Anjney Midha: from a16z Oxygen to AMP's compute-and-capital strategy

Anjney Midha’s documented career links two related strategies: using compute access as portfolio support at Andreessen Horowitz, then combining compute and capital through a separate venture called AMP. The record supports describing him as a technical founder, investor, and infrastructure strategist. It does not support the label “GPU kingmaker,” precise private-return estimates, or claims that one person controls access to the AI market.

The current-role boundary matters. By August 2026, The Washington Post described Midha as a former a16z general partner and the founder of AMP. The Post reported that AMP had raised around $1.3 billion for its venture business, but AMP has not published a full audited account of its capital, infrastructure, or economics.

Answer in brief

a16z hired Midha as a general partner in 2023 and later described him as a leader of Oxygen, a program intended to help portfolio companies obtain more flexible GPU capacity. In August 2026, The Washington Post reported that he was building an independent compute grid through AMP and that the firm had raised around $1.3 billion for its venture business. The article attributed the grid’s design and utilization claims to Midha.

Anthropic independently lists AMP PBC as an investor, corroborating participation but not the reported check size or fund terms. This analysis is current through September 13, 2026 and distinguishes firm announcements, named reporting, and analytical inference.

Verified career record

a16z’s 2023 appointment announcement describes Midha as the co-founder and former CEO of Ubiquity6, later a Discord executive, and an investor focused on AI. It is an employer-written introduction, so its praise and claims about founder relationships should be read as firm positioning.

Stanford’s current profile lists him as a visiting scientist in applied physics. The institutional profile is useful for the affiliation but does not establish a broader research publication record or validate investment outcomes.

Oxygen’s documented purpose

In a 2024 a16z explainer, Midha and Derrick Harris described Oxygen as a program helping portfolio companies navigate compute needs. The stated problem was that startups could be forced into long contracts or unfavorable capacity commitments during a GPU shortage. The firm said Oxygen could aggregate demand and offer more suitable duration and pricing.

That source does not publish audited GPU inventory, utilization, total capital committed, customer-level prices, or profitability. Numbers reported elsewhere should not be presented as official Oxygen disclosures. The defensible claim is that a16z used compute access as a portfolio service and investment tool.

Why compute changes venture support

Traditional venture support emphasizes capital, recruiting, customer introductions, and governance. Frontier-model and infrastructure companies add another scarce input: compute. A startup may have financing but still lack timely access to accelerators, power, networking, or a serving stack.

Compute support can create value in several ways:

• shortening the time between funding and a model experiment;
• reducing exposure to an oversized long-term cloud contract;
• helping a company compare training, fine-tuning, and inference options;
• linking technical plans to capital and utilization milestones.

It also creates conflicts. A fund or affiliate arranging capacity may influence vendor selection, portfolio allocation, or financing terms. Buyers and founders need transparency about ownership, pricing, priority, and data access.

Moving from a16z to AMP

The Washington Post’s August 2026 interview and report describes Midha as a former a16z general partner and founder of AMP PBC. It says AMP aims to pool underused compute into an independent grid for startups, universities, and other labs. The operating claims in the interview are Midha’s account and should not be treated as independently audited utilization data.

The move is strategically coherent: Oxygen addressed compute inside an existing portfolio, while AMP was reported as a separate vehicle designed around infrastructure and investment. Coherence is an analytical observation. The public record does not disclose all commercial relationships between AMP, a16z, capacity providers, and portfolio companies.

Reported AMP fund

The Washington Post reported that AMP had raised around $1.3 billion, including capital from a16z and Y Combinator, for its venture business. The same report said AMP had been investing in companies including Anthropic, OpenRouter, and River AI. The article does not provide audited fund statements, so the amount remains a reported figure rather than a verified close supported by a regulatory filing.

Anthropic’s Series G announcement lists AMP PBC among significant investors. This primary company disclosure corroborates AMP’s participation. It does not state AMP’s check size, ownership percentage, governance rights, or fund structure.

That difference illustrates good source practice: one source can verify participation while another supplies a reported amount. Combining them does not convert the amount into an audited fact.

Infrastructure thesis

The economic problem behind AMP is that compute contracts have multiple risk dimensions at once:

Dimension	Startup risk	Investor or capacity response
duration	demand changes before contract ends	shorter or staged commitments
utilization	paid accelerators sit idle	pooling and workload scheduling
technology	hardware or model architecture changes	optionality across stacks
power and delivery	announced capacity arrives late	milestone and site diligence
financing	large prepayment strains runway	separate asset or credit capital

A combined compute-and-capital vehicle might coordinate these decisions. It could also concentrate risk if utilization, power delivery, or customer demand misses plan. The thesis is not validated by assets under management alone.

Investor diligence requirements

Large infrastructure claims require evidence beyond a press headline. A serious diligence package should include:

1. signed capacity, not only memoranda or planned sites;
2. delivery dates, power availability, and interconnection status;
3. hardware ownership and depreciation assumptions;
4. contracted demand, utilization, and customer concentration;
5. separation between fund capital, credit vehicles, and operating entities;
6. pricing and priority rules when portfolio companies compete for capacity;
7. security boundaries for models, data, and workload telemetry.

Without this evidence, gigawatts, GPU counts, and fund targets remain plans or reports rather than operating capacity.

Oxygen and AMP are not the same thing

Oxygen is described by a16z as its portfolio compute program. AMP is a separate venture reported as founded by Midha. The people and thesis may overlap, but the entities, capital, contracts, and governance should not be treated as interchangeable.

This distinction protects readers from three common errors: assigning Oxygen assets to AMP, treating a16z portfolio investments as AMP investments, and assuming Midha’s continuing venture-partner relationship gives AMP control over a16z programs. None of those conclusions follows from the cited sources.

Company disclosures versus analysis

The source types in this profile carry different weight:

• a16z pages establish what the firm said about Midha and Oxygen;
• Anthropic establishes that AMP participated in its financing;
• The Washington Post provides named journalistic reporting and a direct interview about AMP’s role and fund;
• Stanford establishes a current academic affiliation;
• this article’s explanation of the compute-and-capital model is analysis.

No source here provides audited AMP financial statements or a complete asset inventory. That remains the central evidence gap.

Public evidence gaps

The available record does not establish:

• Midha’s personal investment economics;
• AMP’s full limited-partner list, fee structure, or ownership;
• the exact amount AMP invested in Anthropic as an audited fact;
• delivered data-center capacity, GPU inventory, or utilization;
• confidential a16z or AMP allocation decisions;
• future fund closes, investments, or infrastructure projects.

Reported fundraising is not the same as deployable capital, and planned capacity is not the same as energized capacity.

A decision framework for founders

Founders considering a compute-linked investor should compare the combined offer with separate capital and capacity contracts. Ask whether the package reduces total risk or merely hides one cost inside another.

Key questions include: Can capacity scale down as well as up? Who bears idle hardware risk? Is financing conditional on using an affiliated provider? Can workloads move without punitive exit costs? Who can inspect operational data? What happens if equipment delivery slips? The answers should appear in contracts and technical schedules, not only in an investor narrative.

Bottom line

Anjney Midha’s public significance comes from treating compute access as part of venture strategy. a16z’s Oxygen page documents the original portfolio-service model; The Washington Post documents his work at AMP; Anthropic confirms AMP as an investor. Together they support a credible evolution from compute assistance to a broader compute-and-capital thesis.

They do not support a mythic gatekeeper label or certainty about private fund economics. AMP’s long-term importance will depend on delivered capacity, aligned contracts, transparent conflicts, and returns across a full investment cycle. Those outcomes remain to be demonstrated.