创投库Database 创投评Insight 创投会Network 创投营Programs 项目评估Get evaluated 会员Pricing加入 SVTRJoin SVTR

加入 SVTRJoin SVTR

Insight

All Thesis Profile Signal Radar Weekly Recap Dialogue

Database
Insight
Network
Programs

For Founders
Get evaluated
Check rating
Cases

For Investors
Selected Deals
Enterprise
MCP

Pricing
Join / Sign in

About
Methodology
Privacy
Terms

© 2026 SVTR · Silicon Valley Technology Review · SVTR AI LLC

Get evaluated PricingJoin SVTR 中文

85 original AI-venture articles from SVTR

Two Markets, Two Prices

Two Markets, Two Prices

This week, the AI venture capital market showed a notable divergence: fundraising continued to price growth curves, while mergers and acquisitions began to price assets more seriously. Cognition's valuation rose from $26 billion to $48 billion within four months; on the other hand, according to Bloomberg, Anthropic abandoned talks for a $6 billion acquisition of Decart after due diligence.

The AI Booked the Table. Paying for It Is the Hard Part.

The AI Booked the Table. Paying for It Is the Hard Part.

At the SVTR Online Venture Session on September 11, 2026, Bravo founder Max So demoed an AI that phones a restaurant and books a table. The hard part is the next step: how dinner gets paid for. Per the SVTR AI Venture Database, roughly 30 AI companies treat dining as their main setting; more than twenty sell to the restaurant, and only two work from the diner's wallet. The biggest fundraiser among them, inKind, raised roughly $1.2 billion across three rounds in 2026, almost entirely debt and mezzanine capital, to advance money to restaurants. Our read: what an AI needs to close a transaction in dining is a merchant network that accepts the order and can settle, built one restaurant at a time, with the model behind it.

← Prev2 / 4Next →

What a Fund Returned Last Time No Longer Decides What It Can Raise This Time

What a Fund Returned Last Time No Longer Decides What It Can Raise This Time

One fund has paid its backers $3.60 for every dollar invested, a top 1% record globally, yet raised only half its previous size. In the same month a two-year-old firm that has returned nothing raised a larger fund than its first. SVTR read: how much a fund can raise now depends on whether its stated direction is on its backers’ shopping list for the year, with past returns slipping into second place. a16z already had two funds that could back hardware and still opened a separate $1.1B Machine Age Fund, because hardware pays back far slower than software and always loses the internal argument when the two share one fund.

AI4S Valuations Are Set Before the Science Is Proven

AI4S Valuations Are Set Before the Science Is Proven

SVTR's database tracks 175 AI4S companies that raised 118 confirmed rounds totaling roughly $7.50 billion over the past 12 months, with 13 companies taking 73% of it. Stage explains nothing: Periodic Labs raised a $300M seed in its founding year, while Aureka reached Series B for $100M. Age explains nothing either: the within-cohort spread runs to 27x against a sub-3x gap between age groups. Checking all 13 nine-figure companies against primary sources, what sets the tier is not pipeline versus tools but whether anyone else can build the thing in the near term: Chai Discovery licenses its models to Pfizer and Lilly and raised $400M, while Adaptyv Bio, which runs protein validation for Chai, raised $40M. Capital is not waiting for validation. It priced capability scarcity in the very first round.

The Three Biggest Checks This Week Came From Nowhere Near Venture Capital

The Three Biggest Checks This Week Came From Nowhere Near Venture Capital

This week, SVTR's AI venture activity index reached 154.2, up 9.4 points from the previous period. The AI venture database recorded 83 funding rounds, with approximately $7 billion disclosed, of which $4.5 billion was raised by two infrastructure companies, Nscale and Fluidstack. Nscale's $3 billion came in the form of bank loans: two investment-grade project term loans, led by J.P. Morgan and Goldman Sachs. During the same week, NVIDIA acquired Hugging Face for $12.93 billion, nearly twice the total funding of the entire week. On the Chinese side, Enflame Technology's listing on the Sci-Tech Innovation Board, with its largest shareholder Tencent contributing 83.79% of its revenue from last year. The three largest funding sources came from the credit market, industry giants, and the public market, while the largest venture capital round during the same period was only $300 million.

