Contrary Research Rundown #134
The crisis in drug development finds a solution in space-manufacturing, plus new memos on Windsurf, Formation Bio, and more
Listen to the most important stories in tech on Tech Today with Contrary founder, Eric Tarczynski, with conversations across the founders of Ramp, Flexport, Moveworks, and more.
Designers, start your Config week in SF at Pixel Punch, a curated event of design lightning talks and drinks with senior design leads from Ramp, Windsurf, and Numeric.
Join us on May 6th: https://lu.ma/PixelPunch2025
Research Rundown
The Crisis in Drug Development
Despite exponential advances in scientific knowledge, drug development has become increasingly difficult, expensive, and time-consuming. This phenomenon, known as Eroom's Law (Moore's Law spelled backward), has seen drug development costs skyrocket from a mere $40 million per drug in 1953 to a staggering $2.5 billion per drug in 2023. This represents a fundamental decline in pharmaceutical innovation efficiency that hampers humanity's efforts towards providing effective treatments for diseases.
As promising drug candidates continue to fail in clinical trials and development timelines stretch into decades, the need for new technologies to improve drug development productivity has never been more urgent. The high attrition rates are starkly illustrated in notoriously difficult areas like Alzheimer's disease, where decades of research focusing on targets like amyloid plaques have yielded numerous high-profile clinical trial failures, with failure rates reaching 99.6% and over 400 failed clinical trials since 2003.
A Primer on Drug Manufacturing
The start of each pharmaceutical drug begins with the discovery of a lead active pharmaceutical ingredient (API). These API compounds, like aspirin and penicillin, constitute the core component of any pharmaceutical product, causing the desired therapeutic effect by binding to specific biological targets. Once identified, these API chemical compounds are mixed and heated in solution to achieve dissolution.
However, isolated APIs frequently encounter challenges such as poor solubility and potential degradation into toxic byproducts. To address these issues, APIs are typically combined with excipients, substances that include fillers for tablet formation, binders for API cohesion, disintegrants for tablet breakdown, or specialized coatings for controlled release. A notable example is semaglutide, the API in popular weight loss medications Wegovy and Ozempic. While traditionally administered through injection, Novo Nordisk developed an oral formulation by incorporating SNAC, a permeability enhancer that dramatically improves semaglutide absorption.
Additionally, for roughly 60% of all pharmaceutical drugs, a critical crystallization process must occur to transform dissolved APIs from their powder state into stable crystalline forms. This process involves cooling the solution to trigger recrystallization but is heavily affected by temperature, pH, and convection forces, which in turn affect the purity and performance of the final drug product.
By manipulating these variables, pharmaceutical companies can synthesize crystals of ideal size and structure. This process is crucial for the end effectiveness of a drug: a low degree of crystallinity in carbapenem antibiotics, for instance, is highly correlated with poorer chemical stability. Yet these processes often remain highly variable, difficult to reproduce, and even harder to scale reliably. For example, the resulting crystal solutions often contain a wide size distribution, necessitating additional milling or processing to ensure uniformity.
Some APIs also exhibit polymorphism – the ability of an API to form multiple stable crystal structures depending on processing conditions. Each polymorph possesses distinct properties that can significantly impact drug efficacy and manufacturing costs. For example, specific polymorphs of acetaminophen demonstrate enhanced solubility, resulting in faster pain relief onset.
While artificial intelligence through conformational search algorithms like AlphaFold and RFDiffusion have emerged to tackle the discovery challenge, accelerating the identification of potential drug candidates and predicting their interactions, these AI advancements don't inherently solve the complex physical challenges of actually producing stable, effective drugs through processes like crystallization and formulation.
The Microgravity Revolution
An emerging frontier in pharmaceutical development is the manufacturing of drugs in the absence of gravity. What was once confined to research experiments aboard the International Space Station has become commercially viable thanks to the rapid reduction in launch costs driven by companies like SpaceX.
