They raised 92.8M guided diluted shares, added $15.8M of debt, and told you to focus on the $471M monthly mid-2027 target.
Thesis: The Q3 ramp is real and likely beats. 800G at 10x YoY growth with 5x guided sequentially against a memory-shortage-driven 100G drag is genuine structural demand. But the mid-2027 target of $471M/month in data center revenue requires roughly 13x the current $36M monthly data center run rate in twelve months, funded by $565M quarterly CapEx, 5% dilution per quarter, and inventory building 35% against 27% revenue growth. Every bottleneck AOI names (DSP, TIA, memory, switches) sits with third parties. The equity is a timing trade, not an investment.
Verdict: HOLD — Conviction: MEDIUM
Catalyst: 1.6T qualification completes within 2-3 weeks per Thompson Lin, then $70M+ of 1.6T revenue in Q4 with Q1 doubling if material supply cooperates. A Chinese transceiver ban would re-rate the US manufacturing angle overnight.
Key Risk: 92% of revenue sits with three customers (42%, 26%, 24%), two of them hyperscalers actively insourcing optics. A single qualification loss or insourcing decision removes the thesis. Plus: DSP/TIA vendors rank AI as their 4th priority, so AOI has zero leverage over its critical supply chain.
The Tell: Thompson Lin unprompted reframed Q3 growth: 'because of memory issue for the switch, actually, 100G will decrease by $20 million-$25 million. Otherwise, our Q3 revenue, the growth should be more than 50%.' He is volunteering a number that says the guide understates organic growth, inviting the street to model $25M higher than guided. Then committed beyond guidance on 1.6T: 'If it double, I will not be surprised for Q1 revenue.' Both unprompted. He wants the street to look past the headline guide.
Friction Level: HIGH_FRICTION — Bull sees a capacity-constrained beat-and-raise machine with in-house laser ownership. Bear sees a company funding a forecast it does not control through 5% dilution per quarter with 92% of revenue in three counterparties who are vertically integrating optics themselves.
2026Q2 · Aug 6, 2026
5 of 5 readings agreed5 HOLDRead the full call
Capacity Tripling, Inventory Outrunning Sales, 92% Revenue From Three Buyers
The Q3 ramp is real and likely beats. 800G at 10x YoY growth with 5x guided sequentially against a memory-shortage-driven 100G drag is genuine structural demand. But the mid-2027 target of $471M/month in data center revenue requires roughly 13x the current $36M monthly data center run rate in twelve months, funded by $565M quarterly CapEx, 5% dilution per quarter, and inventory building 35% against 27% revenue growth. Every bottleneck AOI names (DSP, TIA, memory, switches) sits with third parties. The equity is a timing trade, not an investment.
Thompson Lin unprompted reframed Q3 growth: 'because of memory issue for the switch, actually, 100G will decrease by $20 million-$25 million. Otherwise, our Q3 revenue, the growth should be more than 50%.' He is volunteering a number that says the guide understates organic growth, inviting the street to model $25M higher than guided. Then committed beyond guidance on 1.6T: 'If it double, I will not be surprised for Q1 revenue.' Both unprompted. He wants the street to look past the headline guide.
Inventory rose 35% sequentially ($206.2M to $278.8M) against 27% revenue growth. Days of inventory went from ~124 to 132. Management frames it as raw materials staged for the ramp, but revenue has orders to fill. If demand softens, this becomes a write-down.
The $200M+ 1.6T order comes from one hyperscaler with deliveries only beginning late Q3 and no cancellation terms disclosed. Thompson Lin says 'don't get too wrapped up on that particular order, it's just the very beginning' which manages expectations down on firmness.
Management explicitly states 'we have complete book already from now until Q2 next year' and demand runs 20-40% above capacity. They cannot convert excess demand into revenue. High utilization is presented as bullish but it caps near-term upside.
Gross margin at 29.8% is down from 30.4% year-ago while management points to 40% long-term target driven by 1.6T and laser mix. Data center mix blamed for headwind while legacy margin structure hasn't improved. The 40% target requires laser (55-65% GM) and ELSFP (50%+ GM) revenue that is 'very limited production now' and ramps into 2027-2028.
$565.5M of CapEx in one quarter against $1.1B annual revenue, including $280M prepayments on equipment. $538.8M raised on ATM. Management is putting real capital to work against booked backlog, not just talking.
Fifth consecutive record quarter. EPS $0.06 clearing top of $0.03 guidance. Q3 guide midpoint $272.5M is 42% above the quarter just printed. 800G revenue 10x YoY, 2x QoQ, with 5x Q3 ramp guided. This is consistent execution.
CPO and ELSFP revenue is 'very limited production now.' CFO admits 'we can't make enough of them to be involved in their current first-generation deployments.' 400K units/month in 2028. The AI/CPO story is a 2028 story priced into today's capacity claims.
+78
CONFIDENT"Thompson Lin volunteered superlatives unprompted: 'best in the industry, the best in the world,' gave forward commitments beyond guidance ('If it double, I will not be surprised for Q1'), and committed publicly to $471M monthly mid-2027. Only hedged on the Chinese ban: 'it's a little early to say.'"
Full transcript with specific capacity numbers, product mix percentages, customer concentration breakdown, CapEx detail, and direct Q&A. Management gave exact mid-2027 math and named specific bottlenecks. Only gap is cancellation terms on the $200M 1.6T order.
1.6T qualification completes within 2-3 weeks per Thompson Lin, then $70M+ of 1.6T revenue in Q4 with Q1 doubling if material supply cooperates. A Chinese transceiver ban would re-rate the US manufacturing angle overnight.
92% of revenue sits with three customers (42%, 26%, 24%), two of them hyperscalers actively insourcing optics. A single qualification loss or insourcing decision removes the thesis. Plus: DSP/TIA vendors rank AI as their 4th priority, so AOI has zero leverage over its critical supply chain.
The verdict is public. The case behind it is Pro: the forensic reading of guidance and language, the bull and bear cases, the red flags, and the play. It opens to everyone on Nov 4, 2026.