Lol, the market picks a side, and it's easy to tell which one it's chosen right now. Elon told the world a few days ago during SpaceX $SPCX earnings that AI is going to be compute-short for a long time... the market shrugged it off and goes, "sure sure, Elon exaggerates all the time" Elon says many small neoclouds won't survive, and the market triples the weight of these claims: "SEE I TOLD YOU, ELON SAID IT, SO IT MUST BE RIGHT"

Soo the person that wants to turn on over 13GW of compute in a market that is constraints in various things is also trying to scare the market from working with other smaller players. Hey don't work with that neocloud!! They might burst, but you know who won't burst! SpaceX, so don't make power deals with the smaller players, maybe don't sell chips to them, but you can sell it to me!!

This Is What the Chip Happened to Semiconductor and AI Investors Today! NVIDIA $NVDA expands the supply chain: At GTC Berlin, NVIDIA outlined partner plans for 2.5× substrate capacity over three years. Amkor $AMKR plans to increase South Korean interposer-wafer capacity by more than 300% by end-2028, while SPIL targets 15× optical-engine manufacturing capacity by mid-2027. NVIDIA also plans twice the testing capacity in 2027 versus 2026, including a $1.1B capacity-reservation advance supporting new U.S. facilities. Its DSX POD containers arrive integrated, cabled, piped and tested. NVIDIA says deployment can fall from six months to as little as 14 days at a prepared site with power, cooling and networking ready; worldwide partner availability is planned for early 2027. Meta $META pairs ad growth with a bigger AI buildout: Q3 revenue reached $64.60B, up 26%, versus the $63.37B LSEG estimate reported by Reuters. Ad impressions rose 10% and average ad prices increased 14%. Zuckerberg said more than 10M users in the U.S. and Canada had set up Muse agents in about six weeks, with approximately 25% using them on a given day. Transaction fees and additional subscriptions remain planned monetization routes. Full-year capex guidance, including finance-lease principal payments, narrowed to $135–145B from $130–145B. GAAP EPS was $3.83, including a $9.89B legal-agreement charge; removing that charge alone would put EPS at $7.40. Q4 revenue guidance is $73–76B. Lam Research $LRCX sees a larger equipment market ahead: September-quarter revenue reached $8.22B, up 22% sequentially, with non-GAAP EPS of $2.20 versus the $8.14B and $2.17 consensus reported by EarningsWhispers. Systems revenue rose to $5.73B from $4.25B, while customer support and other revenue was nearly unchanged at $2.49B. December-quarter guidance calls for $8.60B ± $400M in revenue and $2.33 ± $0.15 in non-GAAP EPS. Management’s preliminary wafer-fabrication-equipment outlook exceeds $200B for 2027, versus the mid-$150Bs for 2026. That is an industry-spending forecast, not booked revenue, and supply and cleanroom timing could delay shipments. Tesla $TSLA grows storage deployments, but energy revenue falls: Tesla deployed 13.7 GWh of energy storage in Q3, up 10% year over year, while energy revenue fell 16% to $2.881B. Its 33.1% energy gross margin benefited from tariff refunds and warranty reimbursements, so that margin should not be treated as a clean ongoing run rate. Across Tesla, revenue was $28.05B and non-GAAP EPS was $0.48, versus $28.34B and $0.45 in Tesla’s pre-release compilation of analyst estimates. Free cash flow reached $1.53B, with operating cash flow of $6.26B and capex of $4.73B. Tesla also plans 400 MW of training compute by year-end for vehicle and humanoid autonomy software. Micron $MU maps its next memory ramps: Micron’s October 21 technology presentation targets second-half 2027 ramps for 1-delta DRAM and G11 NAND. G11 targets approximately twice G9’s bit density, using more than 400 layers and wafer-on-wafer bonding. The presentation also places 3D DRAM in development on a directional roadmap without a launch date. The 2027 timing belongs to the planned 1-delta and G11 ramps, not a confirmed 3D DRAM launch.

