Menlo Park startup Amp has raised more than $1.3 billion to build a shared “AI grid,” a pooled computing network intended to help startups, universities and other organizations gain access to the scarce data center capacity needed to develop advanced artificial intelligence systems.

The company was founded by Anjney Midha, a serial entrepreneur and former partner at Andreessen Horowitz, who is positioning Amp as a counterweight to the concentration of AI infrastructure among a small number of powerful technology companies. Amp’s backers include Andreessen Horowitz, Y Combinator and several cloud computing providers, according to reports on the financing.

The core idea is to treat AI compute less like a private asset controlled by individual companies and more like a shared utility. Amp plans to buy computing power from data center operators in the United States and overseas, then make that capacity available to a coalition of AI developers and research organizations. Midha has compared the model to an electricity grid: users do not all need to own the power plants, but they need reliable access to the supply.

The raise lands at a moment when computing capacity has become one of the most decisive constraints in AI. Training frontier models requires large clusters of specialized chips, high-bandwidth networking, significant power and long-term data center commitments. Major cloud companies and top AI labs have been striking multi-billion-dollar infrastructure deals, leaving smaller companies to compete for limited capacity in a market where priority access often depends on size, capital and existing cloud relationships.

Amp’s pitch is that collective bargaining can shift that balance. Instead of each startup trying to negotiate separately with data center operators, Amp would aggregate demand from multiple companies and institutions. That could make the group a more serious buyer and allow participants to obtain computing resources that would otherwise be unavailable or too expensive.

The coalition model also broadens what participants may bring to the table. Some companies could contribute money. Others may contribute training data, models, engineering capacity or other resources useful to the group. In theory, the structure could create a shared infrastructure layer for independent AI teams that do not have the balance sheets of Amazon, Google, Microsoft, Meta, OpenAI or Anthropic.

Several AI startups have agreed to participate in or help use the pooled compute effort, including Periodic Labs, which is focused on AI for scientific discovery, and ElevenLabs, the AI voice company. The presence of those companies underscores the range of demand for compute beyond general-purpose chatbots. AI developers working on science, speech, image generation, video and coding systems all face pressure to secure reliable chip access as model performance becomes more compute-intensive.

For Silicon Valley, Amp’s financing highlights a changing venture landscape. Traditional startup funding is no longer only about hiring engineers, building products and acquiring customers. In AI, access to GPUs and data centers can determine whether a young company can train a competitive model at all. That has pushed investors and founders to search for new funding structures tied directly to compute capacity.

The model could also influence how universities and research labs participate in advanced AI. Academic researchers have warned for years that frontier AI development is becoming too expensive for many institutions outside a small set of corporate labs. A shared compute pool could give some researchers access to infrastructure they could not secure individually, though Amp has not yet demonstrated at scale how pricing, allocation and governance would work for a broad mix of commercial and academic users.

There are open questions about whether Amp can acquire enough capacity to make the grid meaningful. Data center operators are already under heavy demand from hyperscalers, AI labs and enterprise customers. Power availability, chip supply, networking equipment and permitting constraints have all become practical limits on AI infrastructure expansion. A pool of financing may help Amp negotiate, but it does not eliminate the physical limits of where new compute can be built and powered.

There is also a strategic question about whether a shared grid reduces concentration or simply creates a new gatekeeper. If Amp becomes an intermediary between smaller AI developers and infrastructure providers, it could expand access while also gaining influence over which companies get compute, on what terms and with what obligations to share data or models. Those details will matter as the company moves from fundraising and coalition-building into actual capacity allocation.

Amp’s official materials frame the effort as a response to what it calls the need for an independent AI grid. The company argues that frontier AI progress increasingly depends on scaling compute and that focused teams can produce major advances if they can obtain the infrastructure required to compete. That argument reflects a broader belief in the AI industry that model capability is closely tied to compute scale, even as researchers continue to debate how much progress can come from better algorithms, data quality and efficiency improvements.

The startup’s timing is notable because AI infrastructure has become a priority for both private markets and policymakers. U.S. cloud providers, chipmakers and data center companies are investing heavily to meet demand from AI model developers. At the same time, the concentration of AI infrastructure has raised policy concerns about competition, national security, energy use and the resilience of the U.S. technology base.

For now, Amp is best understood as an ambitious infrastructure-finance experiment rather than a proven alternative to the hyperscale cloud model. Its $1.3 billion raise gives it credibility and negotiating power, but its long-term impact will depend on whether it can secure enough high-quality compute, price it attractively and build trust among companies that may be both collaborators and competitors. If the model succeeds, it could become a new route for AI startups to access the machinery of frontier development without surrendering the field to the largest technology companies.