Hyperscalers double-dip the chip | PitchBook News Summary

1. Key Themes

Theme: Hyperscalers are turning proprietary silicon into a profit center and a strategic lever against Nvidia

Compute supply, not AI demand, is the binding constraint on hyperscaler growth

As usage shifts toward autonomous agents, scaling depends on solving the chip bottleneck rather than on end-user appetite. Controlling silicon means controlling growth, cost structure and timelines.

"As AI usage shifts from chatbots to autonomous agents, the demand for compute is rising rapidly. For hyperscalers and neoclouds, their ability to scale depends less on appetite for AI and more on how fast they can resolve their own supply bottleneck."
"The more successful firms are at producing in-house silicon, the more control they will have over their own growth, cost structure and build-out timelines."

Custom chips have moved from cost-saving tool to a standalone product line

The evidence is revenue run rates, committed backlog and external sales.

"Proprietary silicon is already becoming a meaningful business, shifting from a cost-saving tool to a proprietary product that puts hyperscalers in more direct competition with Nvidia."
"Amazon's chips business has an annual revenue run rate of more than $25 billion and over $225 billion in revenue commitments for its Trainium chips, including from Anthropic and OpenAI. It also signaled the possibility of selling chips to third parties."
"Google recently began recognizing revenue from TPU system sales, signaling success in the business."
"Over 60% of Alibaba's T-Head chips serve external customers, which the company says helps it avoid the high price premiums of commercial chips."

Scarcity is a source of leverage for chip builders

"The shortage is unlikely to ease soon. For the companies building their own chips, that scarcity is becoming a source of leverage."

Theme: Venture capital is concentrating into fewer, larger rounds anchored on frontier AI models

Frontier-model challengers can command mega-rounds

"More AI model challengers are throwing their hats into the ring. Jev, the model developed by TypeSafe AI, paved the way for the startup to secure a new $870 million round at a $7.5 billion valuation. The round underscores two trends: the prioritization of frontier models and the consolidation of VC funding into larger rounds."
"TypeSafe AI, the developer of Jev, an AI model for developers, raised an $870 million Series A led by Andreessen Horowitz at a $7.5 billion valuation."
Other large AI/compute raises this cycle: "Chinese-founded AI lab Manus... raised a $500 million round led by Boyu Capital and IDG Capital" and "Nvidia-backed data center operator Firmus is considering raising a new round of up to $3 billion after shelving its IPO."

Capital for AI is being raised at sovereign scale

"SoftBank is in talks to raise a new $100 billion fund from investors in the Gulf for AI investments, the Financial Times reported."

Theme: LPs are shifting scrutiny from deal sourcing to portfolio health, valuation and liquidity

Private credit default measurement is the new LP focus

"Every LP used to want to know how direct lenders sourced and originated their deals. Today, LPs are more focused on the health of portfolios."
"Above all, LPs want to know how managers measure defaults, a challenging exercise given that direct-lending stress is generally resolved behind closed doors, through term amendments or by switching from cash interest to payment-in-kind, said Kerry Dolan, managing partner of Brinley Partners."

Regulators are tightening valuation scrutiny

"The regulator has moved from using more holistic models to an assumption-based model when determining whether assets are fairly valued, said Matt Giordano, the leader of KPMG's asset management audit practice."
"Now it's, what is your comp set and how did it change? Why did you not consider this company on the comp set?"

Longer hold periods are breaking LP pacing models

"Late last year, NEPC adjusted its pacing model to reflect longer PE and VC hold periods."
"Without these adjustments, actual fund lives and net asset values extend beyond the model's assumptions and reduce LPs' capacity to commit to future funds, said Sarah Samuels, the investment consultant's chief investment officer."

2. Contrarian Perspectives

The binding constraint on AI is supply, not demand, and chip scarcity benefits the builders rather than hurting them. The consensus worry is whether AI demand is real or a bubble; PitchBook argues the opposite risk, that demand is outrunning supply and that the shortage persists.