$META’s first earnings-call question gets to the economics of scaling Muse. Morgan Stanley’s Brian Nowak asked what constrains user growth and what Meta wants to build in enterprise AI. Zuckerberg said Muse passed 10M setups in the U.S. and Canada in its first six weeks, but compute intensity is one reason it hasn’t launched globally. Local optimization and connectors also need work. Serving a potential billion-plus people depends on hitting compute-efficiency targets. Mark expects that work to remain a limiting factor over the next year, alongside building the business around the product. There’s already a product ladder: Spark powers the free tier, Wonder is available in the paid tier, and Meta is developing a more efficient Wonder version for broader distribution. For enterprise, the near-term offering is model APIs, Muse Code and business agents. Longer term, Mark wants consumer and business agents interacting directly, including for commerce. CFO Susan Li added the financial framework: diversified recurring revenue through subscriptions, usage-based APIs, messaging and other services. She expects that opportunity to develop over multiple years, with investment scaling behind demonstrated customer demand and product-market fit. That answer puts two practical conditions on expansion: cheaper compute and evidence customers want the products. Li didn’t provide a new enterprise revenue target or capex figure.

This is bullish for semiconductor equipment demand. $LRCX now sees preliminary 2027 wafer-fab equipment spending ABOVE $200B, compared with the mid-$150B range in 2026. $ASML $AMAT $TSM CEO Tim Archer expects DRAM to lead percentage growth, followed by leading-edge foundry/logic, then NAND. That puts both AI memory and advanced chip manufacturing behind the spending expansion. $MU Lam also expects calendar 2026 systems revenue growth well above 40%, including the midpoint of its December-quarter guidance. The constraint isn’t simply whether customers want more capacity. Management says cleanroom availability and component supply will continue limiting how quickly the industry can meet demand into 2027. For equipment investors, that’s an important distinction: a strong spending outlook can coexist with bottlenecks that affect delivery timing. My takeaway: the quarterly stock reaction and the longer-term equipment opportunity are two different conversations. Lam’s preliminary outlook points to a much larger equipment market, although it doesn’t guarantee equal growth for every supplie

Texas Instruments $TXN reported $6.06B in Q3 revenue, up 28% year over year and 11% sequentially, ahead of the $5.94B FactSet estimate reported by IBD. Analog revenue grew 32% to $4.91B, while embedded processing increased 21% to $858M. CEO Haviv Ilan described “continued broad growth in industrial, automotive and data center.” The live call adds a useful detail: management says data-center revenue grew approximately 95% year over year and 20% sequentially. Industrial grew more than 40% annually, while automotive grew in the mid-teens. GAAP EPS was $2.62. Q4 guidance is $5.80–6.30B revenue and $2.45–2.79 EPS. The $6.05B revenue midpoint is above the $5.76B FactSet estimate reported by IBD, but roughly flat with Q3. So this isn’t only an AI data-center story. TI is reporting growth across analog, embedded, industrial and automotive, with data centers growing particularly quickly.

$TSLA missed revenue expectations slightly, beat adjusted EPS, and delivered a much bigger cash-flow surprise. Against Tesla’s October 19 company-compiled analyst averages: Revenue: $28.046B vs $28.341B expected. Non-GAAP EPS: $0.48 vs $0.45. GAAP EPS: $0.22 vs $0.23. Gross margin: 19.6% vs 17.9%. Free cash flow: positive $1.531B vs negative $4.506B. The cash-flow difference needs context. Capex was $4.732B versus $7.771B expected, while operating cash flow was $6.263B versus $3.267B. Both lower spending and stronger operating cash generation contributed. Automotive gross margin excluding regulatory credits improved to 18.1% from 16.3% sequentially. Tesla also says Cybercabs entered the Austin Robotaxi fleet, Semi volume production began in Nevada, and Optimus production began in Fremont.

$LRCX delivered strong growth, but the upside versus consensus was much smaller than the growth rate suggests. September-quarter revenue reached $8.219B, up 22% sequentially, versus roughly $8.14B expected. Non-GAAP EPS was $2.20 versus $2.17 consensus, according to EarningsWhispers. That’s a small consensus beat. It was below the separate $2.28 whisper estimate, which is not the same thing as Wall Street consensus. December guidance calls for $8.6B revenue, plus or minus $400M, versus about $8.5B consensus cited by EarningsWhispers. The midpoint implies another 4.6% of sequential growth, not a contraction. Profitability remains strong: non-GAAP gross margin was 52.4% and operating margin 40.2%. December guidance midpoints are 52.5% and 40.5%, respectively. CEO Tim Archer says: “Our strategic investments in technology inflections and manufacturing capacity are paying off”. My read: this is strong business performance without a huge upside reset to expectations. That can help explain why a beat doesn’t automatically produce a positive stock reaction, but it doesn’t prove the cause of the selloff. Management’s call still matters.

