{"id":3491076,"date":"2026-07-17T12:00:00","date_gmt":"2026-07-17T12:00:00","guid":{"rendered":"https:\/\/techingeek.com\/index.php\/2026\/07\/17\/reasons-behind-the-initial-gpu-investors-pivoting-to-inference-chips-in-a-400-million-agreement\/"},"modified":"2026-07-17T12:00:00","modified_gmt":"2026-07-17T12:00:00","slug":"reasons-behind-the-initial-gpu-investors-pivoting-to-inference-chips-in-a-400-million-agreement","status":"publish","type":"post","link":"https:\/\/techingeek.com\/index.php\/2026\/07\/17\/reasons-behind-the-initial-gpu-investors-pivoting-to-inference-chips-in-a-400-million-agreement\/","title":{"rendered":"Reasons behind the initial GPU investors pivoting to inference chips in a $400 million agreement"},"content":{"rendered":"
<\/div>\n
\n

General Compute, a cloud startup specializing in AI inference, has secured a $400 million loan from Upper90, a technology investment firm. This could mark the inaugural instance of using inference-specific chips as collateral \u2014 these chips are engineered to efficiently execute pre-trained AI models quickly, as opposed to the pricier chips utilized for creating the models initially.<\/p>\n

This financing indicates that markets are reacting to concerns regarding the costs of AI tools and tokens by seeking infrastructure capable of running open-source models at lower costs compared to the latest LLMs from leading labs.<\/p>\n

Founded by CEO Finn Puklowski and CTO Jason Goodison, General Compute raised a $15 million seed round in May to develop an inference neocloud powered by silicon from SambaNova, a chip producer backed by Intel. (Neoclouds are specifically designed for AI tasks, contrasting with the general-purpose infrastructure provided by traditional hyperscalers like AWS or Azure.)<\/p>\n

The company’s SN50 chips are crafted for inference. They boast power efficiency and do not necessitate costly water-cooling systems, allowing for faster deployment across a wider range of data centers compared to GPUs. General Compute claims that these new chips will deliver 16 times the inference speed of GPU-based clouds.<\/p>\n

The hurdle lies in acquiring a significant quantity of these chips, particularly for a newly established company.<\/p>\n

Upper90 co-founder and CEO Billy Libby, a former quantitative trader from Goldman Sachs, had a strategy for this: In 2021, his firm financed the GPU acquisitions of Crusoe, the energy-centric data center startup, which he believes was the inaugural loan against advanced chip value.<\/p>\n

Conventional lenders were hesitant about such agreements due to the risks and uncertainties surrounding GPU depreciation. However, as CoreWeave transformed chips-backed loans into a viable business model followed by a remarkable IPO, this type of financing has become increasingly prevalent.<\/p>\n

\u201cWhen we financed Nvidia GPUs as the pioneers in that space, the market was inefficient,\u201d Libby shared with TechCrunch. \u201cWe could really assemble something as early participants and be compensated for the risk.\u201d<\/p>\n

Now that GPUs are much better understood and possibly over-purchased, Upper90 is looking at firms like General Compute to capitalize on the forthcoming wave of the AI expansion. \u201cWe believe open-source models will be significant, and last year we searched for a player focused on inference,\u201d Libby noted. \u201cNot everyone needs a supercomputer, but they do need inference and AI.\u201d<\/p>\n

This perspective has gained momentum, with companies offering access to open models, such as OpenRouter and Fireworks, securing new funding rounds at substantial valuations. New models like Kimi\u2019s K3 have shown they can compete with the latest releases from Anthropic and OpenAI on coding benchmarks. Additionally, emerging chipmakers like Groq and Cerebras have caught the attention of acquirers and public markets.<\/p>\n

General Compute\u2019s capacity to access chips beyond Nvidia\u2019s ecosystem is crucial for the same reasons. TensorWave, another AI infrastructure player, is making a similar wager by partnering with AMD. As more alternatives to Nvidia become available, compute providers not tied to Nvidia agreements might gain an edge in offering cost-efficient inference.<\/p>\n

\u201cSeveral chips are beginning to scale that have excellent [total cost of ownership], or that can operate significantly faster than Nvidia, but the buyer pool is limited,\u201d Puklowski mentioned. \u201cBy collaborating with Upper90, this signifies more than just a \u2018cool startup received funding for compute.\u2019 This represents the initial indication of capital organizing itself and the disintegration of Nvidia\u2019s monopolistic hold.\u201d<\/p>\n<\/div>\n

When you purchase through links in our articles, we may earn a small commission. This doesn\u2019t affect our editorial independence.<\/em><\/p>\n","protected":false},"excerpt":{"rendered":"

