Sequoia Capital's Closed-Door Sharing: From 'Delivering Software' to 'Delivering Results', New Logic of AI Commercialization and Capital Pricing

Source: Video of Pat Grady's "AI for BC IC" presentation

Compiled by: AGI 2050

Editor's Note

At the turning points of technological cycles, the internal assessments of leading venture capital firms often constitute key insights into industry trends.

This article is compiled from a closed-door presentation by Pat Grady, a global partner at Sequoia Capital, to the Boston College Investment Committee. In this presentation, Grady set his observational perspective in the fall of 2026, systematically reviewing the structural changes in global underlying technology evolution, industry landing pace, and capital pricing logic since the explosion of large models.

Unlike generalized public discussions, this report, aimed at core limited partners (LPs), focuses on a series of specific and sharp business substance issues:

• Change in Underlying Logic: The essence of this wave is not an information distribution revolution, but a reconstruction of underlying computational architecture, with its potential market space (TAM) substantially penetrating into the professional services market, which is one to two orders of magnitude larger than the trillion-dollar pure software domain;

• Leap in R&D Paradigms: With the maturity of logical reasoning and Long-horizon Agents technology, the research focus of frontier laboratories has shifted from Artificial General Intelligence (AGI) to Artificial Super Intelligence (ASI) and Recursive Self-Improvement (RSI);

• Commercial Differentiation and Diffusion Gap: There is a significant "technology diffusion gap" between the native capabilities of models and their actual applications in enterprises, with high-value knowledge work (programming, healthcare, tax compliance, etc.) giving rise to a new generation of system-level enterprises, while the high-frequency oscillation of underlying technology foundations forces application-layer enterprises to reconstruct their organizational forms;

• New Forms of Capital Bubble: In the midst of fervent market sentiment, the primary market has evolved a two-step financing model of "separating co-building partners from funding partners," with the valuation leverage and premium rhythm of primary and secondary markets undergoing unprecedented tests.

Below is the full text of our compilation of the presentation:

1. This is not an information distribution revolution, but a computational reconstruction

The materials today are primarily used to respond to a proposition raised by the Boston College Investment Committee: "What is the current state of the AI industry?"

Here, I will record these observations as a snapshot at the node of September 2026.

The venture capital industry often uses a chart to illustrate the overlapping evolution of technological waves over the decades: from the evolution of chips at the lowest level to computing systems, connecting to the public through the internet; then entering mobile devices, giving rise to a complex application ecosystem; until now, AI begins to endow these applications with unprecedented processing capabilities.

The basic trajectory of technological evolution has always been like this. However, this wave centered on AI has three characteristics that are distinctly different from previous cycles:

First, the potential market space (TAM) is unprecedentedly large.

In the past few decades, the main target of industry competition has been the "software market size" (Software TAM); whereas AI is directly cutting into the "professional services market size" (Services TAM) ------ the latter is one to two orders of magnitude larger than pure software.

Second, the speed of technology diffusion far exceeds that of the past.

This high penetration rate is not unique to AI technology but is an inevitable result of infrastructure readiness. At the beginning of cloud computing, the global population accessing the internet was less than 100 million; when mobile internet started, users had to purchase smart hardware one by one. Now, with global network infrastructure and smart hardware being highly popularized, once technology breaks through the usability inflection point, the resistance to large-scale diffusion has dropped to a historical low.

Third, and most importantly: the form of this technological transformation is physically different from the past.

The technological revolutions of the past few decades ------ including personal computers, local area networks, the internet, and cloud computing ------ essentially belong to the "information distribution revolution," with the core being to optimize the efficiency of information transmission in space.

What we are currently experiencing is a "revolution of computation itself." It is no longer about how information is transmitted, but how information is processed and reorganized. This level of underlying reconstruction has not occurred in the industry since the birth of semiconductor integrated circuits in the 1960s and 70s.

This distinction has substantial commercial implications: it means that the technological foundation relied upon by application-layer entrepreneurs is still in a dynamic evolution of high-frequency displacement.

Reviewing the development trajectory of the past few years, the industry has experienced three key turning points:

1. The ChatGPT moment at the end of 2022 established the technical feasibility and expansion potential of unsupervised pre-training;

2. The release of OpenAI o1 at the end of 2024 validated the path of logical reasoning, marking that large models initially possessed the ability to transition from intuitive fast thinking (System 1) to logical slow thinking (System 2);

3. The launch of Claude Code and Opus 4.5 confirmed the usability of "Long-horizon Agents" in complex task environments.

These three stages seem to be on the same technological continuum, but by the time the third stage is completed, the industry has actually crossed a critical threshold ------ Artificial General Intelligence (AGI) has evolved from a technical hypothesis to an industrial reality.

