#365: Flexport Acquires Shopify’s Logistics Business, & More

Last week, Flexport—a late-stage startup focused on digitizing freight-forwarding—agreed¹ to acquire Shopify’s logistics businesses in exchange for 13% of its equity. Flexport will become Shopify’s official logistics partner and take over its technology and employees. The divesture ends Shopify’s venture into the logistics space, which began in 2019² when it bought 6 River Systems, a warehouse robot startup. Then, in 2022, Shopify acquired Deliverr,³ a last-mile delivery provider, for $2.1 billion, and invested⁴ in Flexport. It announced a logistics partnership⁵ with Flexport in early 2023.

In a letter⁶ to its employees, Shopify CEO Tobi Luetke said that Flexport is “the best builder and operator in the world of logistics” and that the divesture would enable Shopify to focus on its core competency—building commerce software.⁷ The stock deal could give Shopify the option to acquire Flexport in the future.

In our view, the partnership is exciting, as it paves the way for Flexport to expand into last mile logistics and e-commerce fulfillment. As a result, Flexport will be able to tap into Shopify’s ~$200 billion⁸ gross merchandise volume (GMV), roughly 15% of which is cross-border.

[1] Shopify 2023a. “Shopify Announces First-Quarter 2023 Financial Results; Agrees To Sell Shopify Logistics to Flexport.” Available here.
[2] Smith, J. 2019. “E-Commerce Platform Shopify Buys Warehouse-Robot Startup.” The Wall Street Journal. Available here.
[3] Marchese, A. 2022. “Shopify to Buy E-Commerce Fulfillment Specialist Deliverr for $2.1 Billion.” The Wall Street Journal. Available here.
[4] DC Velocity 2022. “Flexport lands enormous $935 million backing for freight forwarding platform.” Available here.
[5] Cision PR Newswire 2023. “The Flexport App Launches on Shopify to Make Global Trade Easy for Merchants Everywhere.” Available here.
[6] Shopify 2023b. “Company News: Important team and business changes.” Available here.
[7] Shopify 2023c. “Shopify Inc. (SHOP) Q1 2023 Earnings Call Transcript.” Seeking Alpha. Available here.
[8] Shopify 2023d. “Shopify Announces Fourth-Quarter and Full-Year 2022 Financial Results.” Available here.

During ARK’s weekly Brainstorm, we highlighted a talk delivered by Kfir Schreiber, CEO of DeepCure, at the Broad Institute’s Machine Learning in Drug Discovery Symposium. In What’s Holding Back AI in Drug Discovery,¹ Kfir described the current limits to implementing artificial intelligence (AI) in drug discovery and highlighted key bottlenecks to be resolved before AI can reach its potential.

While AI models have become useful for identifying target candidate molecules, Kfir noted that sometimes they do not predict features like synthesizability, toxicity, or selectivity. In addition, AI-generated compound predictions are not paired widely with the automated custom chemical synthesis necessary to test the efficacy of compounds rapidly in later preclinical steps. By integrating computational prediction, automated synthesis, and downstream assays into one workflow that iteratively informs and improves AI tools, Kfir holds out hope that AI will deliver on its potential for prolific drug discovery.

[1] Schreiber, K. 2022. “What’s Holding Back AI In Drug Discovery?” YouTube. Available here.

Last week, a Google researcher leaked an internal document¹  highlighting the growing threat to both Google and OpenAI from open-source artificial intelligence (AI) models that are emerging in the AI arms race. The document highlights that, in the first of its kind, Meta’s LLaMA foundation model² disclosed the large language model weights that will help accelerate innovation in the open-source community. It also emphasized the importance of Low Rank Adaptation (LoRA)³ in fine-tuning models efficiently and cost-effectively.

Since Meta released LLaMa in February, open-source AI projects have progressed rapidly. The community has developed models running on low-cost devices—enabling the fine-tuning of LLMs on laptops—and released unrestricted models without any content filters. While it might have a slight edge in quality, the gap between Google and open-source AI models appears to be closing.

The leaked document recommends that Google move away from closed-source development and prioritize learning from and collaborating with external open-source projects. It also suggests that Google eschew its focus on larger and larger models, which delays progress, and iterate quickly on smaller models.

While the leaked document paints a dreary picture of two companies—Google and OpenAI—at the cutting edge of AI development, it represents one employee’s point of view and may not reflect the broader views in the AI community.

[1] Patel, D. and Ahmad, A. 2023. “Google ‘We Have No Moat, And Neither Does OpenAI’: Leaked Internal Google Document Claims Open Source AI Will Outcompete Google and OpenAI.” Semi Analysis. Available here.
[2] Cox, J. 2023. “Facebook’s Powerful Large Language Model Leaks Online.” Vice. Available here.
[3] Hu, E.J. et al. 2021. “LoRA: Low-Rank Adaptation of Large Language Models.” Available here.