Together AI $800M 시리즈 C 라운드 완료, Aramco Ventures 주도, NVIDIA/Vista 참여

Four out of every five dollars invested in venture capital this week flowed into artificial intelligence infrastructure. While the headline figure reached $9.9 billion, the underlying market dynamics reveal a highly concentrated landscape. Beyond the mega-rounds dominating the news cycle, here is what unfolded across the broader ecosystem.

**Market Snapshot**

Total disclosed venture capital for the week of June 29 reached $9.94 billion, a 958% increase from the prior week’s $939 million. Disclosed rounds jumped to 32 from 12 (+167%), while the median check size dipped slightly to $21 million from $26.25 million (-20%). Infrastructure captured approximately 79% of total capital, up from 71%. Mergers and acquisitions totaled $11.1 billion, driven by Iridium’s $8 billion exit and Schneider Electric’s $3.1 billion acquisition of industrial AI software firm Cognite. New startup entries in the database surged to approximately 1,930 from 495 (+290%), though this spike largely reflects a batch of automated, minimally detailed submissions, many labeled Israeli, inflating the raw count.

Stripping away the three mega-rounds, the underlying market ran at a typical summer pace of $2.9 billion across 29 rounds. The $9.9 billion headline is accurate but misleading. The critical metric is the $7.8 billion deployed over 48 hours for infrastructure, driven by capital sources such as Aramco Ventures, NVIDIA, General Atlantic, and unnamed sovereign-adjacent funds. These are not traditional venture investors; they are growth equity firms, corporate strategics, and energy-sector players treating AI compute as a utility asset class.

**Infrastructure Thesis: Three Entry Points**

In a span of just two days, four companies raised or sought a combined $7.8 billion. Three focused on the physical layer that powers AI—compute, power, data center fabric, and inference clouds—while the fourth scaled AI video generation. Application-layer companies, which build software atop this infrastructure, were notably absent from the headlines. Their checks averaged $25.8 million at the Series A stage and $12.3 million at seed. Consequently, the infrastructure-to-application funding ratio hit approximately 4:1 by dollar volume, marking the widest weekly divergence recorded this year.

Crusoe is negotiating a $3 billion round that would triple its valuation to approximately $18 billion, according to a July 2 Bloomberg report. Its business model relies on energy arbitrage: identifying stranded power assets—electricity generated far from demand centers—and co-locating AI compute directly at those sites. This strategy yields electricity costs significantly below hyperscaler rates, offset by remote geography and high capital intensity. Investors backing Crusoe are implicitly wagering that AI’s projected electricity consumption—which the International Energy Agency estimates will triple global data center usage by 2030—cannot be met by urban grid infrastructure. Crusoe’s core argument is straightforward: the bottleneck is power, not silicon.

Separately, Las Vegas-based hyperscale data center operator Switch is seeking $2 billion in fresh capital. Operating campuses across the United States and internationally, Switch has benefited heavily from hyperscaler co-location demand. Together, Crusoe and Switch’s July 2 announcements represent $5 billion deployed in a single day for AI’s physical container: power and space.

Together AI’s $800 million Series C, announced July 1 and led by Aramco Ventures with participation from NVIDIA, Vista Equity Partners, and General Catalyst, targets a different layer of the stack: inference cloud. The platform runs open-source AI models on dedicated compute, offering a serverless environment it claims delivers roughly twice the performance of leading alternatives. Aramco Ventures’ leadership is particularly notable; the Saudi state oil company’s venture arm has been systematically accumulating positions in AI infrastructure firms whose compute demands require stable, large-scale power supplies—an area where Aramco holds deep expertise. Together AI reported that annual bookings surpassed $1.15 billion in Q2 2026 and plans to leverage the new capital to expand its public cloud capacity fiftyfold over the next five years.

This close follows Groq’s $650 million raise the previous week. Groq and Together AI now occupy identical market positions with comparable balance sheets, both pitching dedicated AI inference at lower costs than hyperscalers and targeting the same enterprise developers. One will inevitably consolidate the other’s customer base. The compressed announcement timeline—Groq on June 22, Together AI on July 1—suggests parallel fundraising processes that closed within nine days, narrowing the window before competitive dynamics materialize in sales data.

**Kling AI and the Scale of Chinese Video AI Capital**

Kling AI closed a $2 billion round led by General Atlantic, cementing an $18 billion valuation for the Kuaishou-backed video AI company. The July 2 announcement formalized a Series A process initially disclosed in June, which had sought the same $2 billion at the same valuation rather than representing additional capital.

Kling generates video from text and image prompts and has secured meaningful international traction despite geopolitical constraints limiting some Chinese AI firms’ global expansion. At $18 billion, its valuation surpasses Runway’s $4 billion assessment in 2024 and places it squarely in the territory traditionally reserved for foundation model developers rather than application-layer video generators. The valuation gap between Kling and its U.S.-based peers in video AI is large enough to reflect fundamentally different capital structures, regulatory environments, and strategic timelines between the two ecosystems.

**Emerging Microtrends**

Beyond infrastructure and video generation, several notable developments emerged this week. Straiker closed a $64 million round after publishing research demonstrating that 91% of productivity agent attacks result in undetected data theft, confirming that AI agent security has moved past the initial skepticism phase. Meanwhile, three independent startups launched AI search visibility tools this week without coordinating, highlighting an emerging category that currently lacks a standardized name but has already attracted funded participants. Additionally, Schneider Electric’s $3.1 billion purchase of Cognite underscores a broader pattern: industrial incumbents are acquiring the data layer rather than attempting to build it internally.