Earlier this year, Sequoia predicted in a column that "2026 will be the year of AGI." At that time, this assertion was still controversial within the industry; however, as we progress to the present, it has gradually solidified into a common consensus among the tech and investment communities.

2. From "Carriage" to "Automobile": The Reality and Side Effects of Commercial Landing

If we borrow the classic question from Sequoia founder Don Valentine ------ "So what? What substantial value does this create?" ------ the current technological evolution actually means that transportation tools have crossed from the "improved carriage" to the "automobile" stage.

In the past few years, many products labeled as AI essentially still belong to the gradual improvement of traditional software, merely offering better digital tools; whereas now, technology is fundamentally changing the processes and forms of productivity delivery.

In practical business terms, this migration is manifesting in multiple scenarios:

• Software forms are evolving from "toolboxes" to "collaborators": Products represented by Instinct or Muse have transformed from mere task processing tools to end-to-end business collaborators delivering results. Their development logic is similar to that of the video conferencing industry: before Zoom crossed the usability and trust threshold, video conferencing technology existed for many years but was not seriously accepted by mainstream business; once the technology's stability surpassed a specific threshold, online meetings quickly established themselves as the default workflow. Current AI is in a similar trust transition interval.

• Universalization of advanced professional knowledge: In the medical field, primary health institutions and practitioners in remote areas can access industry-level diagnostic knowledge through mobile terminals; in education, innovative teaching systems represented by Alpha School are demonstrating possible paths for AI-native educational models to reshape personalized teaching.

• Infrastructure spillover to the real industry: The wave of building ultra-large-scale data centers is creating sustained industrial-grade demand in the fields of physical engineering, electromechanical equipment, and energy infrastructure.

At the same time, the rapid technological leap is also triggering a series of chain reactions at the industrial and social levels:

• Capital expenditure pressure on cloud infrastructure vendors: Ultra-large-scale cloud service providers (Hyperscalers) are continuously increasing their investment scale in computing infrastructure, and relying solely on free cash flow is no longer sufficient to fully cover Capex; the industry is beginning to commonly finance through bond issuance to sustain the computing power competition.

• Emergence of new cybersecurity threats: The enhancement of foundational model capabilities objectively lowers the development threshold for advanced automated attack tools, and the balance of offense and defense in digital infrastructure is under continuous impact.

• Value reshaping of knowledge-based occupations: Pure text and basic information processing white-collar jobs are being significantly affected. Workers who can actively utilize models to amplify productivity gain a significant advantage in production efficiency, while labor groups that fail to establish collaborative barriers face ongoing marginal substitution pressure.

• Friction in social distribution and governance: The issue of computing power consumption is gradually evolving into a geopolitical and regional governance topic; the differences in capital returns and labor returns brought about by AI have also exacerbated existing wealth distribution inequalities to some extent.

3. Shift in R&D Focus of Frontier Laboratories: Advanced Exploration and Engineering Reality

Delving into the R&D front lines of leading frontier laboratories reveals that researchers' judgments on technological evolution paths are often highly focused and straightforward.

Currently, core laboratories have internally shifted their R&D focus from the engineering realization of AGI to the fields of Artificial Super Intelligence (ASI) and Recursive Self-Improvement (RSI).

Among them, RSI refers to the model's ability to autonomously participate in or even lead the next generation of model code writing, testing, and architecture optimization in a self-enhancing cycle. The emergence of this technological feedback loop has prompted the research and development teams to cautiously assess the "technological exponential takeoff" and the potential risk of alignment failure (p(doom)). This forms the core technical background for the recent public calls by leading figures such as Dario Amodei, co-founder of Anthropic, for the entire industry to calibrate the pace of cutting-edge research and development.

At the specific engineering and commercialization level, frontline laboratories are exhibiting the following definitive trends:

1. Breaking the boundaries of pure natural language: Model training and inference scenarios are accelerating their migration to rigorous deduction fields such as complex scientific exploration and verification of cutting-edge mathematical conjectures, with the underlying computational architecture facing a new round of efficiency optimization demands;

2. Long-term mismatch in supply and demand for computing power: The shortage of computing power remains a core engineering bottleneck that runs through top laboratories and application-layer enterprises, and the trend of self-developed ASICs and other specialized silicon chips across the industry is still strengthening;

3. Intensifying competition for inference APIs: The commercial competition for the consumption of tokens by foundational models is extremely fierce, with API pricing continuously declining. To effectively secure usage from Fortune 500 companies and other large enterprises, leading laboratories are increasingly building their own enterprise-level service teams to promote scenario penetration through high-intensity customized consulting;

4. Normalization of external mergers and acquisitions: As the competitive landscape across fields stabilizes, leading laboratories are gradually establishing a normalized mechanism for talent, team, and asset acquisitions, with the commercial landscape exhibiting characteristics of expansion typical of large enterprises.

