a16z Plans to expand capacity GPU its computing power pool and is in talks with Switch and Volta

AT A GLANCE

Several Silicon Valley venture capital firms have begun securing GPU computing capacity to provide resources for their portfolio AI startups.

Article

Silicon Valley VCs begin stockpiling GPU：a16z 20000 cards and seek further expansion, competing for AI Startups need more than just funding

Dongcha Beating AI breaking news: Several Silicon Valley venture capital firms are competing to secure GPU compute capacity to provide exclusive resources for their portfolio AI startups.a16z As early as 2024, it had already established a compute pool of approximately 20000 units GPU and is now in talks with data center companies Switch and Volta to further expand it. Another venture capital firm, General Catalyst, plans to build a data center for its portfolio companies. This year,May also specifically recruited the former head of NVIDIA DGX Cloud to oversee its AI infrastructure business.

startups may not be able to lease AI in time even after raising funds. This year,GPU Microsoft and other major cloud providers were reported to be prioritizing compute capacity for internal teams and major customers, leaving some startups facing waiting periods of several months.April, Microsoft and other major cloud providers were reported to be prioritizing compute capacity for internal teams and major customers, leaving some startups facing waiting periods of several months.a16z previously disclosed that, amid tight compute capacity, startups seeking prices previously available under -month contracts may be required to sign contracts for 3 years. Yet model training requires intensive short-term use of large amounts of GPU, while inference demand after product launch is difficult to predict; locking in compute capacity years in advance can easily waste capital.

a16z The approach of August，a16z is to centrally procure and lease compute capacity through a program called Oxygen, then allocate it to its portfolio startups, allowing them to receive pricing and contract terms normally available only to major customers. This model is now extending upstream. This year, 300000000 USD co-led Volta's 2400000000 USD financing round, bringing the latter's valuation to AI. It invests both in GPU startups and in data-center companies that provide compute capacity, seeking to control both capital and

supply channels.Crucible Capital In addition, compute capacity itself is becoming a VC business. Small VC 576 built a cluster with B300 GPU Nvidia AI cards for its portfolio startups to test products, while selling part of the compute capacity to a frontier GPU lab. This both secures the needs of portfolio companies and turns idle compute capacity into revenue. For VCs, providing

can also attract founders who need substantial compute capacity, increasing their chances of winning sought-after investment opportunities.GPU However, this model also makes VCs assume risks previously borne primarily by cloud providers and data-center operators.

Key points

a16z-card 2024 compute pool, while General Catalyst is also developing related data-center capabilities. 20000 In GPU, it established a compute pool of approximately
cards and is in discussions with Switch and Volta about expansion.May General Catalyst plans to build data centers for portfolio companies and, this year,DGX recruited the former head of Nvidia AI Cloud to lead its
a16z infrastructure business.
This year August，a16z co-led Volta's 300000000 USD financing round, with Volta's valuation reaching 2400000000 USD。
Crucible Capital built a cluster with 576 NVIDIA B300 GPU and sold part of its computing capacity to a frontier AI laboratory.
GPU updates rapidly, so long-term leasing or building data centers may create a risk of idle computing capacity.

AI-assisted interpretation

The following is analysis, separate from reported facts. Verify important claims independently.

In addition to providing capital, venture capital firms are beginning to centrally purchase, lease, or build GPU resources and then provide them to their portfolio AI startups. This can help startups cope with GPU tight supply, long wait times, and the high cost of long-term contracts, but venture capital firms also need to bear the risks of idle computing capacity and equipment obsolescence.

Why it matters to readers

This shows that GPU computing capacity is becoming AI an important resource and potential business opportunity in venture investing.

For newcomers, the key is to understand:AI The company requires substantial GPU computing power to train and run models; investment institutions may enhance their service capabilities and investment appeal by controlling the supply of computing power.

Risks and unknowns

a16z It is unclear whether the further expansion discussions with Switch and Volta have been finalized; the original text did not provide a formal announcement.
It is unclear whether General Catalyst's data center construction plan has been implemented; the original text did not specify.
Volta financing and 2400000000 USD Whether Volta's financing and valuation have been formally confirmed by the company or lead investor is not stated in the original text.
The original text provided no independent verification materials for claims that cloud providers such as Microsoft prioritize internal teams and major customers, while startups wait for months.

a16z Plans to expand capacity GPU its computing power pool and is in talks with Switch and Volta

Article

AI-assisted interpretation

Why it matters to readers

Risks and unknowns

Related Developments

Related concepts

Keep following

Security Center