Procure.FYI

In a time of Marvell (@MarvellTech, $MRVL) reporting flat data center revenue growth and a disappointing outlook this week, I am once again reminding you to pay attention to the slowing pace of data center CapEx input, things industry talkshop hypebeasts won't tell you: procure.fyi/us-data-center…

Wrote part two of my ongoing series on fiber optics and networking. I think most people underestimate how important and crucial networking is, so I wrote about a somewhat overlooked story: NVIDIA’s 2019 acquisition of Mellanox for $6.9B. This deal fused Israel, the U.S., and China geopolitics, turning networking into NVIDIA’s scale moat amid the AI boom. Link in comment 🔗 AI hype fixates on GPUs and CUDA software, but networking is the glue that enables bottleneck-free, exabyte-scale data flows in massive clusters. Mellanox rose from a 1999 Israeli startup pioneering InfiniBand with RDMA and fiber optics for ultra-low-latency HPC, eventually expanding into Ethernet, RoCE, and SmartNICs, perfectly complementing NVIDIA GPUs. The acquisition faced heavy scrutiny (U.S./EU/China delays), closed in 2020, and is now subject to an ongoing 2024 Beijing probe. But in hindsight, it supercharged NVIDIA’s AI dominance: a $6.9B purchase turned into $20B+ in FY26 networking revenue (56% YoY growth), with 20 to 50% efficiency gains driven by fiber-powered tech.$nvda

A more revealing look at @anduriltech’s R&D engine would explain its roll-up-style acquisitions of companies that secured government contract vehicles—specifically, the 3 to 15 Indefinite-Delivery Vehicles (IDVs) it has amassed over the past five years. @ColossusReview Data via @procureFYI

Ahem, speaking of budget cut, based on contract data from over 70 software vendors tracked by @procureFYI , federal award growth in the first quarter of fiscal year 2025 has already exceeded $1.5 billion, highest in Q1 history (Note: Data is still incoming due to a typical data delay from the #DoD.) —surpassing Q1 of 2024, which totaled under $1.5 billion. #DOGE

If you allow me, I am going to yap about @coreweave working capital, again $crwv. The company posted a record $606 million in adjusted EBITDA on $982 million in revenue, yet only generated $61 million in operating cash flow—a striking gap that points to serious cash-conversion issues. The breakdown shows why: accounts receivable increased by $639 million, pushing DSO (days sales outstanding) to nearly 97 days, meaning roughly a full quarter’s worth of revenue is sitting unpaid on the books. Simultaneously, deferred revenue—an indicator of upfront customer cash—flipped from a $1.44 billion inflow in Q1 2024 to a $15.9 million outflow this quarter. That swing signals a structural shift: newer contracts are increasingly backloaded, with less cash collected upfront. Yet CoreWeave’s spending didn’t slow: It logged $1.41 billion in CapEx for property / equipment / capitalized software usage, and interest expense ballooned to $264 million, up 6.5x from a year ago, as more GPU clusters moved from “construction-in-progress” to active service—cutting off interest capitalisation and pushing full charges to the P&L. To fund this growing working capital gap, CoreWeave raised $785 million in new debt in Q1, and is reportedly lining up another $1.5 billion in unsecured high-yield bonds just weeks after its IPO. This cash cycle—spend first, earn later—only works if every element stays synchronized: hardware deliveries, cluster go-lives, billing ramp-up, and customer payments.

Got time to update my U.S. data center construction spending infograph tonight. March’s Census data feels like a subtle but important inflection point. #Datacenter construction spending reached an all-time monthly high of $34.8B annualized, yet the sharp acceleration we saw through 2023 and early 2024 appears to have stalled. (Data access here: procure.fyi/us-data-center…) Whether it’s supply chain bottlenecks, power/grid constraints, land availability, or simply the market pausing after an aggressive AI-driven buildout—it’s hard to say definitively. But the data shows the momentum has clearly cooled. Here’s what stood out: • U.S. data center construction spending hit a record $34.8B annualized in March 2025 • Year-over-year growth slowed to 33%, the lowest since early 2023 • Monthly growth has hovered around zero for several months The sector remains historically strong, but this could mark a shift from the “explosive build” phase to a more measured, steady expansion.

Palantir @PalantirTech $pltr had one of its strongest Q1 performances from the federal government in recent years—if not its largest ever (over $120M so far). Note that Q1 is typically a quieter period for federal contracting compared to Q3, which marks the end of the fiscal year

Palantir (@PalantirTech $pltr ) FY2025 year-to-date revenue by U.S. federal department

As we noted last month, a resolution of the U.S.–China tariff dispute will likely involve an agriculture‑and‑LNG deal as well open.substack.com/pub/procurefyi…

Palantir @PalantirTech $pltr current fiscal year federal revenue source

Quick reminder that @scale_AI had one large federal contract—W911QX20C0051, 110m, from 2020—which has since ended. I, as @procureFYI, haven’t seen any significant traction since.

@TrendSpider Palantir ( @PalantirTech ) did $99 M in federal‑government contracts in Q1 2024. So far, Q1 2025 is tracking at about $75 M—DoD data for Q1 is still pending, but it’s highly unlikely they’ll match their peak 2022 growth levels $pltr

Last quarter, $PLTR posted its first YoY EPS slowdown since 2022. With earnings just two weeks away, the pressure's on to prove the growth story is still intact.

@qubitium @CoreWeave Yes, you need call them for onboarding open.substack.com/pub/procurefyi…

Pre-review of CoreWeave 😅 ❌ Can't even sign-up. Need invite. ❌ Not a public cloud? Close to useless for SMB. @CoreWeave

@wandb @CoreWeave @l2k truetheta.io/concepts/ai-to… $crwv

Today we announced that we are being acquired by @CoreWeave, the AI Hyperscaler. 🪄🐝 We could not be prouder or more excited to join forces with this team. Our CEO, @l2k, wrote a blog post with more details: wandb.ai/wandb/wb-annou…

@NasdaqExchange @CoreWeave See comprehensive $crwv S-1 analysis here open.substack.com/pub/procurefyi…

Welcoming @CoreWeave to the @NasdaqExchange Opening Bell. twitter.com/i/broadcasts/1…

Want a much deeper (but still plain language understandable) dive on the comparison? See my piece with @WangUWS -> interconnect.substack.com/p/coreweave-vs…

Choosing a GPU cloud provider is a high-stakes decision for any AI/ML team. Specs and technical details matter—but what do they actually mean for your team? It’s not just about clock speed or FLOPS. It’s about whether your models will train faster, deploy more reliably, and scale more affordably in practice—and whether your team has the infrastructure skills to make it all work. To help enterprise teams think through this, @DuaneJRich and I wrote a deep dive on #CoreWeave $crwv — what makes it compelling, what to watch out for, and how to think about pricing, orchestration, and performance trade-offs. Link in comment👇🏽