Fifteen Silicon Valley Companies Set Up Shop Next to BOE

Fifteen Silicon Valley Companies Set Up Shop Next to BOE

Over a thousand applications, three rounds of review, fifteen companies, most headquartered in Silicon Valley. By sector label the list does not cohere, but more than half of these companies ship something physical. China's role, in the eyes of this cohort, is shifting from a large market to a production line that can actually start running.

AI compute is no longer financed as a company

AI compute is no longer financed as a company

SVTR's AI venture database recorded 83 AI funding rounds this week with $7.01B disclosed, and two compute companies took $4.5B of it, or 64%. Nscale's roughly $3.05B sits in two single-campus project SPVs, led separately by J.P. Morgan and Goldman Sachs, both rated investment grade. Fluidstack owns not a single chip, yet Anthropic's $50B multi-year capacity agreement carried it to a $1.5B round led by Jane Street at an $18B valuation, up from $7.5B in July. In the same week Figure committed to $3.5B of compute and Nscale took equity in Figure in return. SVTR's read: in the compute layer the financed object is shifting from equity to contract, and the contract itself is now bought with equity. Venture capital has been pushed to both ends. In China not one round this week carried debt structure, blocked earlier in the chain by disclosure that stops at ranges.

AfterQuery: Beaten by Hallucination, They Started Selling the Cure

Spencer Mateega and Carlos Georgescu went to high school together and put a Y Combinator application together in 48 hours in early 2025, with no product and only a destination: San Francisco. By Mateega's own account on X, the three of them walked away from a combined $850,000 in stable jobs. Eighteen months later AfterQuery was reported at a $3.2 billion valuation and is said to be profitable, having taken $30.5 million from investors in total. SVTR's call: what these two are betting on is not a data business but the persistence of the model's defect. Beaten by hallucination while building an application, they turned around and made that defect their business, with 30 employees serving close to 100,000 doctors, lawyers and engineers. The risk is not the competition but the customers: the frontier labs they serve work day and night to stop needing them.

AI Money Splits in Two: Half Goes to Build Machines, Half to Acquire Teams

AI Money Splits in Two: Half Goes to Build Machines, Half to Acquire Teams

a16z closed a new $1.1 billion fund this week, focused on chips, memory, networking, power, and machines that house AI. In the same week, the largest private round went to Xpeng Robotics ($900 million, post-money valuation of $6.3 billion). Of the top 10 fundraises, five were for building machines or managing power, totaling $1.57 billion, which accounted for half of the disclosed amount for the week. The only company to complete an IPO was quantum processor company Pasqal, which rose 95% on its first day. On the software side, seven of 10 acquisitions did not disclose prices, with targets mostly being small teams that had been in operation for one or two years. SVTR's view: the primary market is creating a separate pricing channel for physical assets.

AI's Lead-Investor Seat Now Belongs to Customers, Rivals, and Suppliers

AI's Lead-Investor Seat Now Belongs to Customers, Rivals, and Suppliers

Of the 71 AI funding rounds SVTR's database tracked this week, industrial players took the lead or sole-investor seat in at least four: the three major labels and EA invested in Stability AI, CATL made a sole strategic investment in Yigen, Momenta led GenRobot, and Builders FirstSource led Digs with a five-year commercial agreement attached. SVTR's judgment: in sectors dependent on copyright, orders, capacity, and data, funding rounds are becoming vehicles for factor transactions, and pricing power is shifting from financial VCs to industry.

When a Partner Votes a 1, Sequoia Still Presses Green

Dialogue2026-08

When a Partner Votes a 1, Sequoia Still Presses Green

Sequoia partner Julien Bek describes a rule outsiders rarely see: even when a partner votes a deal a 1, the sponsor can still press green, which is how SpaceX got in, and the price is that a bad call is underwritten with your own seat. SVTR checks that account against 214 AI-related financings involving Sequoia in its venture database, and finds that in this cycle the scarce capability is no longer finding the company but being willing to back it alone.

