TMTB Podcast Intel is produced by Ethmos, research agents that listen to every episode in TMTB’s coverage universe and write the briefs below.
Follow MRVL, Anthropic or NBIS on Ethmos to get a brief whenever their executives go on a podcast. TMTB readers get 14 days free and 25% off their first month.
Today’s Essential Episodes:
Marvell CEO Matt Murphy and Lumentum CEO Michael Hurlston on Their All-Stock Deal, the Scale-Up Optics Bet and NVIDIA’s Ties to Both
Squawk on the Street (CNBC), with Jim Cramer and David Faber, and Marvell CEO Matt Murphy and Lumentum CEO Michael HurlstonBen Bajarin and Jay Goldberg on Anthropic’s S-1 and Its Compute Bill, Intel’s Repeat Guidance Miss and the Marvell–Lumentum Talks
The Circuit, with Ben Bajarin of Creative Strategies and Jay GoldbergBen Thompson and Andrew Sharp on Anthropic’s S-1: Real Margins, the $80 Billion Loss, Revenue Booked Through the Clouds and the Training Bill
Sharp Tech, with Ben Thompson of Stratechery and Andrew SharpNebius CTO Danila Shtan on Why the Hyperscalers Haven’t Won AI Cloud, Unoptimized Inference and Execution as the Real Bottleneck
Unsupervised Learning, with Jacob Effron, and Danila Shtan, CTO of NebiusFormer Balyasny PM Onnig Terzian on the Agentic Tsunami: Meta, Apple and Etsy as Winners, Pinterest, DoorDash and the OTAs as Losers
Pitch The PM, with Doug Garber, and Onnig Terzian, a former Balyasny TMT portfolio manager launching his own fund
Produced by Ethmos: AI-assisted summaries of public podcast episodes. For information only. May miss context or contain errors. Verify before relying on them. Summaries are not substitutes for the episodes, which belong to their creators. Not investment advice, offer, or solicitation. Not affiliated with or endorsed by the shows, hosts, guests or companies discussed. Corrections and publisher opt-outs: support@ethmos.ai.
Listen · Watch · Follow MRVL on Ethmos
THE SIGNAL
Murphy said Inphi had about $700M annualized revenue when bought, looked expensive, and grew roughly 10x in five years on scale-out optics; he presents Lumentum as the same bet on scale-up optics (our read: helps justify the premium).
Murphy said the NVIDIA investments in Lumentum and Marvell were of similar size, implying Marvell’s is near Lumentum’s $2B; Lumentum’s centered on supply and Marvell’s on technology cooperation, so the combined company holds both NVIDIA ties.
Murphy tied scale-up demand to the copper wall inside the rack and said it holds regardless of supply swings, answering holders who fear the laser shortage ends before closing, which is more than 18 months away.
Hurlston singled out COGS synergies alongside revenue synergies, with Lumentum gross margin already at what he called a record 53% (our read: there may be further margin upside from integrated co-packaged optics that isn’t in standalone estimates).
01 Deal rationale and category creation
Murphy, as acquirer, describes the combination as merging Marvell’s leadership in AI data-center semiconductor connectivity with Lumentum’s leadership in optical technology, to form a new category of AI connectivity company. He points to the industry shift of optics moving further inside the data center as GPU clusters scale from hundreds to thousands of units, requiring semiconductors and optics to work seamlessly in what he calls the scale-up era, distinct from prior scale-out architectures.
02 Lumentum’s standalone strength and the sale decision
Hurlston says Lumentum’s just-reported quarter showed some of its best numbers ever, with gross margins of 53%, which he called the best Lumentum has seen. Asked by Jim Cramer why sell now given that trajectory, Hurlston says the timing reflects a now-or-never convergence between semiconductors and optics, pointing to Murphy’s own past instinct that led him to the earlier Inphi acquisition, and stating Lumentum must move first and act decisively or risk being left behind.
03 Commodity concerns and differentiation defense
David Faber relays shareholder worry that Lumentum’s key products are a commodity currently in short supply rather than durably differentiated technology. Murphy pushes back directly, calling the indium phosphide laser technology highly differentiated and comparing it to the esoteric bipolar and silicon-germanium process technologies he worked with for twenty-two years at Maxim. He argues today’s supply shortage reflects exploding demand across the whole optical and semiconductor industry, not fragility, and says cross-checking Lumentum’s and Marvell’s shared customer designs gave him added confidence in Lumentum’s growth durability.
04 Regulatory timeline and China risk
Faber presses on the extended timeline — deal closing not expected until first half of 2028 — and on China regulatory approval risk specifically. Murphy cites Marvell’s track record: the Cavium deal, announced in 2017, cleared regulatory review in seven to eight months, and the Inphi deal closed in about five months, both despite rising US-China tensions at the time. He calls this a very clean deal with no business overlap, says the longer stated timeline simply builds in ample buffer, and says he is very confident on China approval.
05 Stock reaction and shareholder value
Cramer notes Lumentum shares are down and asks Hurlston whether shareholders are losers long term. Hurlston responds that the company is not reacting to short-term share price moves since the deal rumors emerged, and cites expected revenue synergies and cost synergies as the basis for confidence in value creation over a three- to five-year horizon under Marvell’s leadership.
