AI Power Consumption: Top 1% Companies' Spending Is 8 Times That of the Top 10%, a16z Breaks Down the H1 2026 Market With Hundreds of Charts

TechFlow Summary: a16z released its H1 2026 Market Outlook report, using over 100 charts to reveal a harsh reality—the gap in AI investment between tech companies has widened nearly eightfold. This report not only breaks down the path where hardware replaces software as the market's new protagonist but also directly addresses the two most heated debates of the year: "Will GPUs rapidly depreciate?" and "Is SaaS dead?" For investors focused on AI narratives and US equities, this is a key reference for determining where capital will flow in the second half of the year.

H1 2026 Market Outlook: 100+ Charts

We are pleased to release the second edition of the a16z Market Outlook report.

If you remember the first issue (if not, that's okay), you know what to expect from the second: a feast of charts for H1 2026 viewed through the lens of stock markets and technology.

View all charts

Below is just a small preview.

Technology Is the "Cycle of Everything"

As a force in capital markets, technology truly took off after the financial crisis. Following the bursting of the real estate bubble and credit tightening, technology offered investors wary of heavy assets a high-growth, light-capital alternative. Perhaps more importantly, given massive untapped market space and near-zero marginal costs for software, technology provides enormous embedded operating leverage. It's clear in hindsight, but investors who continually doubted whether loss-making tech companies could ultimately evolve into highly profitable machines missed the defining theme of the past decade.

While it's true that software ate the world, technology's core position in capital markets has elevated to another level:

Technology continues to compound earnings growth. Despite a lower base, it holds a clear advantage over other sectors. Since 2023, one could argue that technology is essentially the story of earnings growth itself—through late August 2026, tech accounted for approximately 76% of total earnings growth for the S&P 500.

The story behind this earnings growth is fascinating in itself, but you can no longer view technology merely as a single sector. Historically, durable goods defined cycles—homes, dishwashers, cars, and so on—but technology has taken their place. Technology is ubiquitous and permeates everything. Now, technology is the cycle of everything.

From Bits to Atoms

Within the tech sector, the theme this year (and beyond) is the rotation from bits to atoms.

Software dominated the last tech cycle, and hardware is now taking center stage, which is easy to understand:

The AI build-out boom has triggered surging demand in traditionally sluggish, cyclical, capital-intensive industries like semiconductors (as well as power and networking equipment). This demand is largely underwritten by historic, massive profits generated by the world's largest tech companies—essentially converting the free cash flow of hyperscale data center operators into the free cash flow of semiconductor firms—though the share of debt financing is also rising.

However, AI isn't the only catalyst reviving the atom economy. Global infrastructure demand runs into the trillions, defense spending is rising, grids are adapting to the electrification of everything, manufacturing is reshoring, and robotics and robotaxis are on the horizon.

Regardless, hardware and infrastructure have become the market's darlings after years of lagging behind software. Both public markets and private equity are investing with unprecedented vigor into innovation across computing, storage, power, robotics, manufacturing, and defense. The bottom line: atoms are back.

The Claim That GPUs Are Rapidly Obsoleted Is Severely Overstated

One thing is clear regarding this capex: demand for compute still outpaces supply.

At least one prominent bear of the build-out expresses skepticism: Given that GPUs become obsolete in three to four years, is deploying all this capital really worth it? No matter how well the Nvidia B200 sells, what does that mean for the A100s installed one or two years ago?

At least for now, the fact that the AI compute demand curve continues to trend upward means the A100 remains highly useful:

Rental prices (and GPU salvage values) should theoretically decline over time, but they haven't. As intelligence becomes cheaper, demand for compute only rises, pushing up prices for the latest chips (while older ones remain resilient). Even A100 prices hold at or above their early-year levels.

The story is far from over, but so far, compute and model advancement are not zero-sum games. Better, cheaper intelligence is creating value across the ecosystem, leaving old chips and models holding substantial value even past the expiration dates cited by bears.

Besides, all of this unfolds against a backdrop where AI adoption remains relatively immature. Adoption is broad, but shallow in most cases:

While nearly 30% of S&P 500 companies reported some "quantifiable impact" from AI, only about 2% track any specific metrics. The same applies to agentic use cases, with only a tiny fraction of user bases actually deploying agents at scale.

The consumer side looks similar, with paid penetration remaining low:

As of April, only about 2% of US households were paying for an AI service. This number is higher and growing now, but remains minuscule in the grand scheme.

The key takeaway: GPUs are running at full capacity, and data shows mature AI adoption and utilization have just begun.

SaaS Must Prove Its Worth, Rather Than Face Apocalypse

Earlier this year, voices claimed software was dead, arguing that AI generates code natively, making software the inevitable victim. The SaaS apocalypse is coming. Run for your lives, enterprise SaaS, and don't let the door hit you on the way out.

The reality is often more nuanced. There was indeed a sell-off, but it was far more selective than an "apocalypse" would imply, and while AI (or the threat of AI) is certainly playing a role, it is not the sole driver.

The truth about public market software companies is that certain parts of this liquidation—or rather, repricing—had been brewing for a while:

Since the end of the zero-interest-rate era, tech companies (and beyond) have traded growth for profitability. In 2022, the market was flooded with high-growth but mostly unprofitable software companies. By 2026, the dynamic has flipped: roughly 75% of these companies are profitable, but only about 30% are growing at rates above 20%.

This makes sense, as higher interest rates inherently make capital relatively scarcer. Companies wisely shifted from loss-funded growth to more sustainable, albeit slower, steady growth. That's fine, but slowing-growing companies cannot sustain high-growth multiples—at least not indefinitely. Eventually, this new "low-growth" norm caught up with the software sector.

Naturally, not every company is affected. High-growth firms' valuation multiples remain in line with historical averages (though nowhere near zero-rate-era peaks), but there are simply far fewer such companies today, leading to a broad repricing of the entire sector.

The software industry didn't face an apocalypse, but it did experience a "show me" moment.

Looking Ahead: Possible Trajectories

Finally, although most of this report looks backward, it would be our dereliction of duty to withhold some perspectives on the future:

We expect AI to expand the reach of demand. This implies maturing adoption on both the enterprise and consumer sides, while simultaneously expanding into entirely new domains like robotics, biotech, health, and autonomous driving.

Overall, technology is advancing at an exponential pace. While no one can predict the future, given the speed and rhythm of change, we are fairly confident this cycle will differ from any before it.

Join the official Coincamps community:

X: https://x.com/coincamps

Telegram: https://t.me/coin_camps