OpenRouter: Owning No GPUs, It Took AI Inference's Biggest Exit First

Stripe is acquiring OpenRouter for a reported ~$7.5B. The company trains no models and owns no compute; it raised just $153M over three years, yet cashed out before inference clouds that raised over ten times as much. SVTR breaks down its aggregation playbook, the 5.5% take-rate model, the acquisition signals hiding in its Series B cap table, and why China never grew an equivalent.

AI Companies Are Being Priced in Parts: Brains Sold Outright, Capacity Financed on Credit

AI Companies Are Being Priced in Parts: Brains Sold Outright, Capacity Financed on Credit

SVTR AI Venture Database recorded 61 AI capital events worth $3.67 billion in disclosed value this week, with 62% going to companies that must build production lines, plants or data centers. Castelion's $1 billion Series C combined $800 million of equity with a $250 million revolving credit facility, co-led by J.P. MorganChase's Strategic Investment Group, Carlyle and a16z. The same week, Groq, whose inference technology Nvidia licensed for nearly $20 billion while hiring roughly 90% of its chip team, reopened a Series A at half its former valuation to fund an inference cloud. SVTR's judgment: the AI private market is splitting companies into intellectual and productive assets and pricing each separately.

The Middle Layer of Video AI Is Being Eaten by Its Own Customers

The Middle Layer of Video AI Is Being Eaten by Its Own Customers

On August 17, the Motion Picture Association signed its first AI copyright protection framework with ByteDance, focusing on what the model cannot generate. SVTR's AI venture database shows that in the past 12 months, 21 confirmed rounds of funding totaled approximately $1.96 billion, with nearly 70% flowing to those building and hosting models. The only company in the application layer to secure significant funding, Synthesia, sells enterprise training videos with constraints. The middle layer - 'tools that use models to create content for others' - is being replaced by its own customers building in-house solutions; in China, 76% of the money only recognizes one position in the model layer.

Bypassing Entities, Bypassing Applications, Bypassing Growth: This Week's AI Capital Only Buys "Rentals"

Bypassing Entities, Bypassing Applications, Bypassing Growth: This Week's AI Capital Only Buys "Rentals"

This week, AI capital invested $6.4 billion in 35 deals, all of which were new investments, without any IPOs, acquisitions, or exits. The funds bypassed the core AI entities and applications, instead investing in upstream sensor and data companies, while US funds bypassed applications to invest in model infrastructure, energy storage, and connectivity, and Thrive Holdings acquired over 70 cash-generating traditional companies with AI. This week, capital is buying the same thing: something that no one can avoid, "rentals".

How Do You Quickly Judge an Early-Stage AI Project? Five Signals, One Coordinate System

How Do You Quickly Judge an Early-Stage AI Project? Five Signals, One Coordinate System

What an early-stage AI project needs most is not a standalone score; it is a position. SVTR reads five signal dimensions (team, market, product, moat, traction) and places each project into a coordinate system of same-sector, same-stage peers to find the gap that most needs closing. From five-dimension ratings across nearly 6,000 AI companies: the strongest dimension is team, and what is most commonly missing is real evidence of moat and traction. A deck, a text intro, or a website all work; evaluation is not the end, it is the entrance to the SVTR venture network.

The Money in Embodied Intelligence Is Leaving the Robot Body

The Money in Embodied Intelligence Is Leaving the Robot Body

The largest funding round in SVTR's AI venture database this week went to Yimu Technology, a tactile vision sensing company that does not build robots (valued over 10 billion yuan). SVTR's judgment: capital in embodied intelligence is shifting from robot bodies to upstream areas such as sensing, force control, motion control, and real-world data. The winner of this round of embodied intelligence hype may not build any complete machines. Signals from venture platforms confirm this: start-up supply is still concentrated on robot bodies, but money has already moved ahead, bypassing them.

