We could, and arguably should have waited with comments before having a clearer picture (we tend to use writing as a thinking tool) but we think the situation is more clear and turns out to be better. Some additional observations.
Project-based
Revenue from the company's rack integration segment is project-based.
TSS closed a new multi-year agreement with Dell smoothing this out and providing TSS with a baseload (Q3CC):
The base case scenario for volume stipulated in agreement is similar to or greater than the peak volume of AI-enabled rack integrations that we delivered earlier this summer.
And from the Q3 earnings PR:
Even more encouraging is the recent signing of an agreement with one of our largest customers to build similar volumes of AI-enabled racks for the next several years. To address both the anticipated volume and the ever-increasing power demands expected from the next several generations of AI racks, we plan to relocate from our current integration facility to a new, larger facility in the area in early 2025.
The big driver behind the recent expansion was the first phase of the buildout of xAI, Elon Musk's AI company, this started in the last month of Q2 (June 2024) but it also ended, from the Q3CC:
The initial program carried well into the third quarter.
If the first phase of xAI was finished before the end of TSS's Q3, this would explain the somewhat disappointing rack integration revenue (insofar as a 361% increase to $7.6M can be disappointing..).
We actually entertained the possibility of xAI being done for now some weeks ago in private with some people familiar with the situation but people disagreed.
Another important takeaway from this is that Dell smoothing out and providing a very high minimum guarantee is itself very positive, it essentially removes the downside of the project nature of its rack integration business.
xAI
The June-September first phase of xAI build-out is going to be followed by a second phase in the coming months, doubling capacity:
In the race to dominate the next wave of artificial intelligence, Elon Musk has taken a bold step forward with the launch of a new supercluster under his xAI venture. The system, powered by a staggering 100,000 Nvidia (NVDA) H100 GPUs, is located in Memphis, Tennessee, and was built in record time—just four months. Dubbed "Colossus," this supercomputer aims to become the most powerful AI training system in the world, accelerating the development of Musk's AI projects, including the free-speech chatbot, Grok… Looking ahead, xAI's plans for Colossus reflect Musk's ambition to lead the AI revolution. The system's capacity will continue to expand, potentially doubling to 200,000 GPUs in the coming months.
What's more, rather than splitting the second phase 50/50 between Dell and SuperMicro, the whole second phase seems to be going to Dell (and by extension TSS, see here and here):
It’s hardly surprising given the execution problems at Super Micro.
TSS Management Guidance
Management guidance was basically flat for the coming quarters (Q3CC):
We expect our volume of rack integrations to be at or above the volumes that we experienced since June 2024… We expect the first half of 2025 to be in line with our second and third quarters of 2024 in aggregate.
This is likely the main reason for investors selling post-earnings:
However, does this flat guidance contain the second phase of the xAI build-out, or is it just the baseload from Dell?
We know, because the guidance is exactly the baseload! If you're still unsure compare the quote about the baseload from the beginning of the article with the quote above about guidance. See the similarity?
Something of the coming second phase of the xAI build-out might be in that baseload as Dell is likely to have better visibility. It's unlikely to be much because however big it will be, it is very time-constrained and they can't guarantee a multi-year baseload on the basis of that.
What we do know is that Dell getting Super Micro's share in the second phase isn't in the guidance, management specifically told us that (Q3CC):
Pradeep brought up the super micro. I don't want to ever speak badly of anybody. I don't wish any ill will on anyone in our industry. If there's a demand increase because of something related to that, we will adapt and adjust to it. And that would be a spike that we're not planning for, but we can prepare and adjust to.
So with guidance containing only Dell’s baseload, investors disappointed with the guidance will soon have something to celebrate when the second phase of the xAI arrives, twice as large as the first one. Very little of that is likely to be in the baseload, if any.
Secular tailwinds
The reasonings above are about a fairly limited time horizon, the next couple of quarters. If you can look beyond that, the picture only gets better, from Morgan Stanley:
"Dell's AI server momentum remains strong, with our recent checks pointing to about $20 billion of AI server revenues in fiscal 2026, 56% higher than prior Morgan Stanley estimates," he said in a client note. That sales target equates to 94% year-over-year growth, thanks to sales of systems with Nvidia (NVDA) AI chips, he said.
Woodring said his $20 billion estimate represents his base case scenario and assumes no market share gains from Super Micro. However, the probability is rising for his bull case scenario for Dell to get $40 billion in AI server sales in fiscal 2026. Dell's fiscal 2025 ends in late January. Under the bull case scenario, Dell would take one-third of Super Micro systems business via competitive wins, Woodring said.
We already know Dell will take share from Super Micro, so Dell's AI server business will grow well beyond that +94% base case. Here is Mizuho:
Mizuho analysts raised their estimates and price target for Dell Technologies (NYSE:DELL), highlighting its increasing share in the AI server market as NVIDIA (NASDAQ:NVDA) leads the charge toward a projected $2 trillion data center AI infrastructure (DCAI) opportunity by 2028-29.
"Dell is gaining share in AI servers as capex percolates and NVDA GPUs supply expands to Tier 2 CSPs and Enterprise/Sovereign customers," Mizuho stated, adding that this trend is expected to significantly boost Dell's position through 2025.
Dell's focus on competitive pricing and expanding GPU offerings are seen as positioning it well to capitalize on these shifts. Mizuho also flagged the potential for Dell to secure future orders from major players like Apple (NASDAQ:AAPL) as the AI server market continues its rapid growth.
Then there are additional opportunities with the replacement market (servers have a three-year life span), the enterprise market (which is likely to be several times the size of the hyperscaler market), and with the enterprise market coming MDC and facilities management opportunities.
Conclusion
It's not surprising that after an epic run-up of the shares, some air went out of the stock.
But insofar as this was the result of the flat guidance, we think that investors will be pleasantly surprised fairly soon.
It's difficult for us to imagine that demand for TSS services will remain flat against a backdrop of strong secular growth in the AI server market with Dell winning share and the second phase of xAI, 2x the size of the first one for Dell/TSS, arriving in the coming months.
It's always difficult to point to the exact bottom, what's less difficult is to argue that after the retreat, the shares are already attractively valued again. This company has a compelling growth opportunity for the coming years.
We made an important change in the article, this part:
["However, does this flat guidance contain the second phase of the xAI build-out, or is it just the baseload from Dell?
We know, because the guidance is exactly the baseload! If you're still unsure compare the quote about the baseload from the beginning of the article with the quote above about guidance. See the similarity?
Something of the coming second phase of the xAI build-out might be in that baseload as Dell is likely to have better visibility. It's unlikely to be much because however big it will be, it is very time-constrained and they can't guarantee a multi-year baseload on the basis of that."]