My read on $META’s new non-voting shares: this could become another way to fund AI capex through future stock issuance while preserving Zuckerberg’s voting control. Semafor reported that exact financing angle today. Meta’s own announcement also connects the new share class to capital raising and AI investment. Here’s the math for existing holders, using a hypothetical $800 price: 100 Class A shares × $800 = $80,000. You keep those 100 A shares and receive 100 new Class C shares. If each class traded at $400 afterward, you’d have $40,000 in A plus $40,000 in C. Still $80,000, not $160,000. Actual prices can differ, and markets move. Class A keeps its votes; Class C has none. That proportional distribution doesn’t itself dilute your percentage ownership. The potential dilution comes later, if Meta issues additional C shares to others. Example: owning 200 of 20,000 shares means 1% ownership. If another 2,000 shares are issued and you receive none, your 200 of 22,000 becomes about 0.91%. That is a hypothetical future issuance, not an announced financing transaction. But it explains the distinction: Meta could raise more equity for AI without issuing more voting power. With 2026 capex guided to $135–145B, including finance-lease principal, that flexibility matters. More shares wouldn’t automatically mean less investment value; the price raised and value created with the money would matter too. January 19, 2027 is the intended charter-change date. The dividend’s record and payment dates haven’t been announced.

$META grew revenue 26% to $64.60B, about 2% above the $63.34B forecast listed by TipRanks. But the AI spending and monetization details deserve just as much attention. Q4 revenue guidance is $73–76B. Full-year capex is now $135–145B, raising the bottom of the previous $130–145B range. That includes finance-lease principal payments. The headline $3.83 GAAP EPS includes a $9.89B litigation charge. Meta says EPS excluding that specific charge was $7.40. On the live call, Zuckerberg says more than 10M people in the U.S. and Canada set up Muse agents in 1½ months, with roughly 25% using them on a given day. The planned business model includes commerce fees paid by businesses, advanced-model subscriptions and business subscriptions. He also expects more than 10× compute cost-efficiency improvement next year. That is management’s forecast, not a claim that Muse is already profitable. Q3 free cash flow was negative $2.12B alongside $34.91B in capex. The spending is happening now; the new agent business still has to turn usage into revenue.

NVIDIA $NVDA is expanding the supply chain behind its AI systems across substrates, packaging, testing and optics. Its partners plan 2.5× substrate capacity over three years. Amkor $AMKR plans a more than 300% increase in South Korean interposer-wafer capacity by the end of 2028. SPIL is targeting 15× optical-engine manufacturing capacity for co-packaged optics by mid-2027. Testing has its own timeline: NVIDIA plans twice the capacity in 2027 versus 2026, including a $1.1B capacity-reservation advance supporting new U.S. testing facilities. High-volume testing there is targeted for April 2027.

NVIDIA IS THE GOAT!!! NVIDIA $NVDA says AI-factory deployment could fall from six months to as little as 14 days with DSX POD. That’s a pretty big change in how these systems get built. The idea: assemble and test the system before it reaches the customer’s site. Vera Rubin computing, networking, power and liquid cooling arrive in partner-built modular containers. Cabling, piping and validation happen at the factory instead of being pieced together on-site. The important detail: those 14 days start at a prepared site. Power, cooling and networking must already be ready. It doesn’t eliminate the work of securing land, permits or electricity. Partners include Dell $DELL, HPE $HPE, Vertiv $VRT and Eaton $ETN. Worldwide delivery is planned for early 2027. This is why I keep coming back to the complete AI system. Faster chips matter, but getting those chips installed and working sooner is another part of the opportunity.

