YC Investment Logic Major Shift: From Software to Hardware, From Models to the Physical World

CN
PANews
Follow
1 hour ago
179506652761543.jpg

Written by: Little Cookie

In the summer of 2026, YC released a list of 15 "Request for Startups," covering lunar manufacturing, reasoning chips, anti-drone defenses, almost laying out the entire spectrum of hard technology.

Three months later, the Fall 2026 batch's list shrank to 13 items.

Three months after that, the Winter 2027 batch had 11 items.

Each round of reductions refined the focus.

From this year's four quarterly RFS lists (YC has changed to four batches per year), a clear investment thesis evolution path can be read: AI's battlefield is rapidly shifting from cloud models to the physical world’s infrastructure and bottlenecks.

Below is a complete breakdown of the most recent lists for Fall 2026 (13 items) and Winter 2027 (11 items).

A Letter from a Secretary of the Army

One of the most eye-catching items in the Fall 2026 list has nothing to do with AI technology: a handwritten submission from the current U.S. Secretary of the Army Daniel P. Driscoll.

This is the first time in YC's history that a serving government official has directly signed a RFS.

The title is "The Future of American Defense," where someone from the Pentagon addresses the startup accelerator's admission brochure: we need you.

This indicates that the U.S. military is bypassing the traditional defense contractor system (the comfort zone of companies like Lockheed Martin) and is directly reaching out to garages in Silicon Valley.

YC's role has therefore subtly changed: it is becoming the interface connecting entrepreneurs with national will.

Looking again at the complete list from Fall 2026, Driscoll's presence is far from an isolated case.

Maritime computing, AI-native compliance infrastructure, physical world operating systems... military and national security undercurrents permeate several directions.

Fall 2026: Three-Tier Structure of 13 Directions

When laid out, the 13 items from the Fall 2026 list reveal a clear three-tier structure.

Physical Layer: AI Needs Hardware, Hardware Needs New Ground

Francois Chaubard's proposal "Compute at Sea" is the most daring imagination at this layer.

Building data centers on land is becoming increasingly difficult: power approvals, environmental compliance, land costs, each of which is slowing down the expansion of AI's computing power.

Moving modular computing to the sea, bypassing terrestrial bottlenecks, YC alumni company Atomarine is already constructing nuclear-powered computing barges.

Charlie Warren's "New Operating Systems for the Physical World" targets another pain point: AI can write code and chat, but it still cannot directly manipulate physical devices.

Robots in factories, irrigation systems on farms, traffic lights in cities, all require a set of AI-native operating systems for coordination.

"Data for the Real World" points to an even more fundamental issue: the internet data used to train AI is about to run out, the next wave of data must be collected from the physical world.

Data generated by sensors, cameras, and industrial equipment is the fuel for AI's next phase.

Trust Layer: The Stronger AI, the More Important "Who You Are" Is

"Proving You’re Human" addresses an accelerating issue: as AI-generated content and behaviors become increasingly indistinguishable from real, identity verification will become an essential infrastructure of the entire internet.

"AI-Native Compliance Infrastructure" targets the enterprise end.

AI is starting to make decisions; who will audit it? Who is responsible when mistakes occur? Traditional compliance frameworks are not designed for the AI era at all.

The logic of this layer can be summarized in one sentence: the greater the capability, the more urgent the demand for trust infrastructure.

Application Layer: From Tools to Services

Andrew Miklas's "The Primer" pays tribute to the AI tutor from Neal Stephenson's sci-fi novel "Diamond Age," aiming to provide each child with a personalized AI mentor.

Raphael Schaad's "AI-Powered Consumer Products for 1 Billion People" targets consumer products for a global user base of one billion.

Max Kolysh's "AI for the Aging Population" touches on a market that is rarely addressed seriously in the tech circle: global aging.

Japan, South Korea, and China's demographic structures are rapidly collapsing; AI caregiving, AI companionship, AI health management represent a highly certain huge market.

Pete Koomen's "A Cloud for Small Software" and Aaron Epstein's "Multiplayer AI" return to the developer tools level.

The former aims to provide dedicated cloud infrastructure for small applications in the AI era, while the latter focuses on the AI experience in multiplayer collaborative scenarios.

Harsha Gaddipati's "Self-Maintaining APIs" suggests enabling APIs to monitor and repair themselves using AI, thereby freeing up developer capacity.

Now Is the Best Time to Build in Crypto

This statement comes from YC visiting partner Nemil Dalal.

He wrote a dedicated direction for Crypto in the Fall 2026 RFS titled "The Best Time to Build in Crypto."

Dalal's confidence primarily comes from stablecoins.

The global settlement volume of stablecoins is catching up to Visa, and the advancement of the GENIUS Act and the CLARITY Act means that the U.S. regulatory framework is shifting from suppression to integration.

Stablecoins have found their PMF and are becoming part of the global financial infrastructure.

He specifically mentioned a concept: Agentic Commerce.

When AI Agents start replacing humans in purchasing and transactions, they won’t apply for bank accounts and credit cards.

On-chain payments are the most natural settlement method for AI Agents.

This cross-narrative (AI × Crypto) is becoming the common frontier for both industries.

The wave of institutionalization is also accelerating.

ETFs, custodians, compliance trading platforms... traditional financial institutions' interest in Crypto assets has already transitioned from exploration to strategic allocation.