This novel approach is the core focus of Varda Space Industries, a startup betting that the future of some pharmaceuticals lies beyond Earth's atmosphere. Founded in late 2020 by former SpaceX engineer Will Bruey and Founders Fund partner Delian Asparouhov, Varda aims to capitalize on the unique microgravity environment by offering comprehensive, space-based manufacturing services for pharmaceutical clients. The company leverages two primary advantages of the microgravity environment: the absence of gravitational forces reduces both convection and sedimentation forces, improving the crystallization process.
In terrestrial manufacturing, rapid convection currents, driven by temperature and density differences in fluids, can disrupt crystal growth in unpredictable ways. But in microgravity, diffusion, which is slower than convection, becomes the dominant force governing crystal formation. This enables more controlled and consistent crystallization, with slower growth resulting in higher purity crystals. Through precise heating and cooling cycles, Varda harnesses this unique environment to control crystallization processes that remain chaotic on Earth, creating larger, structurally superior, and more uniform API crystals.
The elimination of sedimentation forces is a second advantage. Under Earth's gravity, growing crystals inevitably yield density differences with surrounding solvents, causing them to either sink or float. This prevents uniform growth in suspension, as settled crystals develop differently than suspended ones. Crystals that settle become structurally inferior where they contact container surfaces, lacking uniform exposure to the solution.
In microgravity, these sedimentation effects disappear entirely because there is no gravitational force driving crystals to sink. The forming API crystals remain suspended in solution regardless of density differences, allowing each crystal to grow uniformly in all directions without interference from container walls. This uniform growth yields better crystal structures with superior pharmaceutical properties, including improved stability and delivery characteristics.
The evidence for these benefits already exists. In 2019, a study by Merck aboard the ISS confirmed that crystallizing cancer treatments in microgravity produced unusually uniform, stable concentrated crystalline suspensions. For pharmaceutical companies, this translates to greater purity and higher yields for commercialization. Furthermore, microgravity crystallization can favor less-stable but potentially superior polymorphs by preventing the early sedimentation that might otherwise force APIs to nucleate into their standard stable forms on Earth.
Varda Space Industries
Varda currently partners with Rocket Lab to use its Photon Spacecraft and with SpaceX for rideshare services to transport their manufacturing capsules into orbit.
Once in space, Varda's automated capsule becomes a miniature factory. Inside, unprocessed APIs undergo carefully orchestrated procedures, subjected to precise temperature changes and melt-cool cycles designed to manipulate the APIs into crystallizing into their most advantageous solid-state forms. These microgravity-induced structural characteristics become permanently embedded within the material's structure and are retained upon return to Earth.
Once processing is complete, Varda's W-1 Series re-entry capsule plummets back down to Earth at speeds exceeding 18K miles per hour for retrieval of the crystallized APIs. Designed for terrestrial landings, these capsules enable simpler, more cost-effective recovery compared to traditional ocean-based methods.
To date, Varda has launched three missions, each incrementally advancing the frontier of space-based pharmaceutical manufacturing:
W-1, launched in June 2023, successfully synthesized the metastable Form III polymorph of ritonavir in microgravity. Varda's study confirmed that this space-manufactured ritonavir exhibited excellent stability compared to its terrestrially synthesized counterpart, demonstrating the ability to produce specific polymorphs that could serve as intermediates in future manufacturing processes.
W-2, launched in January 2025, expanded the scope by testing multiple APIs across six thermal profiles, with four vials per profile – a significant advancement from W-1's seven vials of a single API under one thermal profile. This mission allowed Varda to assess drug stability under varying temperatures, measuring API degradation rates and thermal responses. These insights are crucial as the company prepares for clinical trials scheduled to begin in 2027, which will require extensive characterization of drug behavior under diverse conditions.
W-3, launched in March 2025, diversified Varda's portfolio by focusing primarily on hypersonic flight testing for the U.S. military. Still, this mission represents a major step for Varda in proving its capability for regular cadence launches of capsules into space for manufacturing purposes. Establishing this operational frequency is important for eventually achieving the economies of scale necessary to drive down the costs associated with pharmaceutical production in orbit.
