufacturing capacity	The cost of migrating data and workflows, plus changes in headcount

The middle question is the crux of what happened this time. For physical equipment, lifespan is set by physical wear and accounting policy. For software, lifespan is set by switching cost. Software where data and processes are deeply embedded—systems of record, security, identity—can’t be ripped out in a single quarter, even when a replacement tool appears. Thin, screen-level functions that a person used to operate manually, on the other hand, see their contracts shrink right along with that person’s workload. What the market cut in February wasn’t future revenue itself—it was the assumption about how many years that revenue would last. That said, this distinction is one I’ve drawn from the data I reviewed; contract terms vary by company, so it doesn’t map cleanly onto every individual stock.

For readers in Korea, there’s one more question to add. In Gartner’s table, IT services is a $1.57 trillion box—bigger than software—yet its growth rate, at 5.3%, is the lowest of any category. That’s also why the February sell-off spread even to Indian IT firms. A business model built on billing for people’s time got hit with the same question as seat-based contracts. Domestic Korean companies, too, tend to weight their software budgets more heavily toward implementation and operations than toward off-the-shelf subscriptions—so the first step is to break down our own budgets into what goes to seats versus what goes to people’s time.

Oswarld’s Lens

In the first draft of this piece, I cited the SaaS Capital Index and lined the multiples up side by side: 16.9x in August 2021, 5.6x at the end of 2025, 3.1x in June 2026, and 4.6x in August. When I checked again, though, the first three numbers were from SaaS Capital — but that last 4.6x actually belonged to a different index, from Aventis. And the trough wasn’t 3.1x either; it was 3.2x. Within a single index, the move was from 3.2x in June to 3.8x in July. But splice two different indices together and it looks like a much bigger jump — from 3.1x to 4.6x. The same problem I ran into last issue, reconciling BNP’s and Epoch’s classification standards, turns up again here — this time in the valuation data.

There’s one more thing I fixed from the draft. I had written that of every $100 a company spends on software, $40 goes to core SaaS — but phrased that way, it reads as if that $40 will still be the same $40 next year. That’s true enough as this year’s cash outflow, but what this whole episode is actually about is that the lifespan of the contracts inside that $40 is starting to split apart. So in this issue, I look at spending through Gartner’s sourced tables, and at that split through data covering 87 companies.

As I see it, what got sold off in February wasn’t software revenue — it was the permanence of seat-based contracts. Infrastructure spending goes out the door first, and software sits on the repayment side; but when the repayment mechanism is tied to headcount, it shrinks the moment headcount does. That doesn’t mean the market’s judgment is correct, of course. In Tom Tunguz’s data, the 48 stocks that took the biggest hit are still growing revenue at 12.5%.

Closing

This year, software spending rose 15.5% to $1.468 trillion, with $197 billion of that being pure new growth. In that same year, market capitalization vanished — anywhere from $285 billion to $2 trillion, depending on how you measure it. Both statements can be true at once because spending and valuation measure different things. Spending measures the cash going out this year; valuation measures how many more years that cash is expected to keep coming in.

So when we look at software, last issue’s three questions still hold up. Which quadrant is it in? Does removing it stop the work? And is what’s shrinking demand, or headcount? If you can answer these three, it also becomes clear why “software is over” and “software is still huge” can both appear in the same year.

---

💬 Among the software contracts Reader‘s company uses, which line items would stay the same price even if headcount drops next year? And which ones would shrink along with headcount — I’d love to hear both.

📨 If you have a colleague arguing in meetings over “SaaS is dead” versus “the software market keeps growing,” send them this piece.

References & Further Reading

Primary sources

• Gartner, “Gartner Forecasts Worldwide IT Spending to Grow 14.2% in 2026, Totaling $6.37 Trillion”, July 27, 2026. Link ··· This is the source table with $1.468 trillion in software spending and $822 billion in data center systems. It lays out 2025 actuals and 2026 forecasts side by side, so you can calculate the increase directly.
• Gartner, “Gartner Forecasts Worldwide IT Spending to Grow 13.5% in 2026, Totaling $6.31 Trillion”, April 22, 2026. Link ··· This is the widely-cited April forecast, showing $1.44 trillion in software spending and 15.1% growth. Placed next to the July table, it shows exactly how the forecast shifted.
• CNBC, “AI fears pummel software stocks: Is it ‘illogical’ panic or a SaaS apocalypse?”, February 6, 2026. Link ··· Original reporting on the February sell-off — which stocks and indices fell how much, and how the panic spread to Asian IT companies.
• Tomasz Tunguz, “The CIO’s Choices are Clear in 2026”, July 13, 2026. Link ··· This splits 87 public companies into five buckets, pairing returns, growth rates, and revenue multiples. It’s the source for this issue’s table.
• Alex Clayton, Austin Wang, “2026 SaaS Rebound”, Meritech Capital, July 2026. Link ··· A tally showing the rebound was concentrated in just 10 stocks — it illustrates the gap between index-level recovery and individual stock recovery.

Background

• SaaS Capital Index explainer page, as of July 31, 2026. Link ··· Explains the source and design (equal-weighted median) behind the 5.6x figure from December 2025, 3.2x from June 2026, and 3.8x from July.
• Aventis Advisors, “SaaS Valuation Multiples: 2015-2026”. Link ··· A separate index that produced the 4.6x median from August 2026. Its constituents differ, so it shouldn’t be compared directly with the SaaS Capital Index.

Related past issues

• What AI Data Centers Cost, and How Often Each Component Gets Replaced ··· Issue 236. This is the $100 map that this issue picks up from.
• $800 Billion: The World’s Most Famous Wrong Number ··· Issue 181. Covers why the numbers diverge for the same market.
• The SaaS Stock Crash Wasn’t Really About AI ··· Issue 23. This is the issue that pinned down the cause of the March decline.

The author is Oswarld (Kwangseob Ahn). Current roles: Adjunct Professor at Sejong University, Strategy Consultant at INLEVEL9. Career, research, books, and recent work are kept current on the About page. Latest · July 2026: HEMA-2: A Consolidation-Aware Tri-Memory Architecture with Multi-Channel Scheduling for Lifelong Conversational AI.

📝 Glossary

Footnotes

1. SaaSpocalypse: A term coined by a Jefferies trader for the stretch in early 2026 when publicly traded software stocks fell one after another. It’s a portmanteau of SaaS and apocalypse — not an official statistical term, just a nickname used in the market. ↩

2. EV/Sales (revenue multiple): Enterprise value divided by annual revenue. It shows how many times the market is valuing the same revenue at, and this figure can move even when revenue stays flat, if expectations shift about whether that revenue will hold up going forward. ↩

3. NTM (next twelve months): A method of calculating the multiple based on expected revenue over the coming year, rather than the trailing 12 months of actual results. The two methods produce different values, so they shouldn’t be mixed when making comparisons. ↩