Sponge (@Sol_Sponge) header
Sponge (@Sol_Sponge) profile picture

In search of asymmetric bets

New York, NY
Joined August 2021
$PENG just became an NVIDIA AI Factory Specialized Partner — an invitation-only designation within the NVIDIA Partner Network recognizing full-stack AI factory design, deployment and operations expertise. This is not a marketing badge. NVIDIA gates this program. The proof points PENG cited to earn it: → Haein AI Factory in Korea (PENG + NVIDIA + SK Telecom) — one of Korea’s largest GPU-as-a-Service sovereign AI clusters → Deepgram enterprise voice AI deployment (PENG + Dell + NVIDIA) Also dropped quietly in the same press release: → PENG won Dell Technologies 2026 Global AI Partner of the Year → Q3 earnings call: July 7 In ~90 days $PENG has: ✅ Named NVIDIA AI Factory Specialized Partner ✅ Named Dell AI Partner of the Year ✅ SKT $200M strategic investment (Haein = the proof of concept) ✅ Stock +20% after Q2 beat + raised guide Analysts still catching up. $7 FY27 EPS at 25x = $175. Clock is ticking.
$PENG Penguin Solutions Becomes an NVIDIA AI Factory Specialized Partner "Joining a select group of NVIDIA Partner Network (NPN) solution providers. Penguin brings proven experience in designing, building, deploying, and operating full-stack, NVIDIA-based AI factory infrastructure for enterprise-scale AI factories that power AI training, inference and agentic AI workloads."
2
706
$PENG just doubled its FY26 growth guide mid-year. AI now 60% of revenue. Memory segment +53% YoY. First production CXL KV cache server shipping today. They’re building the missing tier in the AI memory hierarchy. $53 today. Base $95. Bull $185.
AI has a memory problem. $PENG has the solution. And it’s only a matter of time before the price reflects that. Every long AI conversation builds a “KV cache” inside the GPU. Long docs, agent tasks, big context windows — the cache overflows fast. When it overflows, GPUs recompute, stall, or refuse the request. That’s why inference is so expensive.🧵
1
12
2,507
ServiceNow just raised AI guidance. AI/Now Assist grew by +130% YoY last quarter, with deals including 3+ Now products increasing by 70%. New Guidance: $1.5B (raised +50%) $NOW
1
454
We can only wish brother
1
13
Wait till we find out tomorrow this guy bought $PENG 🐧
1
1
5
952
With the $CBRS IPO this week - The question that naturally arises - how do I position for it? $WYFI is the most interesting pure play. They’re explicitly named as a colocation tenant relationship, and a successful Cerebras IPO validates their pivot from Bitcoin mining to AI infrastructure. Small-cap, so it would move more on sentiment than fundamentals shifting.
2
5
2
16
4,490
Common theme for a lot of these players is execution risk. Lack of bandwidth to meet demand isn’t a bug — it’s the whole game. $AAOI missed earnings and is now pumping up 8% premarket. Mid-2027 capacity: $5.6B annualized. Customer demand: ~$7.3B. The $1.7B gap doesn’t go to AAOI unless they hit every milestone: • InP substrates secured • 1.6T qualification on schedule • Texas facilities online • Customer base diversified If they execute, this re-rates hard.
$AAOI Q1 is in. Here's my take after listening to the full call. Q1 came in at the low end at $151.1M, 29.2% gross margin, -$0.07 EPS. Stock got hit on the print. None of that matters. This is a forward-looking story and the ramp doesn't start until Q3. Management raised FY26 from $1B+ to $1.1B+ revenue and $120M+ to $140M+ non-GAAP operating income. Thompson said internal target is $1.2B. Actual customer demand is $1.4-1.5B. They're guiding the only number they can control, their own capacity. The mid-2027 monthly transceiver run rate went from $378M to $471M. That's ~$1.1B in annualized revenue added to the model in one quarter. 1.6T alone in 2027 is a $2B+ business per Thompson. AAOI's TTM revenue is $507M. Q3 guided up 60-80% sequentially. Q4 similar. If Q2 hits the $189M midpoint, Q4 alone exceeds all of FY25. 40%+ gross margin target by Q4. Three 10%+ customers already in Q1 before the 1.6T ramp. Three multi-year CPO/laser agreements in negotiation. Execution risk is still there. They have to procure InP substrates, hit qualification timelines, and stand up new facilities on schedule. If they execute, this is a very solid long. Sold out through mid-2027 with customer demand 30%+ above what they can produce. Added under $140. Full breakdown linked below. $AAOI $LITE $COHR $AEHR $VIAV
1
673
AI has a memory problem. $PENG has the solution. And it’s only a matter of time before the price reflects that. Every long AI conversation builds a “KV cache” inside the GPU. Long docs, agent tasks, big context windows — the cache overflows fast. When it overflows, GPUs recompute, stall, or refuse the request. That’s why inference is so expensive.🧵
3
5
4
28
29,942
Here’s why the $KOPN base case is $13.50 and bull case $42. AI’s bottleneck isn’t compute — it’s the wires between GPUs (one of many bottlenecks). Copper hit a wall. Silicon photonics is expensive. MicroLED is the dark horse: sub-pJ/bit, no lasers, CMOS-compatible. $90B AI optics TAM by 2030.
4
13
5
73
14,755
Polymarket airdrop is going to hit like crack
131