Sarah Guo and the Price of Conviction

In October 2022, Sarah Guo left behind one of Silicon Valley’s most established investment institutions to build a venture capital firm around a single technological conviction: artificial intelligence would fundamentally change the software industry.

Her new firm, Conviction, launched with approximately $101 million. A month later, OpenAI released ChatGPT, transforming artificial intelligence from a technology largely understood by researchers and specialists into a global commercial phenomenon.

The timing was remarkable. But timing alone does not explain Guo’s investment strategy.

While some of the industry’s largest venture firms were expanding into multibillion-dollar investment platforms, Guo was pursuing a different model. Conviction would concentrate on relatively few companies, specialize in a rapidly developing technology and work closely with the founders it backed.

Four years later, several of those investments have become some of the most valuable private companies in artificial intelligence.

The more interesting question is whether Guo has demonstrated something larger than an ability to recognize a technological transition.

Can concentrated investment judgment still create an advantage in a venture capital industry increasingly defined by scale?

The Business of Being Early

Venture capital has always rewarded investors who recognize important changes before they become obvious.

But identifying a technological revolution and identifying the companies that will profit from it are different problems.

The internet created enormous wealth while destroying countless businesses built on reasonable assumptions about its future. Mobile computing produced similar outcomes. Cloud infrastructure transformed enterprise technology, but not every company that recognized the transition became a successful investment.

Artificial intelligence presents an even more complicated challenge.

Its capabilities are improving rapidly, its costs are changing, and the boundaries between infrastructure providers, foundation-model developers and application companies remain unsettled.

Guo’s decision to specialize in AI reflected a belief that these changes would invalidate some of the assumptions investors had developed during the previous generation of software businesses.

At Greylock, where she spent nearly a decade, Guo had invested across infrastructure, security, enterprise software and other technology markets. By the time she established Conviction, she believed AI represented a sufficiently important departure to justify building an entire investment organization around it.

In her founding announcement, she described the opportunity as the emergence of Software 3.0—a generation of companies in which intelligence would become a fundamental component of software rather than an additional feature.

The distinction matters.

Traditional software automates processes that developers explicitly define. AI systems increasingly perform tasks that previously required human interpretation, reasoning or judgment.

That creates opportunities in markets once considered difficult for conventional software businesses, including law, medicine and other knowledge-intensive professions.

But recognizing that possibility in 2022 required accepting substantial uncertainty about the technology’s reliability, commercial adoption and eventual economics.

Guo was not alone in recognizing AI’s potential. Researchers, founders and investors had been pursuing machine learning for years.

Her distinguishing decision was to organize her investment business around the thesis before its commercial implications became widely accepted.

A Portfolio Built on Concentration

By May 2026, Conviction had invested in 27 startups over approximately three years, according to Forbes.

Among Guo’s investments were Harvey, which develops AI tools for legal professionals; Sierra, which builds AI systems for customer service; Cognition, which develops AI coding technology; OpenEvidence, which serves medical professionals; Mistral, a foundation-model developer; and Baseten, an AI infrastructure company.

Forbes reported that those six businesses had a combined valuation of approximately $62 billion.

The figure illustrates the scale of the companies Guo helped back. It does not establish the financial return earned by Conviction or its investors.

Private valuations reflect financing transactions and investor expectations. They are not equivalent to realized profits, and they can change substantially before shareholders obtain liquidity.

Nevertheless, the portfolio demonstrates the commercial breadth of Guo’s original thesis.

Rather than investing exclusively in foundation models, Conviction has backed companies operating at different levels of the AI economy.

Some provide infrastructure. Others deliver specialized applications. Still others attempt to change how professionals perform work.

The approach suggests that Guo is not simply betting on increasingly capable models. She is investing in the possibility that intelligence will become embedded throughout the economy, creating businesses with different customers, distribution channels and competitive advantages.

That is a broader proposition than predicting the success of any particular AI laboratory.

It is also more difficult to evaluate.

If foundation models eventually absorb capabilities currently sold by specialized application companies, some of today’s apparent advantages could disappear.