A New Era of Pharmaceutical Innovation
While the company currently focuses on small molecule drugs, Varda is eventually looking to expand into the growing biologics market, allowing for the microgravity processing of complex therapeutic proteins and antibodies. In a future where Varda succeeds at scale, rare, expensive treatments for diseases like cancer could become more accessible and affordable. Drugs manufactured in microgravity could offer enhanced efficacy, simplified delivery methods, and extended shelf-life. These benefits will ultimately translate to improved patient outcomes and quality of life.
As Varda pioneers space-based manufacturing, it joins a wave of biotechnology innovation that includes breakthrough gene editing technologies like STITCHR and AI tools like AlphaFold3 for drug discovery. Together, these advances represent humanity's best hope for reversing Eroom's Law, making drug development more efficient, affordable, and accessible once again.
To read our full memo on Varda, you can read it here.
At its core, Windsurf provides AI-powered code completion, refactoring, and search capabilities, enabling developers to write, review, and navigate code more efficiently. To learn more, read our full memo here and check out some open roles below:
AI Product Engineer - Mountain View, CA
Software Engineer, ML Research - Mountain View, CA
Formation Bio is a tech-enabled drug development company that addresses the clinical trials bottleneck in the drug development process by acquiring or in-licensing clinical-stage assets from pharmaceutical and biotechnology companies and conducting clinical trials more cost-effectively and faster. To learn more, read our full memo here and check out some open roles below:
Senior Software Engineer (Applied AI) - New York, NY
Lead Product Manager - New York, NY
Bolt turns natural language prompts into fully deployed, full-stack web applications, running entirely in the browser with no setup, servers, or cloud infrastructure. To learn more, read our full memo here and check out some open roles below:
Senior Software Engineer - Remote (Global)
Director of Engineering, Data & AI - Remote (Global)
Amongst foundation model developers, Anthropic has positioned itself as a company with a particular focus on AI safety and describes itself as building “AI research and products that put safety at the frontier.” To learn more, read our full memo here and check out some open roles below:
Senior Software Engineer, Infrastructure - San Francisco, CA, New York City, NY, or Seattle, WA
Product Manager, API - San Francisco, CA or New York City, NY
Check out some standout roles from this week.
Armada | Bellevue, WA - DevOps Engineer, Security Engineer, Software Engineer (Golang), Product Manager, Cloud Network Engineer, AI Engineer
Magical | San Francisco, CA or Toronto, Ontario - AI Product Engineer
Pave | San Francisco, CA - Engineering Manager (Developer Platform), Senior Software Engineer (Developer Platform)
Ramp has launched an economics lab, giving anyone free access to near-real-time signals pulled from the card and bill-pay data of 30K US companies. Its first releases: an AI Index that measures monthly enterprise AI adoption and an Advertising Index that tracks ad-budget swings. A broader nowcast of overall business spend is slated to follow.
In an open letter, JPMorgan Chase CISO warns that reliance on a handful of SaaS and PaaS giants is creating “single points of failure” that could ripple across critical infrastructure. He urges vendors to ship secure-by-default products, modernize identity and authorization models, and provide continuous proof of control effectiveness—arguing that rushed feature releases and opaque token-based integrations are “quietly enabling cyber attacks” and weakening the global economy.
Hugging Face’s LeRobot team and The Robot Studio rolled out the SO-101, a camera-equipped, reinforcement-learning-trainable arm that enthusiasts 3-D print and assemble. It improves on last year’s SO-100 with quicker build time and stronger low-friction motors. The launch comes as Hugging Face bulks up its robotics push, having just acquired Pollen Robotics to commercialize the Reachy 2 humanoid and open its code to developers.
At a recent conference, Atreides’ Gavin Baker and Fidelity’s Karin Fronczke argued that new AI ventures—even those from ex-OpenAI stars—can’t match incumbents that already control state-of-the-art models, vast user datasets and their own super-computers. Baker called pre-product raises by Ilya Sutskever and Mira Murati “late,” praised xAI’s coming payments layer and Tesla data flywheel, while Fronczke said Fidelity is holding off on “frothy” AI-coding agent rounds for lack of proof points.