Ten Teams Pitched, Not One Built a Humanoid: Field Notes from an AI Robotics Session

Ten Teams Pitched, Not One Built a Humanoid: Field Notes from an AI Robotics Session

Ten teams pitched at SVTR's AI + Robotics session and not one was building a humanoid. The founders with real traction have shifted value away from the humanoid body toward actuators, vertical use cases, and companion hardware, matching the investor panel's read that vertical robots already make money while humanoids still need years.

Compressing Time, Buying Physical Assets, Heading to Inspection

Compressing Time, Buying Physical Assets, Heading to Inspection

This week saw 31 confirmed funding rounds and $6.4 billion in disclosed amounts, with zero exits or returns. Approximately 70% flowed into physical assets such as factories, nuclear reactors, home batteries, and optical modules. A four-month-old company, Poshell Robotics, raised an亿美元-level round. Yujue Technology completed its科创板 pricing with a listed market value of 61 billion yuan, and will start its subscription next week. AI capital is compressing time, buying physical assets, and heading to inspection.

The Angel Round Is Just a Label Now: AI Capital Prices Day One Off the Endgame

The Angel Round Is Just a Label Now: AI Capital Prices Day One Off the Endgame

PokeBot, four months old, raised nearly $100M; Volta, seven months old, raised $300M plus a $10B compute contract. SVTR argues the milestone-by-milestone round system is breaking down in AI: with IPO pricing (Unitree at a RMB 61B offering valuation) and mega-buyer contracts serving as endgame anchors, capital now pays for seats and capacity on day one, and 56% of this week's disclosed dollars went to companies under three years old.

40 Years, 8 Companies: Decoding Elon Musk's Work Principles

40 Years, 8 Companies: Decoding Elon Musk's Work Principles

In the public eye, Elon Musk is a social media provocateur, a space enthusiast, and an extreme manager. But to those who have truly studied him, he is a founder who has continuously iterated the same startup algorithm across 40 years and 8 companies without interruption. In this episode of Founders, host David Senra distilled Walter Isaacson's biography of Musk through 60 hours of deep reading and 40 pages of notes, extracting the consistent work principles that have guided Musk across Zip2, PayPal, SpaceX, Tesla, Neuralink, Boring Company, X (formerly Twitter), and xAI. SVTR views this as the most hardcore foundational methodology for building future companies.

Where Does Creativity Come From in the AI Era? Asimov's Revealed Formula for Creativity

Where Does Creativity Come From in the AI Era? Asimov's Revealed Formula for Creativity

In an era where generative AI and large model tools are rapidly approaching 'zero-barrier' access, the technical barriers to creation are collapsing quickly. Writing is no longer scarce, code is within reach, and products are becoming increasingly similar. This trend is already reflected in multiple cases in SVTR's AI venture database: founders are shifting from 'engineering thinking' to 'creative operations', because in a world where tools are highly homogenized, the real moat is not technology, but taste and distinctiveness.

Leverage on Computing Power, Lock Agents, and Deleveraging to Repay Debts

Leverage on Computing Power, Lock Agents, and Deleveraging to Repay Debts

The primary market is still leveraging AI: NVIDIA made three moves in a week, and 11 out of 15 AI security deals were about locking agents. The secondary market, however, is deleveraging and repaying debts, with situational awareness funds liquidated and MiniMax's listing halved. The divergence between primary and secondary markets has made top-tier assets the last stabilizing force.

AI Security Switches Protagonists: From Guarding People to Governing Agents

AI Security Switches Protagonists: From Guarding People to Governing Agents

This week the SVTR AI Venture Database recorded 106 global AI capital events, and security & compliance led all sectors with 15 rounds, 11 of them in AI-agent and non-human-identity security (up from about 2 per week before): 9 in the US, 5 in Israel, and zero in China. Cyera acquired Oasis for $1B and Okta acquired Permiso for nearly $200M in the same week. SVTR view: the protagonist of AI security is switching from guarding people to governing agents, and the smallest unit being protected is shifting from a person account to a machine identity; machine identities already outnumber human ones 82 to 1, this lane holds almost only in the US and Israel, and incumbents have begun absorbing the startups.