If the costs of operating AI systems remain high, revenue growth may not translate into the profitability investors expect from software.

And if competition drives prices downward faster than costs decline, even rapidly expanding markets may produce disappointing returns.

The question is not whether AI will become important.

It is who will capture the value it creates.

The Advantage Money Cannot Automatically Buy

The modern venture capital industry has developed an increasingly institutional structure.

Large firms can provide founders with access to extensive networks, recruiting resources, technical specialists, business development teams and substantial follow-on financing.

These capabilities can be valuable, particularly for companies competing in capital-intensive markets.

Yet scale introduces its own complications.

A firm managing enormous amounts of capital must deploy that capital across enough opportunities to produce meaningful returns. It may also need to participate in increasingly large financing rounds or invest across multiple stages of company development.

An individual early-stage investment that transforms the economics of a small fund may barely affect the performance of a much larger one.

Guo’s approach raises a different possibility: that a specialized investor can compete by making fewer decisions and devoting more attention to each.

In an April 2025 interview with Fortune, she emphasized the importance of trust and understanding in founders’ decisions about investment partners, questioning whether entrepreneurs necessarily prefer the backing of institutions managing tens of billions of dollars.

Her argument does not imply that large venture firms are ineffective. Nor does it mean that a smaller organization automatically makes better decisions.

It suggests that venture capital contains two businesses that are sometimes treated as one.

The first is allocating money.

The second is helping build companies.

Money can be raised, pooled and deployed at increasing scale. Judgment, trust and individual attention are harder to expand without changing their character.

Guo’s investment in Baseten illustrates the distinction.

She first backed the AI infrastructure company in 2019, years before the commercial market for generative AI accelerated. According to Forbes, she subsequently helped the company retain a customer and supported its financing efforts.

Harvey’s founders have similarly credited her with helping establish an important early customer relationship.

These examples do not prove that Conviction will outperform larger competitors. But they illustrate why a founder might value an investor for reasons extending beyond the size of a check.

For an early-stage company, one important customer or strategic decision can matter more than access to an elaborate institutional platform.

The Problem With Conviction

There is an uncomfortable mathematical reality behind concentrated investing.

The fewer companies an investor backs, the more each decision matters.

Diversification allows venture funds to absorb failures while retaining exposure to unexpected successes. Concentration increases the consequences of both correct and incorrect judgments.

An investor who backs 100 companies can afford to misunderstand many of them if a handful produce extraordinary outcomes.

An investor who backs 20 must be more selective—or more fortunate.

Concentration can also make early performance difficult to interpret.

During a powerful technological expansion, several investments may rise in valuation simultaneously because investors increasingly favor the entire sector.

The apparent success of individual company selection may therefore reflect a combination of genuine insight, favorable market conditions and the broader repricing of technological opportunity.

Separating those effects requires time.

Guo’s early AI investments have attracted considerable attention, and her appearance on Forbes’ 2026 Midas List reflects that recognition.

But the ultimate test of a venture strategy is not how much its portfolio companies are worth during a financing boom.

It is how much capital the strategy returns over a complete investment cycle.

That distinction becomes especially important when companies remain private for extended periods.

A venture fund can report substantial paper gains while its investors have received relatively little cash. Some valuations may eventually be justified by exceptional business performance. Others may not.

Without complete fund-level performance information, it is premature to conclude that Conviction’s concentrated approach has produced superior realized returns.

There is another complication.

Conviction’s strategy depends partly on Guo’s own expertise, relationships and capacity to engage with founders.

Those qualities may be valuable precisely because they are difficult to reproduce.

But that also raises questions about how the organization can grow without diluting what made its approach distinctive.

A strategy built around exceptional individual judgment can become harder to maintain as the number of investments, employees and responsibilities increases.

The constraint is not necessarily capital.

It may be attention.

When the Market Changes the Rules

Guo’s investment philosophy contains an apparent contradiction.

She has built her firm around a strong belief in AI while repeatedly emphasizing the importance of questioning established assumptions.

The contradiction is only superficial.

Conviction about a technological transition does not require certainty about which companies will succeed within it.