IV. The Technology Diffusion Gap and the Commercial Explosive Power of the Application Layer

There exists a significant lag between the explosion of foundational model capabilities and the actual absorption levels in the industry, which is referred to as the "technology diffusion gap." For application-layer startups, this gap constitutes the primary commercialization opportunity.

In this cycle, the most commercially certain entry points are concentrated in the high-value knowledge work sector: software engineering, cybersecurity, medical professional services, finance and investment banking, auditing and compliance, etc. Each sub-sector with high professional cognitive barriers has the potential to nurture new generation platform companies with valuations in the tens of billions of dollars.

Referring to the position of core record systems established by ServiceNow, Workday, and Salesforce during the cloud computing cycle, a new generation of record systems centered around AI as the interaction and logic hub is currently forming in various professional vertical scenarios.

In the exploration of underlying applications, some vertical architectures have demonstrated efficiency ratios far exceeding those of general large models:

• Candidate molecule design in the biopharmaceutical field: Models optimized for specific scientific calculations can now achieve extremely short cycle generation of candidate drug molecules, significantly compressing the time required for molecular design in traditional drug development;

• Explosive single-point architectures: The industry's demand for homogeneous general bases is limited, but vertical models capable of reconstructing single-point tasks with new architectures are growing rapidly. A recent typical case, Jev, is essentially a high-throughput "decision classifier." In actual production, a large number of complex LLM invocation requirements essentially belong to discrete selection classification tasks. This product has reduced the single inference cost of such tasks by approximately 100 times through extreme computational trimming, achieving an extraordinary leap in annual recurring revenue (ARR) in a very short time.

This commercial explosive power indicates that in sub-sectors with high gross margins and significant cognitive barriers, AI-enabled software products are exhibiting abnormal growth slopes.

Companies that can continuously establish barriers at the application layer generally exhibit the following organizational and business characteristics:

First, autonomous control of the intelligent base.

Companies are beginning to proactively migrate their core workloads from closed-source large model APIs to deep post-training based on open-source bases. The core driving force behind this is cost and efficiency: general foundational models primarily serve limited equilibrium points on the general Pareto frontier; to achieve extreme cost-effectiveness in specific commercial tasks, companies holding open-source models that have been specifically engineered are the optimal solution.

Second, organizational structure shifts towards "long-board breakthroughs."

As the underlying technology stack and model capabilities undergo iterations every few months, application-layer companies that survive must maintain extremely short product self-reconstruction cycles. These organizations are gradually abandoning traditional collaborative models constrained by team shortcomings, instead relying on a small number of top core personnel to undertake technical challenges and drive overall business forward at a high speed.

Third, corporate governance evolves into a distributed collaborative network.

Traditional hierarchical command and control systems are giving way to decentralized architectures centered around AI intelligences as information hubs, relying on high-density automated processes and autonomous collaboration to maintain operational efficiency with a very small personnel scale.

V. The Differentiation of Capital Pricing Systems and Leverage Effects

In the capital market, over the past year, the asset pricing in the AI field has shown significant polarization and high structuring characteristics.

To balance risks and returns in an overheated asset environment, some early-stage projects have begun to adopt a "two-step" financing mechanism: structurally separating the strategic co-construction responsibilities of industrial capital from the pure financial capital that provides subsequent expansion ammunition.

In several typical transaction samples that Sequoia has deeply participated in over the past year, the following trends have emerged:

• In the early stages, institutions acting as business co-builders completed early leading investments at relatively reasonable valuation benchmarks (e.g., in the post-investment range of $110 million);

• After a very short business validation period, external pure financial capital quickly followed up, rapidly pushing the target's valuation to several times or even higher levels in subsequent rounds (some sample averages reached over $3 billion).

This rapid switching of high valuation multiples objectively reflects the extreme demand of primary market funds for targets with genuine technological and business barriers, but it also concentrates the high liquidity premium accumulated in the current market.

Facing the future industrial evolution path, no institution can hold a completely certain roadmap. However, the fundamental laws that run through the technology investment cycle still hold:

When the marginal cost of a key productive factor shows an order of magnitude decrease, while the comprehensive processing efficiency continues to improve exponentially, the structural reshaping at the industrial level will become an irreversible objective reality. The technological reconstruction of the global business system is currently only in the unfolding phase of the first half.

This content is provided for general informational purposes only and doesn't constitute financial, investment, legal, or tax advice. Any events, rewards, online promotions, or related information mentioned herein should not be considered a recommendation, solicitation, or invitation to purchase, sell, trade, or otherwise deal in any crypto assets. Crypto assets are highly volatile and may result in loss. The availability of WEEX services, products, and related events may vary by region. You are responsible for ensuring that your participation is in accordance with applicable local laws and regulations.