Renowned VC a16z: The 'Innovation Methodology' of the Past 75 Years Has Been Completely Rewritten by AI

No one really knows what will emerge, "This is a digital product that humanity has never created before."

Written by: Long Yue, Wall Street Insights

Recently, in a16z's latest podcast, general partners Martin Casado and Erik Torenberg, along with board partner Steven Sinofsky, engaged in an in-depth discussion: What do AI's breakthroughs in mathematics mean? Is the underlying logic of innovation being rewritten? How will the competitive landscape between startups and large companies evolve?

The three discussants believe that the fundamental assumptions that have supported the entire computer industry for the past 75 years are being re-examined one by one by AI. The most critical assumption—that innovation is an engineering problem, not a funding problem—has already failed. AI is rewriting the underlying economic logic of innovation: shifting from engineering constraints to capital constraints, where a 20-person team can now effectively deploy $1 billion. This judgment directly impacts the assessment of the competitive landscape between startups and large companies, the logic of venture capital, and even the boundaries of AI capabilities.

From Abacus to AI: A 75-Year Leap in Abstraction Levels

Sinofsky pulled out a brochure from IBM dated 1953.

On the cover, atomic orbits surround a human head, and the first sentence reads: "Humans took millions of years to recognize the utility of the wheel."

In that brochure, IBM dedicated an entire page to explaining what a "digital computer" is—input, storage, computation, control, and output.

Sinofsky said, "This is how we have understood computers for the past 75 years."

From abacuses to slide rules, from differential engines to personal computers, from graphing calculators to cloud computing—each technological leap is essentially an "upward shift in abstraction levels." Lower-level problems are encapsulated, allowing humanity to solve new problems at higher levels.

When the TI-85 graphing calculator appeared, math teachers collectively panicked: "Our profession is doomed." Sinofsky remarked:

But they did not complain about the advent of calculus, as calculus was already a starting point for them. People's reactions to change are far more intense than their reactions to baseline shifts.

The panic triggered by AI solving mathematical problems is reminiscent of the initial reactions to graphing calculators.

AI Solving Mathematics: Breakthrough or 'Game Master'?

Recently, someone asked Claude to attempt solving the Riemann Hypothesis, sparking widespread discussion in the mathematics community.

In response, Casado poured cold water on the excitement.

Adding up the salaries of postdocs who have researched these problems over the years amounts to a relatively small sum. This means the market has never truly prioritized solving these issues. So I’m not sure that solving these problems proves AI has unlocked something of significant economic value.

He likened AI's mathematical abilities to a StarCraft champion: "This is the strongest StarCraft player in history—very impressive, but I find it hard to directly link it to real economic value."

Sinofsky offered another perspective. He believes that breakthroughs in mathematics might be about "giving birth to new abstract tools."

Just like the proof of the four-color theorem—not about writing a beautiful mathematical derivation, but about using a computer to exhaustively check all finite cases, trading computational power for answers.

Once you have a new level of abstraction, everyone no longer has to start from scratch; they can directly build tools at that level.

The Biggest Paradigm Shift: From Engineering Bottlenecks to Capital Bottlenecks

The core insight of this conversation comes from a thought experiment by Casado.

Twenty years ago, if you gave a 10-person startup $1 billion, they wouldn’t even know how to spend it. Buying servers would quickly deplete the funds.

Ten years ago, giving $1 billion would mean hiring engineers to write code, but the myth of man-months is real—more people can actually slow things down.

Now, giving $1 billion to 20 people, they can effectively spend it.

His conclusion is:

We have transformed this industry from an engineering bottleneck problem to a capital bottleneck problem. This is a fundamental difference. We have never had this state before.

Sinofsky added that this is not the first time there has been a capital bottleneck—during the first 30 to 40 years of the computer industry, it was also a time constrained by capital. "If you wanted to do something with a computer, the first step was to get one."

Then came the era of engineering bottlenecks, and now we are back to capital bottlenecks.

Why Haven't Startups Been Crushed by Giants?

By common sense, Microsoft, Google, and Meta, with their capital, data, and distribution channels, should crush everything.

But the reality is: Cursor, Anthropic, and OpenAI are growing at "meteor speed."

Casado provided two reasons.

First, AI has solved the distribution problem.

In the past, you didn’t know if your marketing budget was effective. Now, the demand for computational power and GPUs is infinite; you can directly decide how much to invest to drive growth.

Second, startups can now raise enough money to truly stand on the same starting line as the giants.

And what about the giants? Sinofsky stated bluntly:

Microsoft is more concerned about what Amazon and Google are doing and doesn’t care about any single startup. Startups don’t directly challenge the giants, and the giants don’t notice the startups.

He also revealed a detail: some large companies prioritize providing computational power to enterprise clients, leaving their internal product teams in a state of "AI famine," while none of their competitors are experiencing such a famine.

Sinofsky used his own experience at Microsoft as an example. When he brought the first Surface to meet Intel executives, their attitude turned cold as soon as they heard it contained an ARM chip.

They thought that was a chip for printers. They only do Moore's Law, just as Google only does ultra-large scale—if AI turns to the edge, that’s not their concern.

Sinofsky summarized:

The essence of disruption is the cultural constants of large companies. Scorecards, sales systems, compensation structures, historical baggage, customer commitments—none of these can change. This is a physical law.

$20 billion thrown in, and no one knows what will come out

At the end of the conversation, the three touched on a deeper unknown.

Casado admitted that he had made a mistake before:

I thought recursive self-improvement and rapid takeoff wouldn’t happen, which is true. But I didn’t expect that we could keep pouring money in infinitely, and that the scale law would always hold.

You take $20 billion and throw it into a model. The two of us watch it and casually test it—I don’t think we can understand what that means. So much computational power, so much data, I don’t know what it can achieve.

Sinofsky agreed and likened it to exponential growth: "No one can model exponentials."

He also provided a specific scenario: the exhaustive combinations of proteins, which used to be an "infinite problem" but can now become a "capital problem."

We can turn previously infinite problems into finite ones through capital. This is a very strange thing.

Casado said this is also why he no longer tries to predict the upper limits of AI capabilities:

I have decided that I cannot predict what something created with $20 billion will be capable of.