What's more interesting is YC's own actions.

In February 2026, YC partnered with Coinbase to start using USDC stablecoins to fund all YC startups, supporting Ethereum, Base, and Solana, with transfer costs under $0.01 and settlement times counted in seconds.

Note: This option is available to all YC startups, not limited to the Crypto direction.

On October 28, Dalal further released a more detailed "Crypto Request for Startups," breaking down specific tracks into stablecoin infrastructure, on-chain trading tools, DeFi institutional products, and other subdivisions.

Winter 2027: The Bottleneck Is Not in the Model but in Transformers and Rare Earth Magnets

If Fall 2026 was still outlining the blueprint for AI's advance into the physical world, Winter 2027 directly points to the most painful spot: the bottleneck.

Six partners, 11 directions, almost every one answers the same question: the model is strong enough, what is preventing it from being implemented?

The answer is surprisingly "low-tech."

Jared Friedman’s "Data Center Supply Chain" presents a staggering statistic: the delivery time for large power transformers is 3 years.

No matter how many GPU chips there are, without power access it’s just scrapped iron.

The entire expansion speed of the AI industry is being choked by a category of power equipment from the last century.

Diana Hu's "Robots Made in America" highlights another bottleneck: in 2025, the U.S. deployed only 38,400 robots, while China holds 94% of the world’s sintered rare earth magnet production capacity.

Rare earth magnets are core components of robot motors; with the supply chain in others' hands, the U.S. robotics industry is built on sand.

Transformers, rare earth magnets, sound like topics from the last century, yet they are the directions YC wishes to invest in most in 2027.

Brad Flora proposed two directions, creating an interesting contrast.

"The $15-an-Hour Expert" depicts a scenario of class transition driven by AI: when AI enables ordinary workers earning $15 an hour to perform tasks that previously required professional certification, the entire vocational education and labor market will be rewritten.

"Wall Street for the Physical Buildout" points out that heavy asset projects like robotics factories, GPU clusters, and data centers require a new set of financial products for financing and pricing.

One is liberating labor, the other is liberating capital.

Diana Hu's other direction, "Vertical Foundation Models," steps out of the arms race of general large models, looking at small dedicated models in vertical fields like manufacturing, energy, and biology.

Jared Friedman’s "LLMs That Write Well" points to a counterintuitive phenomenon: models that are becoming stronger in coding and mathematics are actually regressing in writing ability.

This issue has a tangible impact on the content industry and marketing.

Gustaf Alströmer wrote two directions aimed at government.

"Plaid for Government Benefits" creates a unified API for government welfare systems, while "AI for City Hall" aims to use AI to reshape local government administrative efficiency.

The directions are not sexy, but the market is enormous: U.S. government IT spending exceeds $100 billion annually, with the vast majority being consumed by legacy systems from the last century.

Jon Xu's "Own Your Intelligence" raises a strategic issue: almost all companies using AI are "renting intelligence," outsourcing core capabilities to a few model providers.

With open-source model performance catching up, enterprises should consider "owning" their AI capabilities.

Ankit Gupta's "Startups in a World of Abundant Space Access" looks at the cost revolution in space brought about by SpaceX's Starship.

With launch costs reduced by 100 times, things that only NASA could do in the past can now be done by startups.

"AI for the Air-Gapped Economy" focuses on environments that cannot connect to the internet, such as military bases, nuclear facilities, and hospital operating rooms, where offline reasoning and localized deployment is almost a virgin domain.

Two Lists, One Main Line

Looking at the Fall 2026 and Winter 2027 lists together reveals an evolutionary trajectory of an investment thesis.

YC’s focus is undergoing a fundamental gravitational shift: from software to hardware, from models to infrastructure, from the digital world to the physical world.

The maritime computing, defense technology, and physical world operating systems of Fall 2026 have already sent signals; the transformer supply chain, rare earth magnets, and domestic robot manufacturing of Winter 2027 have turned those signals into commands.

These directions share a common feature: each one involves supply chain security and geopolitical maneuvering; YC's investment thesis has become deeply intertwined with U.S. industrial policy.

For the Crypto industry, the cross-signals are also worth paying attention to.

The exclusive Crypto RFS from Fall 2026 and YC's USDC funding plan have already signaled their stance.

Winter 2027, while not directly naming Crypto, has at least three directions that naturally point to on-chain scenarios:

"Own Your Intelligence," which calls for decentralized AI capabilities, has already been addressed by DePIN and decentralized reasoning networks.

The new financing tools required by "Wall Street for the Physical Buildout" may be most efficiently provided by on-chain RWA.

"Proving You’re Human" (Fall 2026) combined with offline AI and Air-Gapped Economy (Winter 2027) forms a complete demand map for decentralized identity and privacy computing.

YC is doing subtraction, and Crypto entrepreneurs should do addition at these intersection points.

免责声明:本文章仅代表作者个人观点,不代表本平台的立场和观点。本文章仅供信息分享,不构成对任何人的任何投资建议。用户与作者之间的任何争议,与本平台无关。如网页中刊载的文章或图片涉及侵权,请提供相关的权利证明和身份证明发送邮件到[email protected],本平台相关工作人员将会进行核查。

Share To
APP

X

Telegram

Facebook

Reddit

CopyLink