美国通信基础设施 CoreWeave (CRWV)：容量争夺持续……人人皆可获得千兆瓦级算力！

CoreWeave (CRWV): The capacity land grab continues...GWs foreveryone! Last week, CRWV announced two new deals, one with Meta for an incremental $21B , andanother with Anthropic for an undisclosed amount (we presume it is not large, but still ameaningful new logo). Our thesis was always that CRWV and other neoclouds would do finein the near term, with longer-term issues to come, but we will concede that the company hasexceeded our early 2026 expectations. We maintain our Under-Perform, but raise our targetprice to $67 on the new signings. Madison Rezaei+1 917 344 8622madison.rezaei@bernsteinsg.com Mark L. Moerdler, Ph.D.+1 917 344 8506mark.moerdler@bernsteinsg.com These deals don’t come in a vacuum - almost daily, we are catching headlines of someonesigning or building or funding new data center capacity to meet demand. Our perspectivecontinues to be that we are in a capacity land grab, and that that isn’t going to last forever.We are forecasting an easing of data center capacity in 2028 - maybe 2029 if you’reextremely bullish on AI demand. Mark Shmulik+1 917 344 8508mark.shmulik@bernsteinsg.com Gautam Chhugani+91 226 842 1416gautam.chhugani@bernsteinsg.com Over the long-term, all hyperscalers will have hybrid approaches to capacity needs. ButGWs aren’t all equal - we believe the prioritization will be: Nancy Wu+1 917 344 8545nancy.wu@bernsteinsg.com 1.Own as much real estate as you canwhere it makes sense - long-term workloads,inference use cases, markets where there is enough demand to justify owning your ownbuilding Firoz Valliji, CFA+1 917 344 8316firoz.valliji@bernsteinsg.com 2.Lease full buildings(powered shell) where you need to trade off capex for opex,need faster capacity than you can build, and/or have attractive economics or strongdevelopment partners Wenhuan Chang+1 917 344 8546wenhuan.chang@bernsteinsg.com 3.Fill in with smaller real estate leases(wholesale colo) where you have specific marketneeds but long-term visibility into requirements and can’t justify a full building Shelly Tang, CFA+1 917 344 8342shelly.tang@bernsteinsg.com 4.Use traditional cloud (if appropriate) or neocloud as a band-aidwhen you can’t getcapacity fast enough, need it in a specific location that they happen to have, are not sureyou will need a long-term workload in that location, and/or want to trial deployments ina sandbox before committing to your own build Deeksha Pandey+1 917 344 8447deeksha.pandey@bernsteinsg.com In our current land grab state, options 1-4 are happening, but as things start to ease (notyet), we maintain our conviction that the fourth category (neocloud) is the fastest and worsthit. It’s far easier to churn out of a neocloud relationship than a data center - the contractsare shorter and you don’t have to move any hardware. Eventually the LLM strategy should be similar to hyperscale, but today, these megaliths arebypassing Option #1 (and often Option #2) as they can’t yet support large-scale facilityownership or builds. They also are consuming significant capacity from traditional clouds -both GPUs and CPUs. All of this underpins our call on CRWV (UP, $67), and our long-term bearishness onthe neocloud sector, though we anticipate continuing to see stock pops on newsupdates for many months to come. BERNSTEIN TICKER TABLE PRICE TARGET CHANGE / ESTIMATE CHANGE IN BOLD O - Outperform, M - Market-Perform, U - Underperform, NR - Not Rated, CS - Coverage SuspendedGOOGL, META, AMZN estimate is Reported EPS; CRWV valuation is EV/EBITDA (x); GOOGL, META, AMZN valuation is Reported P/E (x);Source: Bloomberg, Bernstein estimates and analysis. INVESTMENT IMPLICATIONS CoreWeave (UP, $67): We are Underperform on CRWV with a revised price target of $67. We value the company using an EV/Adj. EBIT multiple(28.4x) on 2027 E Adj. EBIT per share of $5.81. Our case hinges on our longer-term revenue expectations for CRWV,which are significantly lower than street expectations. In general, we question CRWV’s ability to continue to signlarge-scale hyperscale deals as data center supply becomes more readily available (2028+) and are concerned thathyperscalers are likely to attempt head-on competition with neoclouds for enterprise business. We disagree with CRWV’s premise that its existing software moat will hold up to hyperscaler attacks, especially givenhyperscalers’ scale and ability to spend. We do not believe the structural concerns abate with increased AI demand,though the timing for our call may be pushed further out. Importantly, we do not believe the tide will shift for CRWV in the very near term and anticipate them signing anincremental $45B of deals between now and the end of 2027. We are roughly in line with consensus for 2026expectations. By 2027, we begin to diverge, and by 2028, we are ~15% below consensus revenue estimates,accelerating through the forecast window. Since our last model refresh, we have increased our Price Target by $11, reflecting a Meta deal that exceededexpectations