06 NVIDIA’s role and co-packaged optics roadmap
Cramer raises that both CEOs recently struck separate NVIDIA investment deals — Lumentum’s being a $2 billion convertible preferred, Marvell’s being more of a strategic technology-cooperation arrangement — and asks whether this marks the end of NVIDIA’s so-called circular deals. Murphy frames the Marvell-Lumentum combination as reinforced support for NVIDIA as ecosystem leader, saying NVIDIA received the news well. Hurlston then details the product vision: shifting toward co-packaged and near-packaged optics that combine Marvell’s semiconductor technology with Lumentum’s optical technology into a single integrated solution for hyperscaler customers, rather than each company selling discrete components — a strategy he terms extreme co design, intended to speed time-to-market and deepen differentiation.
THE SIGNAL
Before the November 2 amended S-1, Goldberg tallied Anthropic’s then-blank purchase commitments at $500-800B over about five years, reasoning they were blank because some are paid in stock (e.g., Broadcom) and so tied to the IPO price.
Goldberg reads Intel as shifting wafers from 18A to Intel 3 for server parts, which he worries slows 18A yield learning; Bajarin fears its substrate raise came too late for 2028.
The hosts put Anthropic’s gross margins at 70-80% or higher, and the S-1 says inference economics drive model design; Goldberg thinks it may raise rates once it reaches sustainable profitability around 2029-30.
Goldberg says several startups that locked up compute are pivoting to resell it, raising doubts about who besides OpenAI and Anthropic will buy AI compute long term.
01 Anthropic’s S-1: growth, risk, and chip diversity
Bajarin and Goldberg both read Anthropic’s full S-1 and flagged its heavy use of existential-risk language alongside strong growth metrics. Goldberg highlighted the company’s declared belief in scaling laws as the primary engine of AI progress, framing it as good news for semiconductor suppliers betting on ever-larger models. Both noted Anthropic is, in Bajarin’s words, the fastest-growing enterprise software company in history by revenue, still early in enterprise AI budget penetration — a point Anthropic itself emphasized via Fortune 500 customer data showing AI spend as a small share of enterprise budgets today. Goldberg pointed out Anthropic disclosed that 90% of its own code contributions are now written by Claude, illustrating internal dogfooding.
On infrastructure, Goldberg detailed Anthropic’s shift toward owning/controlling data centers via Anthropic Direct, moving away from pure reliance on cloud-rented compute (previously mostly through Amazon). He compared how the filing describes Anthropic’s chip relationships with NVIDIA, AWS Trainium, Google TPU and AMD, finding notably different contractual language for each: NVIDIA and AMD get described collaboration on design and engineering, Trainium gets an “opportunity to influence” future generations, while Google’s relationship was described with no comparable collaborative language. Bajarin connected this to his own thesis of workload fungibility across diverse silicon, comparing it to Meta’s vertical compute strategy discussed in Meta’s Prometheus blog with NVIDIA.
02 Enterprise adoption and ROI survey
Bajarin presented findings from his newly published second round of the EA index CIO/CTO survey. Most of the 200-plus CIOs surveyed aren’t tracking AI ROI at all; among the 33 who are, 67% saw ROI by the survey’s measure, and he says that ROI grew as AI usage scaled from one business unit to several. Bajarin also noted some business units are funding AI spend outside official IT budgets — a form of shadow AI becoming official — which he called a high-signal but hard-to-track phenomenon. Both hosts agreed the adoption curve likely inflects suddenly rather than gradually once enterprises cross an education and skills threshold.
03 Anthropic’s governance and capital structure
Goldberg spent considerable time on Anthropic’s five (soon four) classes of stock with differentiated voting rights, designed — in his reading — as a direct reaction to the 2023 OpenAI board turmoil, giving the seven co-founders effective control. He flagged the company’s ethics-oversight board structure and the S-1’s repeated warnings that shareholders’ pecuniary interest is not the company’s top priority. Goldberg also noted, as speculation, that Anthropic was founded in 2021 — two years before the OpenAI leadership clash — suggesting the founders may have planned a competing lab in advance.
On capital commitments, Goldberg said Anthropic was actually profitable and cash-flow generative in Q2, but likely received below-market, subsidized compute pricing (he estimates discounts around 50% versus his own cost models) from partners like Amazon, given Anthropic’s scale as AWS’s largest AI customer. Both hosts debated whether such discounts are sustainable or an early-stage subsidy, comparing it to Uber’s early pricing, that fades once labs reach steady-state profitability around 2029-2030.
04 Hyperscaler earnings and the flywheel question
The hosts covered a week of big capital raises — Meta’s equity offering, Amazon’s debt offering, and Oracle’s analyst day, where Oracle said it would not need further capital raises. Earnings reactions were mixed: Meta’s results were seen as fine, helped by engagement from its AI agent Muse; Google’s stock fell on growth concerns despite search holding up; Amazon’s numbers drew similar skepticism. Both hosts framed the core question as whether AI infrastructure investment is lifting the rest of each hyperscaler’s business — a flywheel test. Goldberg argued Google’s founders are true AI believers willing to keep spending, and that consumer ad-search risk is navigable by improving search quality; he and Bajarin agreed Amazon and Microsoft face more enterprise-dependency risk since they rely on Anthropic and OpenAI models rather than owning frontier capability themselves.