"Their ability to scale depends less on appetite for AI and more on how fast they can resolve their own supply bottleneck."
Evidence: Amazon's "$25 billion" run rate and "$225 billion in revenue commitments" for Trainium, plus the statement that "that scarcity is becoming a source of leverage."

Don't try to be tactical on vintage years: pull back on check size, not on skipping funds. LPs under liquidity pressure often skip vintages or sit out funds, but the advice here is that doing so means missing top funds.

"Don't try to time vintage years, and don't skip funds. If you do need to pull back, pull back in check size," said Sarah Samuels of NEPC.
Supporting reasoning: failure to adjust models for longer hold periods "reduce[s] LPs' capacity to commit to future funds," and the recommended fix is to update the pacing model rather than abandon funds.

Private credit stress is largely invisible in headline default rates. Defaults are the standard metric, yet stress is resolved quietly, so reported numbers may understate problems.

"Direct-lending stress is generally resolved behind closed doors, through term amendments or by switching from cash interest to payment-in-kind."

3. Companies Identified

Amazon

Description: Hyperscaler with an in-house chip business (Trainium).
Why mentioned: Cited as the leading example of proprietary silicon scaling into a major business.
Quotes: "Amazon's chips business has an annual revenue run rate of more than $25 billion and over $225 billion in revenue commitments for its Trainium chips, including from Anthropic and OpenAI. It also signaled the possibility of selling chips to third parties."

Google

Description: Hyperscaler with in-house TPUs.
Why mentioned: Evidence that custom silicon is now monetizing directly.
Quotes: "Google recently began recognizing revenue from TPU system sales, signaling success in the business."

Alibaba (T-Head)

Description: Chinese tech giant with an in-house chip unit.
Why mentioned: Example of a custom chip business serving mostly external customers.
Quotes: "Over 60% of Alibaba's T-Head chips serve external customers, which the company says helps it avoid the high price premiums of commercial chips."

Nvidia

Description: Incumbent AI chip leader; also an investor in AI infrastructure.
Why mentioned: The competitor hyperscalers are increasingly challenging; also backer of Firmus.
Quotes: "...puts hyperscalers in more direct competition with Nvidia." "Nvidia-backed data center operator Firmus is considering raising a new round of up to $3 billion after shelving its IPO, Bloomberg reported."

TypeSafe AI

Description: Developer of Jev, an AI model for developers.
Why mentioned: Frontier-model challenger with a large round.
Quotes: "raised an $870 million Series A led by Andreessen Horowitz at a $7.5 billion valuation."

Firmus

Description: Nvidia-backed data center operator.
Why mentioned: Considering a raise of up to $3B after shelving its IPO.
Quotes: "is considering raising a new round of up to $3 billion after shelving its IPO."

Manus

Description: Chinese-founded AI lab.
Why mentioned: Raised $500M after its planned Meta acquisition was unwound.
Quotes: "whose planned acquisition by Meta was unwound by Chinese authorities, raised a $500 million round led by Boyu Capital and IDG Capital."

Anthropic and OpenAI

Description: Frontier AI labs.
Why mentioned: Anthropic and OpenAI are named as Trainium customers; Anthropic also in a side letter on model welfare; OpenAI on math proofs.
Quotes: "...over $225 billion in revenue commitments for its Trainium chips, including from Anthropic and OpenAI." "Anthropic wants to protect Claude from verbal abuse." "OpenAI's attempts to provide mathematical proofs for long-held problems don't make the cut yet."

SoftBank

Description: Global investor.
Why mentioned: Raising a mega-fund for AI.
Quotes: "is in talks to raise a new $100 billion fund from investors in the Gulf for AI investments, the Financial Times reported."

Catalyst

Description: Developer of AI trading technology.
Why mentioned: Seed round led by Sequoia.
Quotes: "raised a $30 million seed round led by Sequoia."

Medical Microinstruments

Description: Surgical robotics startup.
Why mentioned: Large Series D.
Quotes: "raised a $125 million Series D led by BioStar Capital."

Konst

Description: Taiwan-based AI compute operator.
Why mentioned: Series B.
Quotes: "secured a $30 million Series B led by Adata Technology."