<\/div>\n
\n

General Compute, a cloud startup specializing in AI inference, has secured a $400 million loan from Upper90, a technology investment firm. This could mark the inaugural instance of using inference-specific chips as collateral \u2014 these chips are engineered to efficiently execute pre-trained AI models quickly, as opposed to the pricier chips utilized for creating the models initially.<\/p>\n

This financing indicates that markets are reacting to concerns regarding the costs of AI tools and tokens by seeking infrastructure capable of running open-source models at lower costs compared to the latest LLMs from leading labs.<\/p>\n

Founded by CEO Finn Puklowski and CTO Jason Goodison, General Compute raised a $15 million seed round in May to develop an inference neocloud powered by silicon from SambaNova, a chip producer backed by Intel. (Neoclouds are specifically designed for AI tasks, contrasting with the general-purpose infrastructure provided by traditional hyperscalers like AWS or Azure.)<\/p>\n

The company’s SN50 chips are crafted for inference. They boast power efficiency and do not necessitate costly water-cooling systems, allowing for faster deployment across a wider range of data centers compared to GPUs. General Compute claims that these new chips will deliver 16 times the inference speed of GPU-based clouds.<\/p>\n

The hurdle lies in acquiring a significant quantity of these chips, particularly for a newly established company.<\/p>\n

Upper90 co-founder and CEO Billy Libby, a former quantitative trader from Goldman Sachs, had a strategy for this: In 2021, his firm financed the GPU acquisitions of Crusoe, the energy-centric data center startup, which he believes was the inaugural loan against advanced chip value.<\/p>\n

Conventional lenders were hesitant about such agreements due to the risks and uncertainties surrounding GPU depreciation. However, as CoreWeave transformed chips-backed loans into a viable business model followed by a remarkable IPO, this type of financing has become increasingly prevalent.<\/p>\n

\u201cWhen we financed Nvidia GPUs as the pioneers in that space, the market was inefficient,\u201d Libby shared with TechCrunch. \u201cWe could really assemble something as early participants and be compensated for the risk.\u201d<\/p>\n

Now that GPUs are much better understood and possibly over-purchased, Upper90 is looking at firms like General Compute to capitalize on the forthcoming wave of the AI expansion. \u201cWe believe open-source models will be significant, and last year we searched for a player focused on inference,\u201d Libby noted. \u201cNot everyone needs a supercomputer, but they do need inference and AI.\u201d<\/p>\n

This perspective has gained momentum, with companies offering access to open models, such as OpenRouter and Fireworks, securing new funding rounds at substantial valuations. New models like Kimi\u2019s K3 have shown they can compete with the latest releases from Anthropic and OpenAI on coding benchmarks. Additionally, emerging chipmakers like Groq and Cerebras have caught the attention of acquirers and public markets.<\/p>\n

General Compute\u2019s capacity to access chips beyond Nvidia\u2019s ecosystem is crucial for the same reasons. TensorWave, another AI infrastructure player, is making a similar wager by partnering with AMD. As more alternatives to Nvidia become available, compute providers not tied to Nvidia agreements might gain an edge in offering cost-efficient inference.<\/p>\n

\u201cSeveral chips are beginning to scale that have excellent [total cost of ownership], or that can operate significantly faster than Nvidia, but the buyer pool is limited,\u201d Puklowski mentioned. \u201cBy collaborating with Upper90, this signifies more than just a \u2018cool startup received funding for compute.\u2019 This represents the initial indication of capital organizing itself and the disintegration of Nvidia\u2019s monopolistic hold.\u201d<\/p>\n<\/div>\n

When you purchase through links in our articles, we may earn a small commission. This doesn\u2019t affect our editorial independence.<\/em><\/p>\n","protected":false},"author":2,"featured_media":3491077,"comment_status":"open","ping_status":"closed","sticky":false,"template":"Default","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[],"class_list":["post-3491076","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-uncategorized"],"_links":{"self":[{"href":"https:\/\/techingeek.com\/index.php\/wp-json\/wp\/v2\/posts\/3491076"}],"collection":[{"href":"https:\/\/techingeek.com\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/techingeek.com\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/techingeek.com\/index.php\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/techingeek.com\/index.php\/wp-json\/wp\/v2\/comments?post=3491076"}],"version-history":[{"count":0,"href":"https:\/\/techingeek.com\/index.php\/wp-json\/wp\/v2\/posts\/3491076\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/techingeek.com\/index.php\/wp-json\/wp\/v2\/media\/3491077"}],"wp:attachment":[{"href":"https:\/\/techingeek.com\/index.php\/wp-json\/wp\/v2\/media?parent=3491076"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/techingeek.com\/index.php\/wp-json\/wp\/v2\/categories?post=3491076"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/techingeek.com\/index.php\/wp-json\/wp\/v2\/tags?post=3491076"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}