AI Startups

orbital data centers, a chipmaker taking on Nvidia, and a Sequoia-backed security model: AI Startups to Watch for August 22, 2026

AI startups to watch August 22, 2026: orbital data centers, a chipmaker taking on Nvidia, and a Sequoia-backed security model. Five disruptive moves.

Daniyal

· 4 min read
A startup rocket launching toward orbit above a city skyline at dusk

The pattern in the last 24 hours is clear: the money is betting against the Nvidia monoculture and against the congested power grid. A UK inference-chip startup is circling a six-billion-dollar valuation after landing an Anthropic order, a Tel Aviv foundation-model shop grabbed a Sequoia-led seed, and a startup put a GPU in orbit. Here are five moves worth your attention on August 22, 2026.

Starcloud: a $250 million bet on orbital inference

Starcloud raised $250 million in a funding extension led by Manhattan West Ventures, with participation from Nvidia and Cisco, per TechCrunch. Nvidia put in about $25 million, according to a person familiar with the deal; Benchmark, EQT, Soma, NFX, 776, Cedar Capital, and Goanna Capital also joined.

Starcloud is the only company known to run a terrestrial Nvidia H100 GPU in orbit and the first to train a model on one, versus other space GPUs built for lightweight edge processing. It is sharing those results with Nvidia as the chipmaker develops its first purpose-built space GPU. Why it matters: orbital compute sidesteps the terrestrial constraints of power, land, and water that are throttling data centers on Earth.

The risk is real. The plan depends on SpaceX rapidly reusing Starship, and SpaceX this week said it will push the next catch-and-reuse attempt to year-end or early 2027. Co-founder Philip Johnston put it plainly: no SpaceX launch capacity in 2029, and Starcloud struggles. The upside, if the launch economics work, is an orbital inference layer no terrestrial player can easily match.

Fractile: taking a swipe at Nvidia's inference business

London-based Fractile is in advanced talks to raise roughly $600 million at a $6.5 billion pre-money valuation, per Bloomberg, up about sixfold from its roughly $1 billion valuation in May. The jump follows a reported $250 million initial order from Anthropic to supply inference chips, with a larger contract under discussion.

Founded in 2022 by Oxford PhD Walter Goodwin, Fractile designs in-memory-compute inference chips it says run LLMs 25 to 100 times faster at around a tenth of the cost of GPUs. Its May round, about $220 million near $1 billion, was led by Accel, Founders Fund, and Factorial. Why it matters: a small UK lab is landing orders before shipping a production chip, which targets 2027, and mounting a direct challenge to Nvidia's inference franchise alongside Cerebras, Etched, and Groq. Flag this as a reported round still in discussion, not a closed deal.

Corma: a foundation model for the defense side of cyber

Corma raised $60 million in seed funding led by Sequoia Capital, with Khosla Ventures and Coatue, emerging from stealth. No valuation was disclosed. The company, based between Tel Aviv and San Francisco, builds a foundation model purpose-built to mine audit logs, security events, and network flows, rather than a general-purpose LLM. It says early Fortune 100 and Fortune 500 deployments cut threat-response times by around 94 percent. Its "Corma Workforce" agent layer runs on-prem, which matters for regulated customers.

Why it matters: this is a bet on the defense gap, the idea that attackers using general AI outrun defenders who do not. Vertical security models could end up as the moat, and Sequoia is putting its name behind it. Note that coverage is fresh, but some outlets dated the seed earlier in August, so treat it as a this-week announcement.

Apollo Atomics: shrinking the nuclear reactor for the power crunch

Apollo Atomics raised $31 million in seed funding, $26 million in equity and $5 million in debt, led by FCVC with Y Combinator, Paul Graham, Alumni Ventures, and Nucleation Capital. The MIT spinoff is redesigning the steam generator to make pressurized-water reactors about 40 times smaller and factory-built, with a roughly two-year install timeline. It claims more than 20 gigawatts in letters of intent, and former NRC chair Christopher Hanson sits on its advisory board.

Why it matters: data centers are running out of power, and this startup is aimed squarely at the bottleneck, competing with Oklo, Kairos, and Radiant. If factory-built small reactors work, they could become the default power source for the next wave of AI infrastructure.

Domyn: Europe's check on frontier-model sovereignty

Italian LLM builder Domyn disclosed it has raised $1.1 billion in combined debt and equity over two years, per its founder, including about 650 million euros in project financing closed in 2024 at a 1.7 billion euro valuation. It is reportedly seeking an additional billion euros.

Why it matters: this is a test of whether Europe can sustain a capital-intensive frontier model developer on a debt-plus-equity structure instead of ceding the frontier to US and Chinese labs. Domyn, founded in 2016, is fronting the European AI-sovereignty push. It has the balance-sheet ambition; the open question is whether it has the talent and compute to match.

The takeaway

Across these five moves, one theme dominates: the new front lines are physical and geopolitical. Compute is moving to orbit, chips are being designed to unseat the incumbent, power is being re-engineered, and capital is polarizing around sovereignty. The startups that win the next phase may not look like software companies at all. Yesterday's August 20 roundup captured the same shift, and the momentum has only accelerated.

Background worth having before you act on the above:

More on AI Startups

Browse every AI Startups story →