Indeed, the faster a technology develops, the more dangerous rigid assumptions about individual business models may become.

Consider the changing economics of software.

For decades, successful software companies benefited from relatively low marginal distribution costs. Once a product had been developed, additional customers could often be served without proportionate increases in operating expenses.

AI applications can have a different cost structure.

Generating answers, processing documents and executing complex tasks may require substantial computing resources each time a customer uses the service.

Those costs may decline as hardware and software become more efficient. But the competitive implications are not straightforward.

Lower inference costs can improve margins. They can also make it easier for new competitors to enter a market.

Better foundation models can improve specialized applications. They can also make some of those applications less differentiated.

A legal AI company, for example, may develop valuable customer relationships, proprietary workflows and integration with professional systems.

But it must continue demonstrating why customers should purchase its product rather than use increasingly capable general-purpose AI tools.

The same challenge applies to customer service, coding and other emerging categories.

For Guo, the investment question is therefore more demanding than identifying where AI can perform useful work.

It is identifying where companies can build lasting economic advantages as the underlying technology improves.

That may prove to be the defining challenge of AI venture capital.

The Value of Being Wrong

One revealing feature of Guo’s approach is her willingness to discuss the possibility that established investment assumptions no longer apply.

In a September 2026 conversation on the Invest Like the Best podcast, she discussed investing without a reliable historical model, the importance of computing infrastructure and the difficulty of identifying non-obvious AI opportunities.

These questions expose a limitation shared by nearly every investor attempting to understand a technological transition.

Historical data can explain what previously worked. It cannot reliably establish which economic structures will emerge from a technology that changes the underlying conditions.

The companies that appear strongest today may not occupy the most valuable positions tomorrow.

An infrastructure provider can become indispensable—or face commoditization.

An application company can establish a durable relationship with customers—or discover that its core functionality has become a standard feature elsewhere.

A model developer can build an extraordinary technical advantage—or find that maintaining it requires continually increasing investment.

Conviction, in this environment, should not mean refusing to reconsider an investment thesis.

It should mean developing a sufficiently clear understanding of the opportunity to recognize when the evidence changes.

That distinction separates disciplined investment judgment from simple confidence.

It also explains why specialization can be valuable.

An investor who spends years studying one technological field may develop a better understanding of its technical constraints, commercial opportunities and emerging competitors.

But specialization can create blind spots of its own.

Deep familiarity with a sector may encourage an investor to interpret every development as confirmation of its importance.

The challenge is maintaining enough conviction to act before consensus forms, without becoming so committed that contradictory evidence loses its influence.

What Conviction Is Really Testing

Sarah Guo’s career offers an interesting counterpoint to the assumption that the future of venture capital belongs exclusively to institutions with the most money.

Her firm has demonstrated that a focused investor can establish meaningful positions in companies operating at the center of a major technological transition.

It has also shown that founders may value specialized knowledge and direct involvement alongside financial resources.

What remains unproven is whether those advantages translate into consistently superior investment returns.

The answer will depend on outcomes that cannot yet be fully measured: the durability of AI business models, the prices paid for ownership, the eventual liquidity of private investments and Conviction’s ability to maintain its approach over time.

There is no reason to assume that small venture firms will replace large ones.

The two models may serve different purposes.

Large institutions can finance companies through expensive growth phases, provide substantial resources and support ambitious projects requiring billions of dollars.

Specialized investors can identify emerging opportunities, develop close relationships with founders and make decisions that would be economically insignificant to larger funds.

The more interesting possibility is that technological change increases the value of both models at different stages of company development.

Guo’s strategy is a test of where independent judgment fits into that structure.

Four years after Conviction’s founding, the artificial intelligence revolution she anticipated is no longer a contrarian proposition.

Capital is flowing into the industry, competition is intensifying, and the number of investors claiming expertise in AI has expanded dramatically.

Being convinced that artificial intelligence matters is no longer particularly distinctive.

Determining which companies will capture its value remains extraordinarily difficult.

And that may be the real price of conviction.

The investor must be willing to make a consequential decision before the evidence is complete—and remain willing to change that decision when the evidence changes.