Venture Bytes #119: The Silent War Above

Great power rivalries have been confined to land, sea, and air for millennia. But today, the most consequential battles may unfold far above, with space becoming a key warfighting frontier where dominance over satellites can determine the outcome of conflicts before the first shot is fired on the ground. From intelligence gathering to communications to missile defense, space systems are now the invisible backbone of modern warfare. To stay ahead, the US can’t rely solely on traditional defense primes. The future of space superiority will hinge on VC-backed start-ups with their potential to become a vital tool in America’s future defense playbook.

The militarization of space originated in the 1950s, with early rockets initially designed as weapon-delivery systems. During the Cold War era, both the US and USSR explored space's military potential, developing anti-satellite capabilities and other space-based military technologies. The 1967 Outer Space Treaty attempted to ensure peaceful use of space while prohibiting nuclear weapons in orbit, but notably did not prevent conventional weapons or military activities

For the US, Space doctrine has evolved from “protect and defend” to planning offensive counterspace options. Earlier in April 2025, the US Space Force (USSF) released its Space Warfighting framework, describing how the US forces might seize the orbital high ground through both defensive and offensive measures. Today’s armed forces rely on space like never before. Satellites provide communications, navigation, and intelligence across all war theaters. The US Army, for example, notes that “approximately 70% of the Army’s weapons are space-enabled.”

‍

Space superiority is now a prerequisite for battlefield success – Gen. Chance Saltzman, the USSF’s chief of space operations.

‍

This strategic shift is driving demand for new hardware and technologies. The Pentagon is explicitly seeking satellite systems that can maneuver, inspect, and even neutralize other satellites. Concepts like on-orbit servicing drones, robotic refueling tugs, and agile small-sat constellations are now tied to military objectives as much as commercial ones. Even cyber warfare and electronic attack now extend into space. As DoD notes, China and Russia have fielded “weapons aimed at US space capabilities” including lasers to blind sensors and anti-satellite missiles.

‍

‍

‍

The technological boundary between civilian and military space is blurring – a missile-warning satellite is a legitimate target, and a commercial launch provider might soon ferry US spacecraft to defend a failing national asset. The 2022 use of SpaceX's Starlink in Ukraine was a watershed moment, proving that commercial infrastructure can become mission-critical in warfare.

Beyond military considerations, the economic stakes of space development have grown dramatically. The overall global space economy reached an impressive $596 billion in 2024 and is projected to grow to $944 billion by 2033, according to Space Economy Report 2025 by space consulting and market intelligence firm Novaspace. This growth reflects a sweeping fusion of economic and security interests in orbit. Notably, 17% of 2024’s space demand was defense-related, and that is expected to surge as the Pentagon deepens its partnerships with private firms through USSF and Space Development Agency.

Venture capital is following this strategic momentum. U.S. VC funding in space-tech more than doubled in one year—from $5.3 billion in FY2023 to $11.8 billion in FY2024, per Space Capital. The surge reflects a structural maturation of the sector with investors increasingly backing companies with clear revenue paths, strong DoD engagement, and defensible dual-use technology that can plug directly into military needs.

Several defense-focused space startups are already emerging as strategic assets. True Anomaly, founded by Space Force and Air Force veterans, is a standout. The company has developed a full-stack platform—spanning both software and spacecraft hardware—designed for rendezvous and proximity operations. These maneuvers, which include close-range monitoring and potential intercepts of other satellites, are vital for countering threats from adversarial co-orbital systems. True Anomaly fills a crucial gap in the US national security space architecture, where such dynamic space operations remain underdeveloped. The company, which recently raised a Series C round, has secured contracts across hardware, software, and services.

Slingshot Aerospace, based in Colorado, is building the digital backbone for space domain awareness. Its proprietary software ingests data from hundreds of sensors to create a real-time, continuously updating digital twin of the space environment. Slingshot’s virtual reality simulations allow warfighters and mission planners to “see” space in ways never before possible.

As threats in orbit multiply, commercial surveillance firms are becoming essential eyes in the sky. California-based LeoLabs leads the charge in this domain. The start-up runs a 10-site global radar network that tracks over 20,000 objects in low Earth orbit, delivering precision data on satellite trajectories and real-time collision risks. Its intelligence is now critical not just for commercial operators, but also for allied defense forces demanding sharper visibility in space.

‍

‍

ZeroSearch – AI’s Next Efficiency Play?

What if AI could search like Google, without ever touching a search engine? Alibaba’s ZeroSearch does exactly that, with up to 88% less cost and offering the same intelligence.

Currently, many AI systems rely on external search. ChatGPT and Bing Chat hit Microsoft’s Bing API and Google’s Bard taps Google Search. Startups like Perplexity AI merge multiple search results to answer queries. All of these incur non-trivial costs with commercial search APIs ranging from $3 to $15 per 1,000 queries, depending on volume and plan. For example, Alibaba notes that making ~64,000 Google search requests via SerpAPI costs about $586.7, whereas ZeroSearch’s 14B model can handle that in-house for about $70.8, leading to an 88% reduction in training expenses.

ZeroSearch’s novelty lies in internalizing search. Traditionally, an LLM-based assistant that needs outside facts either uses an index (retrieval-augmented generation) or fires off web searches. ZeroSearch avoids that by training an LLM to simulate web search internally. In practice, the system lightly fine-tunes the base LLM to generate relevant documents (snippets, web pages, etc.) for any given query, mimicking a search engine. This is followed by running reinforcement learning rollouts where the simulation LLM produces answers. Alibaba’s researchers “gradually degrade the quality” of the synthetic documents during training, forcing the model to adapt and improve.

ZeroSearch’s ripple effects could be broad. By lowering both cost and technical barriers, ZeroSearch could level the playing field. ZeroSearch breaks the dependency on cutting reliance on “expensive and gated platforms” such as Google or Bing. In future, we might have AI systems to develop increasingly sophisticated capabilities through self-simulation. A startup or research lab might no longer need a six-figure Google search bill to train an advanced assistant.

Additionally, the controlled training environment means the model can learn from consistent data. This may improve answer relevance since it is not misled by irrelevant or malicious web pages. By filtering out irrelevant or misleading content at the source, ZeroSearch has the potential to produce models that are not only more accurate, but more dependable.

That said, ZeroSearch is not without trade-offs. The simulated search model is inherently bound by its training data. Unlike a live engine like Google, it cannot dynamically pull in fresh information or respond to rapidly evolving events. This limitation is particularly relevant in use cases that demand real-time awareness, such as financial news or breaking policy changes. However, Alibaba has hinted at future “hybrid” systems that could use occasional real searches when the model is unsure.

ZeroSearch opens intriguing questions. If search costs can be brought down materially, new startup categories will emerge. For example, companies building domain-specific AI assistants (legal, medical, etc.) could use ZeroSearch-style methods to ingrain specialized search without expensive data licensing. Open-source AI vendors may incorporate similar techniques, offering cloud services that are cheaper per query.

The frontier of AI innovation is shifting from raw scale to design-level breakthroughs. ZeroSearch is a reminder that breakthroughs often come from rethinking the underlying approach as much as they come from scaling data or compute. Even with current constraints, ZeroSearch offers a blueprint for making models more autonomous, cost-effective, and less dependent on proprietary external APIs. Alibaba’s approach does not necessarily dethrone Google or Bing today, but it gives the industry a powerful new tool. As the company’s researchers themselves put it, ZeroSearch “enables developers and businesses – especially [smaller] enterprises – to independently develop their own RL framework without costly search engine interactions.” That autonomy could be a game-changer for many.**