Kapta Space

Description: Aerospace and defense startup building spaceborne radar payloads.
Why mentioned: Series A.
Quotes: "secured a $24 million Series A led by Washington Harbour Partners and Bison Ventures."

Rocscience / TA Associates / Hexagon

Description: Geotechnical and engineering software provider; its PE owner; and the Sweden-listed acquirer.
Why mentioned: Notable PE exit.
Quotes: "TA Associates agreed to exit geotechnical and engineering software provider Rocscience to Sweden-listed Hexagon in a $535 million deal."

Zetwerk

Description: Lightspeed India-backed manufacturing platform.
Why mentioned: Planned India IPO.
Quotes: "is planning an India IPO that could raise $525 million, Bloomberg reported."

City Therapeutics

Description: Arch Venture Partners-backed RNA interference drug developer.
Why mentioned: IPO.
Quotes: "is looking to raise $184.7 million in its IPO."

TRex Bio

Description: Eli Lilly-backed biotech.
Why mentioned: Completed IPO.
Quotes: "raised $116.7 million in its IPO."

vVardis

Description: Apollo-backed Swiss dental products maker.
Why mentioned: Filed for US IPO.
Quotes: "a Switzerland-based maker of dental products, filed for a US IPO."

Everton FC / The Friedkin Group

Description: English Premier League club and its owner.
Why mentioned: Control sale exploration.
Quotes: "is exploring the sale of a controlling interest in Everton FC, an English Premier League football club."

Consonance Capital Partners

Description: Healthcare-focused PE firm.
Why mentioned: Fundraise close.
Quotes: "closed on $1.35 billion for its third PE fund."

4. People Identified

Kerry Dolan

Description: Managing partner, Brinley Partners.
Why mentioned: Explains the private credit default-measurement challenge.
Quotes: "...a challenging exercise given that direct-lending stress is generally resolved behind closed doors, through term amendments or by switching from cash interest to payment-in-kind, said Kerry Dolan, managing partner of Brinley Partners."

Matt Giordano

Description: Leader of KPMG's asset management audit practice.
Why mentioned: Describes the SEC's shift in valuation scrutiny.
Quotes: "Now it's, what is your comp set and how did it change? Why did you not consider this company on the comp set?"

Sarah Samuels

Description: Chief investment officer, NEPC.
Why mentioned: Advice on adapting LP pacing models.
Quotes: "Don't try to time vintage years, and don't skip funds. If you do need to pull back, pull back in check size."

5. Operating Insights

Build or secure your own compute supply to control growth. The companies that own supply set their own timelines and costs, and can monetize it externally.

"The more successful firms are at producing in-house silicon, the more control they will have over their own growth, cost structure and build-out timelines."
"Over 60% of Alibaba's T-Head chips serve external customers, which the company says helps it avoid the high price premiums of commercial chips."

Fund managers: be ready to explain how you measure defaults and defend valuations comp-by-comp. LP and regulator questions have shifted from origination to measurement and methodology.

"LPs want to know how managers measure defaults."
"What is your comp set and how did it change? Why did you not consider this company on the comp set?"

Use third-party valuation providers for complex events like fund wind-downs. Heightened scrutiny is pushing the market toward outside validation.

"The deeper scrutiny is pushing market players to hire third-party valuation providers for more complicated processes, such as fund wind-downs."

6. Overlooked Insights

Geopolitics can unwind cross-border AI M&A, creating funding opportunities for the targets. Meta's planned acquisition of Manus was reversed by Chinese authorities, and the lab then raised $500M from Chinese investors instead, a signal about the limits of exit paths for China-founded AI labs.

"Chinese-founded AI lab Manus, whose planned acquisition by Meta was unwound by Chinese authorities, raised a $500 million round led by Boyu Capital and IDG Capital."

Emerging-manager capital formation is getting harder as large hedge funds absorb capital. Even star traders struggle to launch, which may matter for seeding strategies and for LPs looking for differentiated alpha.

"The big fish of hedge funds are leaving no room for the little guys. Large players are eating up capital, making it harder than ever to launch a hedge fund, even for star traders."