What the Chip Happened to AI Semiconductor Investors Today! ELON SAVED THE DAY!! SpaceX $SPCX reported Q3 revenue of $14.6B, up 161% year over year. Its Super Intelligence segment generated $7.6B, up 830%, with new cloud-services agreements, the Cursor acquisition and growing Grok adoption contributing. Nameplate compute reached 1.7 GW at September 30, versus 1.4 GW at June 30. Starlink consumer subscribers reached 13.6 million, up 1.6 million from the previous quarter. Supermicro $SMCI estimates revenue of $18.5–19.5B for its first quarter of fiscal 2027, ended September 30, above its previous $14.5–15.5B guidance. Estimated gross margin increased to 12.0–12.5%, versus 10.4–10.8% previously. Management attributes the revenue increase primarily to AI-platform demand and the margin improvement to customer and product mix. These figures are preliminary and unaudited; the earnings call is scheduled for November 3. The Wall Street Journal reports that Anthropic’s draft IPO prospectus shows $16.3B in first-half 2026 revenue, versus $1.1B a year earlier. Its reported $80.7B net loss included $72B in accounting charges tied to changes in the value of convertible notes and other financing instruments. That loss is not a measure of cash burn. Anthropic declined to comment to the Journal. Amazon $AMZN announced plans to invest more than €16B in Germany in 2027, including €4.8B to expand the AWS Frankfurt Region for cloud and AI services. The wider plan also covers satellite internet, logistics and Prime Video, so the €16B-plus is not exclusively AI or data-center spending. Salesforce $CRM agreed to acquire Braze $BRZE for approximately $5.66B in enterprise value, with Class A shareholders receiving $47.50 per share in cash if the deal closes. Salesforce plans to connect Braze’s real-time customer engagement capabilities with Data 360, Marketing Cloud and Agentforce. Closing is expected in Salesforce’s fiscal 2028, subject to shareholder and regulatory approvals; Salesforce anticipates no change to fiscal 2027 guidance.

SpaceX $SPCX gave a more specific Terrafab roadmap on its Q3 call. Musk’s internal target is a functional test chip around May–June 2027 at the Austin research fab. He described an SRAM-type chip used to test the production line, not a finished AI processor. For the main Terrafab site at Gibbons Creek, he initially targeted volume production around Q3 2028, then described Q3 or Q4 2028 as possible outcomes. He called those dates extremely aggressive and stressed the large uncertainties around achieving them. On equipment, Musk thanked ASML $ASML and other suppliers for accelerating deliveries. But the analyst’s specific question about EUV machine orders and delivery timing did not receive an explicit answer. No EUV order quantity, model or delivery date was confirmed in that exchange. The longer-term ambition is much larger: Musk said Terrafab ultimately aims to produce a terawatt of AI chips per year. He did not provide a date for that goal. The milestones are distinct: an Austin test chip first, then volume manufacturing at Gibbons Creek. Late-2028 production remains an aspiration,

SpaceX $SPCX outlined how it wants to monetize its expanding AI compute capacity. Musk’s rough allocation goal: 50% for internal use and 50% rented to third parties. That mix can move to 60/40 in either direction as demand changes. The contract strategy is just as important as the split. Musk said SpaceX favors short-term agreements at effectively spot-market rates rather than locking capacity into long-term deals. His rationale: better AI models increase the useful intelligence produced per watt, supporting higher compute pricing. He said existing short-term contracts are renewing at higher dollars per watt and claimed demand could absorb available capacity several times over. Earlier, CFO Bret Johnsen said new compute deployments were generating payback periods of less than a year at current pricing. Musk also floated a possible six-month payback in 2027, but explicitly said he was NOT predicting that outcome.

SpaceX $SPCX and Tesla $TSLA have formed independent special committees to discuss a potential merger, according to Elon Musk on SpaceX’s Q3 call. Musk said he cannot participate in those committee discussions because he is a conflicted party. He supports combining the businesses and highlighted potential cooperation across AI software, chip design, Terrafab, robotics, connectivity and space infrastructure. But the committees have NOT agreed on transaction terms. Musk explicitly pushed back on treating a merger as certain. He said any proposal would need shareholder support, describing Tesla’s approval hurdle as 50% of all shares, rather than only shares voted. No agreed exchange ratio or transaction timetable was disclosed in that answer.

SpaceX $SPCX is planning a massive AI infrastructure buildout, and the spending numbers now put scale behind it. The CFO expects $35–40B of total company capex in Q4 2026, versus $19.59B in Q3. That would be roughly 79–104% higher sequentially, with additional compute capacity coming online this quarter. For context, $16.44B of Q3’s spending was in SI. Looking into 2027, Musk says SpaceX is aiming to add 20GW of power and cooling infrastructure. The goal is to have the infrastructure ready for the GPUs, rather than let it become the bottleneck. Keep the numbers separate: $35–40B = Q4 total-company capex guidance, not AI-only spending. 20GW = additional power/cooling infrastructure targeted for 2027, not operating compute already installed. Musk separately discussed roughly 13GW of compute by the end of 2027. More than 20GW of compute was described as a small, non-zero possibility, not the expected outcome. The AI buildout is about the chips AND the infrastructure needed to run them. SpaceX is putting enormous spending plans behind both. The CFO says new compute deployments are generating payback periods of less than one year at current pricing.