Speaking of late entrants to the AI race, draft terms for the a16z-led $2 billion raise give CEO Mira Murati a seat whose single vote equals all other directors plus one, while founders’ stock carries 100× normal voting power—an “unorthodox” setup that lets her hire-fire directors or block a sale at will, investors say.
Valar Atomics, Deep Fission, Last Energy and five Sun-Belt states say the Nuclear Regulatory Commission’s “one-size-fits-all” rules make small-modular reactors impossible to license, so they’re asking a federal court to let states take over. Supporters call it the only way to get new reactors on the grid; critics warn a patchwork of state rules could undo hard-won public trust in nuclear safety.
Unveiled at LlamaCon, the Llama-4–powered assistant breaks out of Facebook, WhatsApp & Instagram into its own app with a social “Discover” feed, taking direct aim at ChatGPT, Gemini, Claude and Grok. The move supports Mark Zuckerberg’s goal of reaching over 1 billion users and follows Meta’s plan to pour up to $65 billion into AI infrastructure this year.
A report in the Wall Street Journal says Sam Altman and Satya Nadella’s relationship is fraying: Altman wants more Azure capacity and freedom to restructure OpenAI, while Nadella is funding in-house models (via Mustafa Suleyman’s $650 million Inflection hire) and pushing Copilot. Contract clauses let Microsoft block OpenAI’s for-profit flip—and let OpenAI cut Microsoft off from its top models—leaving the 2019-era allies preparing for life apart even as they still depend on each other.
Waymo signed a preliminary pact with Toyota to co-develop an autonomous-vehicle platform and eventually bring Waymo’s “generalizable” driver to next-gen personally owned Toyotas, while also adding Toyota models to the Waymo One robotaxi fleet. Woven by Toyota will supply software muscle as both firms pitch the tie-up as a step toward a zero-accident future.
Flower and Vana crowd-train “Collective-1” LLM using Flower’s open-source Photon framework to spread training across hundreds of internet-linked GPUs and Vana’s trove of user-contributed private posts, producing a 7-billion-parameter model without a data center; they’re now scaling toward 30 billion and eyeing a 100 billion-parameter multimodal model later this year, a route that could let smaller players pool compute and data to rival today’s AI giants.
Glean, now past $100 million ARR, is finalizing about $150 million in new funding that would lift its price tag from December’s $4.6 billion to roughly $7 billion. Wellington Management will lead, joining backers Kleiner Perkins, Lightspeed and Sequoia
In Epic Games v. Apple, Judge Yvonne Gonzalez Rogers found the iPhone giant “willfully” defied her 2021 order and banned it from charging any commission or limiting links for purchases made outside iOS apps, even referring Apple for possible criminal-contempt review. Apple will appeal, while Epic says Fortnite will return to the U.S. App Store next week.
With GMV now past $3 billion and a ~$5 billion valuation, Whatnot, the collectibles-to-clothing marketplace says users watch livestreams 80 minutes daily and the average buyer snaps up 12+ items/week. Sign-ups in 2025 already exceed all of last year as niche communities, referral credits and new seller perks fuel growth—while TikTok’s uncertain U.S. fate and tariff talk could further tilt shoppers toward Whatnot’s second-hand focus.
Apple taps Anthropic’s Claude for a “vibe-coding” Xcode variant where Claude Sonnet can write, debug and UI-test code, replacing the glitch-prone Swift Assist; a public launch is TBD, underscoring Apple’s shift toward external AI partners alongside ChatGPT and (soon) Gemini.
At Contrary Research, we’ve built the best starting place to understand private tech companies. We can't do it alone, nor would we want to. We focus on bringing together a variety of different perspectives.
That's why applications are open for our Research Fellowship. In the past, we've worked with software engineers, product managers, investors, and more. If you're interested in researching and writing about tech companies, apply here!